Independent Research · AI Shopping
AI Shopping Recommendations: What Tarte and Belgian Boys Show Beyond Brand Counts
An identical brand-recommendation count does not mean two AI systems gave shoppers the same options. In two separate exploratory US API-answer studies, Kojable examined 20 selected Tarte Cosmetics answers collected on 3 November 2026 and 60 Belgian Boys replication answers collected on 5 November 2026. Each provider produced the same target-brand recommendation total within each study, but the shopper questions and product choices behind those totals differed. Belgian Boys' three responses per question add a limited view of recurrence within one collection window. These are observations of saved answers, not market-share estimates or evidence of sales.
Key finding
Both providers recommended Tarte in 2/5 selected unbranded answers and Belgian Boys in 6/15 unbranded replication answers. The questions and products behind the equal totals differed within each study.
Research disclosure: This was an unpaid Kojable investigation. Both brands were subjects of the research, not participants or endorsers. The findings concern recorded OpenAI and Meta model API answers under the stated conditions; they do not represent personalised ChatGPT or Meta AI conversations. Product claims made by those answers were not independently verified.
- Original Research
- Tarte Cosmetics
- Belgian Boys
The finding in brief: the same count can hide different answers
The two cases have a simple, revealing feature in common: for each brand, the two model providers returned the same total number of qualifying target-brand recommendations across that study's selected unbranded shopping questions.
| Separate study | OpenAI saved answers recommending the target brand | Meta Muse Spark saved answers recommending the target brand | What the count hides |
|---|---|---|---|
| Tarte Cosmetics — five unbranded questions, one selected answer per provider/question | 2 of 5 | 2 of 5 | Which needs produced a Tarte recommendation and which named product or variant appeared. |
| Belgian Boys — five unbranded questions, three fresh answers per provider/question | 6 of 15 | 6 of 15 | How recommendations were distributed across breakfast needs and whether an inclusion recurred in the three saved answers to the same question. |
How to read this table: The numerator counts saved answers containing a qualifying recommendation of the named brand, once per answer. Each denominator is the eligible unbranded answer panel for that provider in that study. Neither row measures shopper impressions, market share, retail purchases or a platform-wide probability of recommendation. The two studies used different products, questions and repetition rules. Their totals must not be combined into a single 80-answer recommendation rate.
The useful conclusion is about measurement resolution. A brand count is an efficient summary of whether the brand appeared as a qualifying choice. It does not, by itself, tell us for whom, in which product form, in what recommendation role or how often the outcome recurred when the same question was asked again.
That distinction matters in AI-mediated product research. A positive recommendation for a specific mascara under one set of conditions is not interchangeable with a conditional recommendation for a lip product under another. Nor is a single saved answer interchangeable with three fresh answers collected to the same prompt.
Two exploratory studies, two different collection designs
Kojable used deliberately selected shopping questions relevant to each brand. The studies were not a random sample of shoppers, a single harmonised cross-industry experiment, or a controlled comparison of personalised consumer apps.
| Design field | Tarte Cosmetics | Belgian Boys |
|---|---|---|
| Data collected | 3 November 2026, 09:52–10:09 UTC | 5 November 2026, 13:24:09–13:43:01 UTC |
| Market and question context | United States; en-US; USD | United States; en-US; USD |
| Recorded answer providers | OpenAI API gpt-6-luna; Meta API muse-spark-1.3-contributor | OpenAI API gpt-6-luna; Meta API muse-spark-1.3-contributor |
| Question panel | Ten purposive questions: five unbranded, five branded | Ten purposive questions: five unbranded, five branded |
| Response rule | One selected usable answer per provider/question; 20 selected answers | Three fresh answers per provider/question in one collection window; 60 replication answers |
| Collection qualification | OpenAI's selected answers came from the initial collection. Initial Meta requests failed, so the selected Meta answers came from a later first-successful-response recovery. Provider capture was not simultaneous. | Three responses per provider/question did not share conversation history. An earlier 20-answer pilot was separate and excluded from the 60 replication answers. |
| Coding and review | Saved answers were analysed with automated Claude-based coding and the report was reviewed by Kojable. The report does not attest to a separate independent human coding review. | Automated coding used Claude Sonnet 5.5. Piush, Kojable's founder, documented a spot-check of 20 of the 60 replication answers and their coding, with no corrections. |
| Inference limit | One selected answer per provider/question cannot estimate repeatability. The Meta recovery complicates direct timing comparisons. | Three answers in one short window allow a description of within-window recurrence, not long-term consistency or market probability. |
Both projects recorded API-model answers with native-auto search/grounding under the settings reported in their source records. Their model IDs identify the collected configurations, not universal behaviour of every OpenAI or Meta product. Search/source annotations are discussed separately from recommendations; a retrieved record is not automatically a source used in a final answer.
The differences between the collection designs matter as much as the commonality in their headline brand counts. If we ignore those designs, we risk treating a small selected answer as evidence of a stable provider preference, or treating repeated answers to five breakfast questions as a representative survey of an entire grocery category.
What counted as a recommendation?
Study definition
A qualifying AI shopping recommendation in these studies is a saved model answer's own positive or conditional product choice for a defined shopper need, counted once per target brand per assessable answer.
The unit was an assessable saved model answer. For an unbranded discovery question, we counted the target brand once when the answer's own narration offered one of its products as a positive or conditional choice for the shopper's stated need. An eligible alternative could qualify; the model did not need to place the brand first.
We did not count a passive brand mention, text quoted from a reviewer, an option the answer rejected, or buying instructions for a product the shopper had already chosen. Product names and variants were retained as the model communicated them. A long product label in an AI answer should not be mistaken for an independently reconciled manufacturer SKU.
This distinction also separates two tasks that look similar in a dashboard:
Unbranded discovery: A shopper asks about a need or category without naming the target brand. Does the brand enter the model's recommended options?
Branded consideration: A shopper names the brand or products and asks for a comparison, suitability explanation or purchase guidance. The answer can discuss the brand because the prompt introduced it. That is not an unaided discovery event.
The 2/5 and 6/15 figures in this report refer only to the unbranded panels. Findings from branded comparison questions are labelled separately.
Tarte Cosmetics: equal totals, different needs and product names
The original Tarte case used five selected US shopping questions that did not name Tarte: full-coverage concealer, dry under-eyes, oily/combination foundation, tubing mascara and a glossy, slightly plumping lip product. Each provider contributed one selected saved answer per question. The resulting inclusion pattern was:
| Unbranded question | OpenAI's selected answer | Meta Muse Spark's selected answer |
|---|---|---|
Q01 — Full-coverage concealer for dark circles without a cakey appearance | Tarte recommended | Tarte recommended |
Q02 — Concealer for dry under-eyes and fine lines | No qualifying Tarte recommendation | No qualifying Tarte recommendation |
Q03 — Long-wearing foundation for oily/combination skin within a US budget | No qualifying Tarte recommendation | No qualifying Tarte recommendation |
Q04 — Tubing mascara within a US budget, with wear and removal needs | Tarte recommended | No qualifying Tarte recommendation |
Q05 — Glossy, slightly plumping lip product within a US budget | No qualifying Tarte recommendation | Tarte recommended |
| Total across these five selected unbranded answers | 2/5 | 2/5 |
At the brand level, the totals are equal. At the shopper-question level, they are not the same pattern.
Both Q01 answers included Tarte for the full-coverage concealer question, but they did not communicate the same product name. The OpenAI answer used the saved label “Maracuja Creaseless Full Coverage Radiant Undereye Concealer”; Meta used “Shape Tape Concealer”. We preserve those historical answer labels. The longer OpenAI phrase has not been independently matched to a current official Tarte SKU.
In Q04, OpenAI named Tartelette XL Tubing Mascara as a recommendation. Meta's selected answer did not recommend Tarte for that question. In Q05, the direction reversed: Meta included Maracuja Juicy Lip Plump, while OpenAI's selected answer did not include a qualifying Tarte option. The answers differed in the shopping situations where Tarte entered the selection—not merely in the number of times the word “Tarte” appeared.
The zeroes in Q02 and Q03 are equally bounded. They mean no qualifying Tarte recommendation in those two selected answer pairs. They do not show that the brand lacks a suitable product, that the answers were correct, or that either provider would omit Tarte when asked again.
A brand being recognised is not the same as being recommended without prompting
The original Tarte investigation also included five branded questions. One asked about choosing between versions of Shape Tape for different skin conditions; another placed Shape Tape alongside named alternatives. Those answers discussed product variants and trade-offs after the brand was introduced by the shopper.
That creates an important interpretive boundary. In the unbranded dry-under-eye question (Q02), neither selected answer recommended Tarte. In a branded variant question (Q07), the answers nonetheless discussed original and Creamy Shape Tape options. This shows a difference between answering a question about a named brand and selecting that brand when it was not named. It does not prove a content gap, a suitability claim or a reason for the omission.
The retained Tarte source also examined citations and source visibility. Those annotation counts are different measurements from recommendation inclusion. Provider-specific capture differences, including missing citation annotations for some saved responses, mean that a larger number of citation events should not be treated as stronger evidence, a higher-quality provider or the cause of a product recommendation.
The selected Tarte results were collected on 3 November 2026; the original standalone case was published on 5 November 2026. The Meta answers came from a later recovery after initial failures, and the study did not run repeated responses for each question. Its product names, suitability statements, retailer details and policy descriptions are recorded model output, not independent product verification.
Read the original Tarte findings, answer pairs and source tables.
Belgian Boys: the same total came from different breakfast questions
The Belgian Boys replication asked five unbranded questions about everyday breakfast and snack needs, with three fresh answers from each provider for each exact question. That makes the denominator 15 answers per provider, rather than Tarte's five.
| Unbranded shopper question | OpenAI answers recommending Belgian Boys | Meta Muse Spark answers recommending Belgian Boys |
|---|---|---|
U01 — Store-bought breakfasts for school-age children, ready in under five minutes | 0/3 | 0/3 |
U02 — Ready-made mini pancakes for busy weekday mornings | 1/3 | 0/3 |
U03 — Ready-made Belgian-style waffles without making batter | 2/3 | 3/3 |
U04 — Ready-made crêpes for an easy weekend brunch | 3/3 | 3/3 |
U05 — Ready-to-eat sweet snacks for children, served cold or quickly warmed | 0/3 | 0/3 |
| Total across these 15 unbranded saved answers per provider | 6/15 | 6/15 |
The totals are again equal, but the routes to those totals differ. For the mini-pancake need (U02), Belgian Boys appeared as a conditional choice in one of OpenAI's three saved answers and none of Meta's three. For the waffle need (U03), the brand appeared in two of three OpenAI answers and all three Meta answers. For the crêpe need (U04), it appeared in all three answers from each provider.
Conversely, Belgian Boys was not a qualifying recommendation in either provider's three saved answers to the broad five-minute breakfast question (U01) or the quick sweet-snack question (U05). That is an absence in this purposively chosen question panel and short collection window. It is not a verdict on the brand's whole grocery range or future visibility.
The difference between U02, U03 and U04 is a concrete example of why one brand-level percentage is not enough to interpret AI shopping answers. A team reading only the equal 6/15 totals would miss the variation between mini pancakes, waffles and crêpes. A team reading only the product names would still need to know which shopper need and response each name belonged to.
Repetition adds information, but only within the observed window
Belgian Boys' three requests per provider/question allow a limited answer to “did this inclusion recur when the same question was asked again?” They show recurrence in U03 and U04, and non-recurrence in U02, within the 5 November 2026 collection window.
The study did not repeat the same panel across several days or months, randomly sample real shopper questions, or experimentally manipulate a product page. Three answers to one question are not evidence of a permanently stable model preference. Tarte, by comparison, has only one selected response per provider/question and cannot answer even this short-window recurrence question from its retained selected panel.
The Belgian Boys research also had an earlier 20-answer pilot. The pilot informed the process but was analysed separately and is not included in the 60 replication answers or any unbranded recommendation numerator in this article.
Read the exact Belgian Boys questions and replication evidence.
Recommendation role matters: a choice is not always a first choice
A qualifying recommendation can be a primary pick or a conditional alternative, depending on how the answer is worded. Counting both can be useful for one narrowly defined presence measure, but those roles are not identical shopper-facing outcomes.
Belgian Boys' U02 mini-pancake result demonstrates this: the brand's one OpenAI inclusion was conditional, not evidence that it was the primary product selected across the three responses.
A separate branded Belgian Boys question (B02) compared Original Belgian Waffle with Bite-Sized Waffles for busy mornings. In the saved answers to that brand-named comparison, Original Belgian Waffle was the primary recommendation in 3/3 OpenAI and 3/3 Meta responses, while Bite-Sized Waffles was treated as a conditional option for a different use. Those counts describe the models' stated recommendations in that exact prompted comparison. They do not establish actual product superiority or independently verified suitability.
Because B02 named Belgian Boys products in the question, its six answers were not counted in the unbranded 6/15 discovery panel. Adding branded comparison outcomes to unbranded recommendations would change the construct being measured without warning the reader.
This is why a useful interpretation of an AI recommendation should preserve both a presence flag and a role label, alongside the question and product identity. A brand may be present as a conditional backup, as a directly favoured option, or simply as text that is not a narrator recommendation at all. A single aggregate can obscure those states.
What these two studies can—and cannot—tell us
These cases support a bounded measurement lesson: for the saved answers examined here, provider-level target-brand recommendation counts hid differences in shopper needs, named products and recommendation roles. The Belgian Boys design additionally exposed short-window recurrence for the exact same questions. That is the cross-case finding, not a claim that every AI system behaves this way for every product.
| What is visible in the retained studies | What cannot be concluded from it alone |
|---|---|
| Whether the target brand was a qualifying recommendation in a specified saved answer | Its probability of appearing for the entire market or all shopper queries |
| Which need, product label and recommendation role accompanied an inclusion | That the label is a verified current SKU, or that a claimed feature, ingredient or suitability is true |
| Whether Belgian Boys appeared in zero, one, two or three saved responses to the same question during one short window | Long-term stability across days, models, updates, geographies or personalised apps |
| Recorded source and citation annotations where available | The complete hidden retrieval pool, internal reranking, source causation or model training provenance |
| Different products and competitor selections in two selected providers' answers | Which provider is generally “better”, which brand deserved recommendation, or which answer produced a purchase |
Nor should the two fractions be read as comparable brand performance merely because 2/5 and 6/15 happen to represent the same mathematical fraction. They arose from different questions, product categories, response rules and collection dates. There is no pooled 80-answer visibility rate, no randomised benchmark and no statistical basis for ranking Tarte against Belgian Boys.
The results also cannot tell us that a manufacturer's page, retailer listing, citation or SEO change caused a specific recommendation. Both studies observed answers. Neither randomised source availability nor tested a controlled before-and-after content intervention. The visible answer and source-annotation layers are evidence about what was recorded, not full telemetry of why an AI model selected an option.
This matters when interpreting recommendations as business signals. A model describing a product favourably is not itself a sale, a qualified lead, an independently verified product claim or a measured commercial outcome. Those would require different evidence.
Methods, source records and reproducibility
Kojable designed these as two separate exploratory investigations. Both used purposive questions, United States shopping context, en-US and USD. The model IDs recorded in the source collections were gpt-6-luna through OpenAI's API and muse-spark-1.3-contributor through Meta's API, using the saved collection's native-auto web-search configuration. The answers should not be presented as personalised ChatGPT, Instagram, WhatsApp, Facebook or Meta AI application sessions.
Answer and metric rules. The brand-inclusion numerator is the number of distinct assessable saved answers in which the answer narrator offered a qualifying positive or conditional target-brand product choice for the shopper's stated need. A brand is counted at most once per answer. The denominator is the eligible unbranded answer count within that named provider and study. Branded questions, quote-only appearances, passing mentions, rejected options and buying instructions for an already-selected item are excluded. The main Tarte and Belgian Boys tables display exact eligible answers and per-question values.
Selection and collection. Tarte retained one selected usable response per provider/question. After initial Meta requests failed, the saved Meta selections were taken from a later first-successful-response recovery; this was not a simultaneous or repeated-response provider test. Belgian Boys made three fresh response requests per provider/question in one window on 5 November. Its earlier pilot was excluded.
Analysis and review. The Tarte report used automated Claude-based statement coding and underwent Kojable report review; it does not document a separate independent human check of the automated recommendation classifications. Belgian Boys used automated Claude Sonnet 5.5 coding; Piush, Kojable's founder, spot-checked a preselected set of 20 of 60 replication answers and coding covering both providers and the question panel, and recorded no corrections. These are different review histories, not one shared verification standard. Neither review established independent manufacturer or retailer fact checking.
Record retention and scope. Kojable's original project records retain the Tarte selected dataset, analysis tables and 20 selected raw-response files, as well as Belgian Boys' saved dataset, analysis and supporting exports. This synthesis uses the recorded findings and reconciled question summaries; its preparation did not independently replay every collection, parsing and analysis stage. The technical evidence tables below make the public figures inspectable. Any promise of a fully reproducible public raw-data package would require checking the precise exports, transformations, access conditions and redistribution rights first.
Funding and relationships. This is an unpaid Kojable investigation. The original Tarte case reported no customer relationship, commissioning, participation or endorsement from Tarte Cosmetics. Belgian Boys was neither a Kojable client nor a prospect. No claim is made here about Tarte's prospect status. Brand inclusion in a research sample does not signify the brand's endorsement or involvement. “Unpaid” is a funding disclosure, not a waiver of evidential standards or a reuse licence.
Product-fact boundary. Saved shopping answers contained assertions about prices, stock, dietary characteristics, suitability, benefits and policies. Such statements were not independently verified at the collection time or for present-day shopping. Where product names appear in this article, they are historical model-output labels, not our recommendation to buy those products.
Frequently asked questions
Were the 80 saved answers one AI-shopping experiment?
No. 80 is only an inventory: Tarte retained 20 selected answers from ten questions with one answer/provider/question; Belgian Boys retained 60 replication answers from ten questions with three responses/provider/question. The panels, product categories and collection dates differed. The earlier Belgian Boys pilot was excluded, and no pooled recommendation percentage is reported.
Why are equal brand-recommendation totals not equivalent results?
Within Tarte's unbranded panel, both providers returned 2/5, but the mascara and lip questions produced different Tarte inclusions. Within Belgian Boys' panel, both returned 6/15, but mini-pancake and waffle inclusion differed. The same total can therefore conceal different question distributions, named products and roles in these saved samples. It does not establish a general market rule.
Which Tarte products did the models recommend?
In the selected Tarte unbranded answers, both providers included Tarte for the full-coverage concealer question but used different saved product labels. OpenAI included Tartelette XL Tubing Mascara in the tubing-mascara answer, and Meta included Maracuja Juicy Lip Plump in the glossy-lip answer. These are dated saved-answer observations, not current SKU or performance verification.
Which Belgian Boys shopper needs produced recommendations most often?
In the specified unbranded panel, Belgian Boys appeared in 3/3 answers from each provider for ready-made crêpes (U04). For ready-made Belgian-style waffles (U03), it appeared in 2/3 OpenAI and 3/3 Meta answers. The mini-pancake need (U02) yielded 1/3 OpenAI and 0/3 Meta. All denominators concern exactly three saved answers to that question/provider on 5 November 2026.
Does three out of three mean a product will always be recommended?
No. Three out of three describes these three responses to this exact question in a short collection window. It cannot establish future behaviour, performance in another market, a changed prompt or a personalised app session. Tarte's selected answers were not repeated in this way.
Were the product facts and retailer offers checked independently?
No. The studies analysed what model answers said. Prices, inventory, ingredients, product suitability, dietary labels, shipping information and related claims were not independently validated. They should not be treated as current consumer advice.
Did a citation or a product page cause a recommendation?
The studies do not establish that. Visible source annotations identify some cited material, but not a complete causal path through source retrieval, ranking and synthesis. Neither investigation altered a page and conducted a controlled comparison that isolated a source's effect on recommendations.
Do these results measure what personalised ChatGPT or Meta AI users would see?
No. They describe the recorded OpenAI and Meta API-model responses under the source collection's settings and dates. The consumer apps can have different access, retrieval, context and personalisation conditions, none of which were established as equivalent here.
What the findings suggest for future research
The next rigorous test would start by matching the question families and collection conditions across brands or categories, then repeat comparable questions over more than one collection window. It would preserve raw outputs, recommendation roles, product identity and citation/source records, and independently verify the product claims that matter to the shopper. A controlled content intervention, if justified by a diagnosed evidence gap, would be a different experiment with a documented baseline and retest.
These are proposals, not fixes shown to work by the Tarte or Belgian Boys collections.
For the wider measurement concepts, see Kojable's AI Visibility measurement framework and AI Brand Monitoring guide. For consumer-product research questions, AI Product Discovery explains Kojable's scoped approach to monitoring and diagnosing product representation across defined AI surfaces. The platform’s broader process is See → Align → Prove → Compound. A third-party model’s recommendation and any resulting sale remain independent outcomes.
Technical appendix A — Tarte Cosmetics evidence preserved for the replacement report
This appendix is part of the consolidated public evidence destination, not a second Tarte page. Its original source was Kojable's single-brand Tarte study, labelled published 5 November 2026, based on data collected 3 November 2026. The following prompt strings are transcribed from Kojable's retained original Tarte research HTML. They should still be matched to saved API request and response identifiers during publication QA; this HTML transcription is not a full independent raw-record replay.
A1. Tarte original question text and what each panel can establish
| Original question ID | Question text from the saved research HTML | Panel | Research role |
|---|---|---|---|
Q01 | “I'm shopping in New York and need a full-coverage concealer for dark circles that won't look cakey. Which three products from different brands should I consider, and where can I buy them?” | Unbranded | Discovery and named concealer products |
Q02 | “I'm 45 with dry under-eyes and fine lines. What concealer gives good coverage without settling into lines? Compare three options.” | Unbranded | Need-specific discovery/absence |
Q03 | “What's the best long-wearing foundation for combination or oily skin under $45 USD in the United States? Compare coverage, finish and value across three brands. The budget is before sales tax and shipping.” | Unbranded | Discovery/absence within a budgeted need |
Q04 | “I want a tubing mascara under $35 USD, available in the United States, that won't smudge during a long day and removes without harsh rubbing. What should I buy? The budget is before sales tax and shipping.” | Unbranded | Distinct Tarte inclusion in one selected provider answer |
Q05 | “I'm looking for a juicy, glossy, slightly plumping lip product under $35 USD, available in the United States. Which three brands and products are worth trying? The budget is before sales tax and shipping.” | Unbranded | Distinct Tarte inclusion in the other selected provider answer |
Q06 | “Is Tarte Shape Tape still worth buying, or should I choose NARS Radiant Creamy Concealer or e.l.f. Hydrating Camo Concealer? What are the trade-offs for dry under-eyes?” | Branded | Consideration of an already named brand, not discovery |
Q07 | “Tarte has several Shape Tape concealers. Which version suits oily, blemish-prone skin versus dry under-eyes, and how should I choose my shade?” | Branded | Variant discussion/conditional product framing |
Q08 | “I've seen Tarte Maracuja Juicy Lip recommended on social media. Is it worth the hype, and what do reviewers say compared with similar lip products?” | Branded | Review-attribution/claim interpretation, not independent product validation |
Q09 | “I only buy cruelty-free makeup and prefer vegan formulas. Is Tarte entirely vegan, and how can I verify whether a specific concealer or mascara is vegan?” | Branded | Policy explanation as stated by models, not independently verified policy truth |
Q10 | “I'm in the United States and ready to buy Tarte Shape Tape concealer and Tartelette tubing mascara. What are the current prices in USD ($), total cost, US shipping threshold and most direct place to purchase? Stop before checkout.” | Branded | Buying-facilitation response as recorded, not live retailer facts |
A2. Complete target-brand discovery inclusion matrix: Q01 yes/yes; Q02 no/no; Q03 no/no; Q04 yes/no; Q05 no/yes, in OpenAI/Meta order. Each cell is one selected saved answer. Sum: 2/5 OpenAI, 2/5 Meta, with no repeats. The main-body table supplies the full labelled rows.
A3. Named product examples: In Q01, the recorded OpenAI concealer name differed from Meta's Shape Tape label; in Q04, the saved OpenAI answer named Tartelette XL; in Q05, the saved Meta answer named Maracuja Juicy Lip Plump. These are model-output strings at collection, not verified exact catalogue matches. Named alternatives in other brands appeared in the original case but do not change Tarte's target-brand numerator.
A3a. Qualifying narrator-selected brands in the five unbranded Tarte answers. This preserves the comparison behind the target-brand numerator. These are coded historical answer choices, not current purchasing advice or an ordered product ranking. The saved record excluded a Maybelline mention in Meta's Q01 answer from its narrator-only recommendation table because the recommendation voice was unclear.
| Question | OpenAI narrator-selected brands | Meta Muse Spark narrator-selected brands |
|---|---|---|
Q01 Full-coverage concealer | Bobbi Brown; HUDA BEAUTY; Tarte Cosmetics | NARS; Tarte Cosmetics |
Q02 Dry under-eyes | Dior; Kosas; NARS | Kosas; Maybelline; NARS |
Q03 Oily/combination foundation | L'Oréal Paris; Maybelline; Revlon | L'Oréal Paris; Maybelline; Revlon — different named Maybelline formula |
Q04 Tubing mascara | Tarte Cosmetics (Tartelette XL) | Thrive Causemetics |
Q05 Glossy/plumping lip | Fenty Beauty; Milani; Too Faced | Maybelline; Tarte Cosmetics; e.l.f. |
A4. Branded versus unbranded nuance: The saved Q06/Q07 answers discussed Shape Tape variants after being asked about the brand, while the two Q02 unbranded answers did not choose Tarte. This is a difference in prompting/answer scope, not proof that a page fix would cause future discovery.
A5. Historical Tarte source/annotation measurements. The original case had a separate source-analysis component. To preserve its headline trace while keeping the two-study recommendation synthesis central, these values are reproduced strictly as source-reported saved annotations:
| Original Tarte source measure | OpenAI | Meta Muse Spark |
|---|---|---|
| Selected usable answers, all ten questions | 10 | 10 |
| Saved citation occurrences, all ten answers | 43 | 155 |
| Saved retrieved-source records, all ten answers | 597 | 629 |
| Unbranded panel citation occurrences | 17 | 82 |
| Distinct URLs in those unbranded citation annotations | 16 | 7 |
Unbranded answers with a saved tartecosmetics.com citation | 1/5 | 0/5 |
Branded answers with a saved tartecosmetics.com citation | 5/5 | 3/5 |
Interpretation boundary: Citation occurrences count repeated emitted annotation events, whereas distinct URLs and domain reach are different, separately deduplicated measures. Retrieved-source records are not a list of sources proved used in final answer synthesis. Provider capture formats differed; some saved responses lacked comparable annotations. Neither the higher 155-event total nor a domain's presence or absence demonstrates evidence strength, product truth, retrieval causation or recommendation quality. The original case's detailed source-by-domain classifications and source-type chart require faithful migration into an accessible archive before the historical case page can be redirected without losing its material source-analysis content.
A6. Collection/review boundary: One selected answer/provider/question; OpenAI initial collection and Meta later first-successful recovery, after initial failures; automated statement coding and Kojable report review without a documented separate independent human coding audit. The relevant source dataset, analysis tables and 20 selected raw responses are retained in Kojable's research workspace. This draft has not independently replayed all those files.
Open the historical Tarte evidence, including the full source and citation analysis.
Historical Tarte evidence
The following material comes from the original Kojable Tarte study, collected on 3 November 2026 and published on 5 November 2026. The answer summaries, excerpts and coded tables document the selected API answers and retain the original study’s units and caveats.
Open a question to read the two provider summaries and their paired interpretation. The original figures are represented by their underlying tables and captions.
What the saved answers show
The same brand appears equally often in the two saved discovery panels—but the products and shopper situations differ.
Kojable examined ten US shopping questions: five without Tarte in the prompt and five naming the brand. We compared one saved answer from each provider for each question, giving 20 answers collected on 3 November 2026. These are observations from a small sample, not a ranking of the providers.
Equal frequency conceals different choices
Tarte is recommended in 2/5 OpenAI and 2/5 Meta unbranded answers. Both include Tarte for full-coverage concealer. In this sample, the OpenAI mascara answer selects Tartelette XL, while the Meta glossy-lip answer includes Maracuja Juicy Lip Plump. Neither provider’s answer selects Tarte for dry under-eyes or oily/combination foundation. Q01 · Q05
The shopper’s need changes the choice
Both answers distinguish original Shape Tape for coverage and wear from Creamy for dry under-eyes. Recognising Creamy in a branded question does not mean selecting it when the shopper asks without naming Tarte. Product variants and conditions carry more meaning than a single brand sentiment. Q02 · Q06 · Q07
Different choices are not necessarily factual conflicts
The foundation answers share three brands but choose different Maybelline formulas and emphasise different benefits. Other pairs agree on their core answer, including variant guidance, vegan-policy explanation and purchase facilitation. Reviewer disagreements remain attributed experiences. Q03 · Q07–Q10
More annotations do not mean broader source visibility
Meta retains 155 citation occurrences, versus 43 for OpenAI. Yet in the unbranded panel Meta’s 82 occurrences reference seven URLs, while OpenAI’s 17 reference sixteen. In branded answers, tartecosmetics.com appears in 3/5 eligible Meta answers and 5/5 OpenAI answers. These are different measures, not evidence-strength rankings. Source analysis
How to read this study. There is one answer per provider and question, with no repeated trials. Meta’s answers were collected later, after initial failures. The findings describe these saved responses; they do not establish consistent provider preferences or sales effects.
1 · Brand discovery
| Unbranded question | OpenAI | Meta Muse Spark |
|---|---|---|
| Q01 Full-coverage concealer | ● Tarte recommended | ● Tarte recommended |
| Q02 Dry under-eyes & fine lines | ○ No Tarte recommendation | ○ No Tarte recommendation |
| Q03 Oily / combination foundation | ○ No Tarte recommendation | ○ No Tarte recommendation |
| Q04 Tubing mascara | ● Tarte recommended | ○ No Tarte recommendation |
| Q05 Gloss & slight plumping | ○ No Tarte recommendation | ● Tarte recommended |
2 · Different choices for the same need
3 · Target-domain visibility
| Provider | Panel | tartecosmetics.com reach |
|---|---|---|
| OpenAI | unbranded | 1/5 eligible answers |
| Meta Muse Spark | unbranded | 0/5 eligible answers |
| OpenAI | branded | 5/5 eligible answers |
| Meta Muse Spark | branded | 3/5 eligible answers |
01 / Brand discovery
What do AI shopping systems recommend when people ask questions relevant to Tarte Cosmetics?
Only the five questions that do not name Tarte count toward discovery. A recommendation means the answer itself presents the product as a suitable choice for the shopper. A passing mention, quoted review, rejected option or instruction to buy an already-chosen product does not count.
| Unbranded question | OpenAI | Meta Muse Spark |
|---|---|---|
| Q01 Full-coverage concealer | ● Tarte recommended
| ● Tarte recommended
|
| Q02 Dry under-eyes & fine lines | ○ No Tarte recommendation
| ○ No Tarte recommendation
|
| Q03 Oily / combination foundation | ○ No Tarte recommendation
| ○ No Tarte recommendation
|
| Q04 Tubing mascara | ● Tarte recommended
| ○ No Tarte recommendation
|
| Q05 Gloss & slight plumping | ○ No Tarte recommendation
| ● Tarte recommended
|
Interpretation. The two panels have the same Tarte frequency but cover different needs. Absence here means absence from these saved selections; it does not show that Tarte lacks a suitable product or should have been selected. Named formulas are preserved.
Brand frequency — complete unbranded table
| Provider | Brand | Answers recommending | Panel denominator | Answers with any brand-linked role |
|---|---|---|---|---|
| Meta Muse Spark | Kosas | 1 | 5 | 1 |
| Meta Muse Spark | L'Oreal Paris | 1 | 5 | 1 |
| Meta Muse Spark | Maybelline | 3 | 5 | 4 |
| Meta Muse Spark | NARS | 2 | 5 | 2 |
| Meta Muse Spark | Revlon | 1 | 5 | 1 |
| Meta Muse Spark | Tarte Cosmetics | 2 | 5 | 2 |
| Meta Muse Spark | Thrive Causemetics | 1 | 5 | 1 |
| Meta Muse Spark | e.l.f. | 1 | 5 | 1 |
| OpenAI | Bobbi Brown | 1 | 5 | 1 |
| OpenAI | Dior | 1 | 5 | 1 |
| OpenAI | Fenty Beauty | 1 | 5 | 1 |
| OpenAI | HUDA BEAUTY | 1 | 5 | 1 |
| OpenAI | Kosas | 1 | 5 | 1 |
| OpenAI | L'Oreal Paris | 1 | 5 | 1 |
| OpenAI | Maybelline | 1 | 5 | 1 |
| OpenAI | Milani | 1 | 5 | 1 |
| OpenAI | NARS | 1 | 5 | 1 |
| OpenAI | Revlon | 1 | 5 | 1 |
| OpenAI | Tarte Cosmetics | 2 | 5 | 2 |
| OpenAI | Too Faced | 1 | 5 | 1 |
Product-level discovery — including competitor selections
| Provider | Brand | Exact communicated product / variant | Answers / 5 | Questions |
|---|---|---|---|---|
| OpenAI | Bobbi Brown | Skin Hydrating Full Coverage Longwear Concealer | 1 | Q01 |
| OpenAI | Dior | Dior Forever Skin Correct Concealer | 1 | Q02 |
| OpenAI | Fenty Beauty | Gloss Bomb Heat | 1 | Q05 |
| OpenAI | HUDA BEAUTY | #FauxFilter Matte Buildable Coverage Waterproof Concealer | 1 | Q01 |
| OpenAI | Kosas | Kosas Revealer Concealer | 1 | Q02 |
| OpenAI | L'Oreal Paris | Infallible 32 Hour Fresh Wear Foundation | 1 | Q03 |
| OpenAI | Maybelline | Super Stay Lumi-Matte Foundation | 1 | Q03 |
| OpenAI | Milani | Keep It Full Nourishing Lip Plumper | 1 | Q05 |
| OpenAI | NARS | NARS Radiant Creamy Concealer | 1 | Q02 |
| OpenAI | Revlon | ColorStay Longwear Makeup for Combination/Oily Skin SPF 15 | 1 | Q03 |
| OpenAI | Tarte Cosmetics | Maracuja Creaseless Full Coverage Radiant Undereye Concealer | 1 | Q01 |
| OpenAI | Tarte Cosmetics | Tartelette XL Tubing Mascara · variant: XL | 1 | Q04 |
| OpenAI | Too Faced | Lip Injection Lip Gloss | 1 | Q05 |
| Meta Muse Spark | Kosas | Revealer Super Creamy + Brightening Concealer | 1 | Q02 |
| Meta Muse Spark | L'Oreal Paris | Infallible 32 Hour Fresh Wear Foundation · variant: 32 Hour | 1 | Q03 |
| Meta Muse Spark | Maybelline | Fit Me Matte + Poreless Liquid Foundation | 1 | Q03 |
| Meta Muse Spark | Maybelline | Instant Age Rewind Eraser | 1 | Q02 |
| Meta Muse Spark | Maybelline | Lifter Gloss Plumping Lip Gloss with Hyaluronic Acid | 1 | Q05 |
| Meta Muse Spark | NARS | Radiant Creamy Concealer | 2 | Q01 · Q02 |
| Meta Muse Spark | Revlon | ColorStay Longwear Makeup for Combination/Oily Skin SPF 15 · variant: Combination/Oily Skin SPF 15 | 1 | Q03 |
| Meta Muse Spark | Tarte Cosmetics | Maracuja Juicy Lip Plump | 1 | Q05 |
| Meta Muse Spark | Tarte Cosmetics | Shape Tape Concealer | 1 | Q01 |
| Meta Muse Spark | Thrive Causemetics | Liquid Lash Extensions Tubing Mascara | 1 | Q04 |
| Meta Muse Spark | e.l.f. | Pout Clout Lip Plumping Pen | 1 | Q05 |
Recommended brands by question and provider
Frozen original figure. All ten questions are shown; only Q01–Q05 measure unbranded discovery. Cell values count qualifying narrator statements, not rank or share of shoppers. Read the answer-frequency table separately. Underlying values are retained in the accompanying tables.
| Question | Panel | OpenAI | Meta Muse Spark |
|---|---|---|---|
| Q01 | unbranded | Bobbi Brown (1); HUDA BEAUTY (1); Tarte Cosmetics (1) | NARS (1); Tarte Cosmetics (2) |
| Q02 | unbranded | Dior (1); Kosas (2); NARS (1) | Kosas (1); Maybelline (1); NARS (1) |
| Q03 | unbranded | L'Oreal Paris (1); Maybelline (1); Revlon (1) | L'Oreal Paris (1); Maybelline (2); Revlon (2) |
| Q04 | unbranded | Tarte Cosmetics (5) | Thrive Causemetics (2) |
| Q05 | unbranded | Fenty Beauty (1); Milani (2); Too Faced (1) | Maybelline (2); Tarte Cosmetics (2); e.l.f. (2) |
| Q06 | branded | NARS (1); Tarte Cosmetics (2); e.l.f. (1) | NARS (2); Tarte Cosmetics (2); e.l.f. (3) |
| Q07 | branded | Tarte Cosmetics (4) | Tarte Cosmetics (5) |
| Q08 | branded | Clinique (1); Summer Fridays (1); Tarte Cosmetics (2) | LANEIGE (1); NYX (1); Tarte Cosmetics (1) |
| Q09 | branded | None (0) | None (0) |
| Q10 | branded | None (0) | None (0) |
The five discovery questions
Q01 · Full-coverage concealer
I'm shopping in New York and need a full-coverage concealer for dark circles that won't look cakey. Which three products from different brands should I consider, and where can I buy them?
Open the paired explorer for both summaries, product choices, conditions and short answer excerpts.
Q02 · Dry under-eyes & fine lines
I'm 45 with dry under-eyes and fine lines. What concealer gives good coverage without settling into lines? Compare three options.
Open the paired explorer for both summaries, product choices, conditions and short answer excerpts.
Q03 · Oily / combination foundation
What's the best long-wearing foundation for combination or oily skin under $45 USD in the United States? Compare coverage, finish and value across three brands. The budget is before sales tax and shipping.
Open the paired explorer for both summaries, product choices, conditions and short answer excerpts.
Q04 · Tubing mascara
I want a tubing mascara under $35 USD, available in the United States, that won't smudge during a long day and removes without harsh rubbing. What should I buy? The budget is before sales tax and shipping.
Open the paired explorer for both summaries, product choices, conditions and short answer excerpts.
Q05 · Gloss & slight plumping
I'm looking for a juicy, glossy, slightly plumping lip product under $35 USD, available in the United States. Which three brands and products are worth trying? The budget is before sales tax and shipping.
Open the paired explorer for both summaries, product choices, conditions and short answer excerpts.
02 / Model-to-model comparison
How do OpenAI and Meta Muse Spark differ when answering the same shopping question?
Explore all ten matched questions. Branded questions are labelled by their purpose and are not added to the discovery counts. Blue identifies OpenAI; orange identifies Meta, with text labels throughout.
| Question | Purpose | Recorded comparison | Shared qualifying brands |
|---|---|---|---|
| Q01 Full-coverage concealer | Unbranded discovery | preference difference | Tarte Cosmetics |
| Q02 Dry under-eyes & fine lines | Unbranded discovery | conditional difference | Kosas, NARS |
| Q03 Oily / combination foundation | Unbranded discovery | preference difference | L'Oreal Paris, Maybelline, Revlon |
| Q04 Tubing mascara | Unbranded discovery | preference difference | No shared qualifying product recommendations |
| Q05 Gloss & slight plumping | Unbranded discovery | preference difference | No shared qualifying product recommendations |
| Q06 Shape Tape versus alternatives | Branded · consideration | preference difference | NARS, Tarte Cosmetics, e.l.f. |
| Q07 Shape Tape variants & shade | Branded · product evaluation | agreement | Tarte Cosmetics |
| Q08 Maracuja value & reviewer opinions | Branded · product evaluation | preference difference | Tarte Cosmetics |
| Q09 Cruelty-free & vegan guidance | Branded · policy explanation | agreement | No shared qualifying product recommendations |
| Q10 Purchase facilitation | Branded · purchase facilitation | agreement | No shared qualifying product recommendations |
Shared brands and exact-product names
Frozen overlap figure, all ten matched questions. Ratio = shared distinct qualifying recommendations / union within each pair. Empty unions are not applicable. Naming and role conventions can reduce exact-name overlap even where the reviewed prose finds substantive agreement; Q07 is an example. Underlying values are retained in the accompanying tables.
| Question | Panel | Brand overlap | Exact product overlap |
|---|---|---|---|
| Q01 | unbranded | 1/4 | 0/5 |
| Q02 | unbranded | 2/4 | 2/4 |
| Q03 | unbranded | 3/3 | 2/4 |
| Q04 | unbranded | 0/2 | 0/2 |
| Q05 | unbranded | 0/6 | 0/6 |
| Q06 | branded | 3/3 | 3/4 |
| Q07 | branded | 1/1 | 1/4 |
| Q08 | branded | 1/5 | 0/4 |
| Q09 | branded | Not applicable (empty union) | Not applicable (empty union) |
| Q10 | branded | Not applicable (empty union) | Not applicable (empty union) |
Q01 Full-coverage concealer (Unbranded discovery)
Unbranded discovery
I'm shopping in New York and need a full-coverage concealer for dark circles that won't look cakey. Which three products from different brands should I consider, and where can I buy them?
OpenAI
gpt-6-luna
The answer recommends three distinct-brand concealers (Tarte, Bobbi Brown, Huda Beauty) as options for full coverage without a cakey look, with brief formula descriptions, approximate Sephora prices, a general New York Sephora store list, and an application tip. All three products are treated as co-equal primary recommendations for the shopper's stated need; no product is rejected or treated as inferior. Descriptions of finish/texture are attributed to the brand or to Sephora's own listing language rather than independently verified by the narrator. Purchase guidance is general (Sephora online + NYC store list with Buy Online & Pick Up) rather than a confirmed in-stock assertion for any specific product or shade.
Qualifying narrator selections in the saved table: Bobbi Brown — Skin Hydrating Full Coverage Longwear Concealer; HUDA BEAUTY — #FauxFilter Matte Buildable Coverage Waterproof Concealer; Tarte Cosmetics — Maracuja Creaseless Full Coverage Radiant Undereye Concealer
Tarte recommended: Yes in this saved answer.
A creamy, hydrating, radiant-finish option made for under-eyes and dark circles.
Excerpt from the saved OpenAI answer. Product claims are unverified.
Meta Muse Spark
muse-spark-1.3-contributor
The answer recommends three concealers (NARS Radiant Creamy Concealer, Tarte Shape Tape Concealer, Maybelline Instant Age Rewind Eraser Dark Circles Treatment Concealer) as full-coverage, dark-circle-targeting, non-cakey options from three different brands, and supplies online retailers plus three Manhattan store addresses for purchase. Claims about coverage and crease/cakey resistance are largely marketing-style descriptions for each product individually; a caveat limits the Tarte product's non-cakey claim to thin application on non-dry/mature skin, creating a conditional nuance rather than a contradiction.
Qualifying narrator selections in the saved table: NARS — Radiant Creamy Concealer; Tarte Cosmetics — Shape Tape Concealer
Tarte recommended: Yes in this saved answer.
Buildable, full coverage with a radiant, natural-looking finish
Excerpt from the saved Meta Muse Spark answer. Product claims are unverified.
Recorded comparison
Selections and preferences. Meta's narrator recommends three concealers from three brands: NARS Radiant Creamy Concealer , Tarte Shape Tape Concealer (with a conditional caveat), and Maybelline Instant Age Rewind Eraser Dark Circles Treatment Concealer . OpenAI's narrator recommends Tarte Maracuja Creaseless Full Coverage Radiant Undereye Concealer , Bobbi Brown Skin Hydrating Full Coverage Longwear Concealer and HUDA BEAUTY #FauxFilter Matte Buildable Coverage Waterproof Concealer . The product sets do not overlap. Only Tarte appears in both answers, and it is a different formula in each: Shape Tape in Meta, Maracuja Creaseless in OpenAI.
Alternatives and conditions. Meta has no rejected options. Its conditions are that Tarte Shape Tape needs a thin layer and can crease on dry or mature skin when over-applied, and that shades should be checked under bright store lighting. OpenAI frames Bobbi Brown as a pick for shoppers wanting a natural finish. It frames Huda as a pick for those wanting matte, long-wearing, buildable coverage. It also notes that stock and shades vary by store, and that prices are approximate and before tax. Meta's purchase guidance is broad: several online retailers per product, plus three Manhattan store addresses (Sephora Times Square, Ulta Herald Square, Target Herald Square). OpenAI's is Sephora-only: about $31–$35 per product, pickup guidance, and a store-list link with Manhattan and Brooklyn locations. Meta gives no prices.
Reasons, caveats and attribution. Meta relies on marketing-style descriptions for coverage and non-cakey claims, and several of its slogan-like lines have an unclear speaker. OpenAI attributes some claims to Tarte or to Sephora's listing. Both answers rely on unverified product claims, and neither verifies stock. OpenAI's Bobbi Brown and Huda entries do not explicitly mention dark circles, and Huda is described as medium-to-full coverage. Meta's three products are all described as full-coverage and as covering dark circles. Both answers give thin-application advice. No same-product, same-condition narrator conflict exists. External verification was not performed.
Interpretation for this shopper. The answers differ mainly in which products they surface and in how they support purchase. Meta favours a wider retailer and in-store set. OpenAI gives Sephora pricing and pickup. The difference is in selection and presentation, not a demonstrated factual contradiction. The two Tarte products are different formulas, so they cannot be compared as one product.
Shared and exclusive — brand
| Saved overlap category | Qualifying records |
|---|---|
| Shared | Tarte Cosmetics |
| Meta Muse Spark only | NARS |
| OpenAI only | Bobbi Brown; HUDA BEAUTY |
Shared and exclusive — exact product
| Saved overlap category | Qualifying records |
|---|---|
| Shared | None / not applicable |
| Meta Muse Spark only | nars | radiant creamy concealer | variant unstated; tarte cosmetics | shape tape concealer | variant unstated |
| OpenAI only | bobbi brown | skin hydrating full coverage longwear concealer | variant unstated; huda beauty | #fauxfilter matte buildable coverage waterproof concealer | variant unstated; tarte cosmetics | maracuja creaseless full coverage radiant undereye concealer | variant unstated |
Q02 Dry under-eyes & fine lines (Unbranded discovery)
Unbranded discovery
I'm 45 with dry under-eyes and fine lines. What concealer gives good coverage without settling into lines? Compare three options.
OpenAI
gpt-6-luna
The answer compares three named concealers (Kosas Revealer, NARS Radiant Creamy Concealer, Dior Forever Skin Correct) for a 45-year-old with dry under-eyes and fine lines who wants good coverage without creasing. It gives a table of coverage/finish, reasons, and watch-outs for each, plus a bottom-line set of conditional picks (Kosas for hydration, NARS for balanced everyday wear, Dior for stronger coverage) and general application advice. It also states an explicit caveat that no concealer guarantees zero creasing and that list prices shown are current U.S. prices found during the search. No merchant, stock, or promotional facts are stated.
Qualifying narrator selections in the saved table: Dior — Dior Forever Skin Correct Concealer; Kosas — Kosas Revealer Concealer; NARS — NARS Radiant Creamy Concealer
Tarte recommended: No in this saved answer.
My first one to try for your dry under-eyes.
Excerpt from the saved OpenAI answer. Product claims are unverified.
Meta Muse Spark
muse-spark-1.3-contributor
The answer compares three concealers (Maybelline Instant Age Rewind Eraser, NARS Radiant Creamy Concealer, Kosas Revealer Super Creamy + Brightening Concealer) for a 45-year-old shopper with dry under-eyes and fine lines seeking good coverage that resists settling into lines, explicitly fulfilling the 'compare three options' request. Each product gets a coverage/finish description, benefit bullets tied to anti-creasing/hydration, and a caveat, followed by a conditional pick-by-scenario summary and general anti-settling tips. No price, stock, merchant, promotion or fulfilment facts are present anywhere in the answer.
Qualifying narrator selections in the saved table: Kosas — Revealer Super Creamy + Brightening Concealer; Maybelline — Instant Age Rewind Eraser; NARS — Radiant Creamy Concealer
Tarte recommended: No in this saved answer.
**Want crease-resistant + brightening on a budget:** Maybelline Age Rewind
Excerpt from the saved Meta Muse Spark answer. Product claims are unverified.
Recorded comparison
Selections and preferences. Meta (narrator) presents three products: Maybelline Instant Age Rewind Eraser (conditional, budget, crease-resistant and brightening, not for very dark circles), NARS Radiant Creamy (conditional, for fuller coverage of dark circles while hydrating), and Kosas Revealer (the pick for the driest skin with the most lines and a fresh look, effectively the primary match). OpenAI (narrator) presents Kosas Revealer as its 'first one to try' for dry under-eyes, NARS Radiant Creamy as a balanced everyday option, and Dior Forever Skin Correct for when coverage is the priority. Both answers lead with Kosas for the shopper's dry under-eyes and both include NARS. They differ on the third option: Meta chooses Maybelline and OpenAI chooses Dior.
Alternatives and conditions. Meta's picks are conditioned on sub-scenarios (driest skin and most lines, dark-circle coverage, budget). It also adds a rejected-option caveat for Maybelline and Kosas on very or extremely dark circles. OpenAI's picks are conditioned on the hydrated feel it attributes to Kosas, a balanced everyday result for NARS, and coverage priority for Dior. It also flags that NARS may be less comfortable on very dry under-eyes. The two answers frame NARS differently. Meta calls it the higher-coverage, hydrating option for dark circles, while OpenAI calls it a middle-ground, medium-coverage option. Meta describes Kosas as medium to full coverage, while OpenAI describes it as medium and buildable.
Reasons, caveats and attribution. Meta reasons from hydrating ingredients and long-wear and anti-crease claims. It also quotes an unnamed review, 'users over 50', and unnamed 'experts'. It contains an unreconciled tension on NARS between 'looks better over time' on very dry skin and a review reporting creasing after 4-6 hours. OpenAI says outright that no concealer can promise zero creasing. It gives US list prices and attributes some support to unnamed 'beauty editors' and 'some people'. Meta gives no price or merchant information. External facts and the claimed wear times were not verified. The two answers also differ on how well NARS suits very dry under-eyes: Meta presents it as suited to dry or mature under-eyes, while OpenAI cautions about comfort. Each side's NARS statement is partly tied to a reviewer or unnamed group, so this is not a clean narrator-versus-narrator conflict.
Interpretation for this shopper. The answers largely agree on Kosas as the best fit for dry under-eyes and on NARS as a viable option. They diverge in the third product, in how they position NARS's coverage level, and in their NARS dryness caveats. The differences are mainly differences in preference and conditions, plus differences in the evidence and caveats each provides. They are not established factual conflicts.
Shared and exclusive — brand
| Saved overlap category | Qualifying records |
|---|---|
| Shared | Kosas; NARS |
| Meta Muse Spark only | Maybelline |
| OpenAI only | Dior |
Shared and exclusive — exact product
| Saved overlap category | Qualifying records |
|---|---|
| Shared | kosas | revealer concealer (reviewed naming group) | variant unstated; nars | radiant creamy concealer | variant unstated |
| Meta Muse Spark only | maybelline | instant age rewind eraser | variant unstated |
| OpenAI only | dior | dior forever skin correct concealer | variant unstated |
Q03 Oily / combination foundation (Unbranded discovery)
Unbranded discovery
What's the best long-wearing foundation for combination or oily skin under $45 USD in the United States? Compare coverage, finish and value across three brands. The budget is before sales tax and shipping.
OpenAI
gpt-6-luna
The answer recommends Maybelline Super Stay Lumi-Matte as the best overall long-wearing foundation for combination/oily skin, with Revlon ColorStay Longwear (Combination/Oily, SPF 15) positioned as a conditional alternative for maximum shine control and L'Oréal Infallible 32 Hour Fresh Wear positioned as a conditional alternative for a more natural finish. It presents a three-row comparison table (coverage, finish, listed price/value) and a closing 'quick guide' with explicit caveats that wear-time figures are brand 'up-to' claims and that prices are pre-tax/pre-shipping. No stock, merchant, or purchase-link information is given; only brand-site citation URLs are present.
Qualifying narrator selections in the saved table: L'Oreal Paris — Infallible 32 Hour Fresh Wear Foundation; Maybelline — Super Stay Lumi-Matte Foundation; Revlon — ColorStay Longwear Makeup for Combination/Oily Skin SPF 15
Tarte recommended: No in this saved answer.
For most combination or oily skin, it offers the best balance of long wear, buildable coverage, a not-too-flat finish, and price.
Excerpt from the saved OpenAI answer. Product claims are unverified.
Meta Muse Spark
muse-spark-1.3-contributor
The answer compares three specific under-$45 US foundations (L'Oréal Infallible 32 Hour Fresh Wear, Revlon ColorStay Combination/Oily SPF15, Maybelline Fit Me Matte + Poreless) on coverage, finish, wear time and per-ounce value, mostly restating brand-page language, and ends with a three-way verdict assigning each product to a different shopper priority (longest wear, matte+SPF, lowest cost). It also contains one ambiguous sentence referencing a differently-named 'L'Oréal Infallible 24H Fresh Wear' and a separate Fit Me Primer SPF 20, both incidental to the three main picks. All factual wear/coverage/shade/price claims originate from brand copy and are not independently verified here.
Qualifying narrator selections in the saved table: L'Oreal Paris — Infallible 32 Hour Fresh Wear Foundation · variant: 32 Hour; Maybelline — Fit Me Matte + Poreless Liquid Foundation; Revlon — ColorStay Longwear Makeup for Combination/Oily Skin SPF 15 · variant: Combination/Oily Skin SPF 15
Tarte recommended: No in this saved answer.
For **longest wear** under $45 for combination or oily skin, choose **L'Oréal Infallible 32 Hour Fresh Wear** - 32-hour transfer-resistant wear vs 24-hour for Revlon and 12-hour shine control for Maybelline, while still suited to combination to oily skin struggling with midday shine
Excerpt from the saved Meta Muse Spark answer. Product claims are unverified.
Recorded comparison
Selections and preferences. Meta (narrator) picks L'Oréal Infallible 32 Hour Fresh Wear for longest wear . It names Revlon ColorStay Combination/Oily as the conditional pick for matte oil control plus SPF 15 and Maybelline Fit Me Matte + Poreless for lowest cost and everyday medium coverage . OpenAI (narrator) picks Maybelline Super Stay Lumi-Matte as best overall . It names Revlon as conditional for oilier skin or maximum matte control and L'Oréal Infallible 32 Hour Fresh Wear as conditional for a more natural finish . The answers share Revlon and L'Oréal but differ on the third brand (Fit Me Matte + Poreless versus Super Stay Lumi-Matte) and on the top pick.
Alternatives and conditions. Ranking review: OpenAI does not state that L'Oréal is ranked last. It places L'Oréal third in the table and quick guide, but its saved recommendation is a need-specific conditional pick: "L’Oréal for a more natural finish", and "a good choice if you want long wear without a strongly matte look". The only explicit ranking language is Maybelline as "best overall", so L'Oréal's third position is list order, not a stated rank. Revlon is conditional for oilier skin or maximum matte control. Meta makes L'Oréal its longest-wear primary pick, while Revlon and Maybelline are need-specific alternatives. Meta's Maybelline product is Fit Me Matte + Poreless, a different formula from OpenAI's Super Stay Lumi-Matte. Meta also adds a note that Fit Me has no SPF and suggests a separate SPF 20 primer.
Reasons, caveats and attribution. Meta emphasizes wear hours (32/24/12), per-ounce price, SPF and matte oil control, mostly restating brand-page language. OpenAI emphasizes balance of wear, buildable coverage, a not-too-flat luminous-matte finish and price, with an explicit caveat that wear times are brand "up-to" claims and prices may differ by retailer. Meta has an unresolved internal naming inconsistency, an unexplained "Infallible 24H Fresh Wear" sentence, and gives no merchant or URL. OpenAI cites brand-site URLs but gives no stock or merchant details, and US availability is only implied. Meta says Revlon is "matte oil-absorbing" and L'Oréal "natural", and OpenAI says the same, so these descriptions broadly agree. No factual verification was performed.
Interpretation for this shopper. This is a preference and criteria-weighting difference, not a factual conflict. Meta prioritizes maximum wear time, OpenAI prioritizes overall balance with a less flat finish, and the two answers evaluate a different third formula. OpenAI's L'Oréal third position should not be coded as a last-place ranking. Its explicit statements are Maybelline Super Stay Lumi-Matte as best overall and L'Oréal as the pick for a natural finish.
Shared and exclusive — brand
| Saved overlap category | Qualifying records |
|---|---|
| Shared | L'Oreal Paris; Maybelline; Revlon |
| Meta Muse Spark only | None |
| OpenAI only | None |
Shared and exclusive — exact product
| Saved overlap category | Qualifying records |
|---|---|
| Shared | l'oreal paris | infallible 32 hour fresh wear foundation | variant in communicated name; revlon | colorstay longwear makeup for combination/oily skin spf 15 | variant in communicated name |
| Meta Muse Spark only | maybelline | fit me matte + poreless liquid foundation | variant unstated |
| OpenAI only | maybelline | super stay lumi-matte foundation | variant unstated |
Q04 Tubing mascara (Unbranded discovery)
Unbranded discovery
I want a tubing mascara under $35 USD, available in the United States, that won't smudge during a long day and removes without harsh rubbing. What should I buy? The budget is before sales tax and shipping.
OpenAI
gpt-6-luna
The answer recommends a single product—Tarte's Tartelette XL Tubing Mascara—as the primary purchase, citing its $28 Ulta price (under the shopper's $35 budget), in-stock status, Allure's endorsement as a top tubing-mascara pick with smudge/flake resistance, and removal instructions requiring only warm water and gentle pressure (no hard rubbing). The narrator adds a personal caveat that results can vary and that this is 'the one I'd try first,' blending the narrator's own recommendation with reviewer- and brand-sourced claims.
Qualifying narrator selections in the saved table: Tarte Cosmetics — Tartelette XL Tubing Mascara · variant: XL
Tarte recommended: Yes in this saved answer.
**Buy Tartelette XL Tubing Mascara**
Excerpt from the saved OpenAI answer. Product claims are unverified.
Meta Muse Spark
muse-spark-1.3-contributor
The answer recommends one product — Thrive Causemetics Liquid Lash Extensions Tubing Mascara — as a direct purchase pick, supporting the recommendation with a quoted price ($26.00 USD, stated as under the $35 budget before tax/shipping), a quoted tubing-formula description, quoted brand/product-page language about smudge/flake resistance, quoted removal instructions emphasizing warm water rather than harsh rubbing, and a brand address plus a single variant's 'Available' status as US-availability support. Most explicit shopper needs are addressed through quoted product-page language, but the US-availability requirement and general purchase guidance are only partially supported, since a corporate address and one variant's status are not firm evidence of nationwide stock or purchasability. Several supportive quotes are of unclear or brand-sourced origin and are not independently verified.
Qualifying narrator selections in the saved table: Thrive Causemetics — Liquid Lash Extensions Tubing Mascara
Tarte recommended: No in this saved answer.
“tubing formula wraps each lash in lightweight ‘tubes’”
Excerpt from the saved Meta Muse Spark answer. Product claims are unverified.
Recorded comparison
Selections and preferences. Both answers give one primary "buy" pick, and they pick different products. The Meta answer tells the shopper to buy Thrive Causemetics Liquid Lash Extensions Tubing Mascara . The OpenAI answer tells the shopper to buy Tartelette XL Tubing Mascara, "the XL version, not the original" . Both place their pick under $35: Meta lists $26.00 USD and OpenAI lists $28 at Ulta. Neither names a ranked alternative.
Alternatives and conditions. Neither answer offers named alternatives. The OpenAI answer only contrasts the XL with an unspecified "original" Tartelette. The Meta answer's availability support is a corporate address and one variant, Crystal (Brown Black), marked "Available" . It gives no purchase link or retailer destination, and the variant looks illustrative rather than recommended. The OpenAI answer names Ulta, gives a literal ulta.com product URL as a citation beside the buy sentence, and says the item is currently listed in stock. The Meta answer says explicitly that $26.00 is before tax and shipping. The OpenAI answer does not state whether $28 excludes tax or shipping.
Reasons, caveats and attribution. The Meta answer relies on quoted product-page or brand-style language for tubing formula, smudge-proofing and warm-water removal, and most of those quotes have unclear voice. Its US availability support is partial and has a source presentation limitation: a corporate address sits under the heading "U.S. availability". The OpenAI answer relies on Allure, attributed as reviewer, for top-pick status, smudge and flake resistance, and warm-water removal. It adds narrator removal instructions, a caveat that "Individual results can vary," and a first-person "I'd try first." On the markup review focus, the Ulta citation is ordinary valid Markdown: [ulta.com](URL) inside text parentheses. The utm_source=openai parameter does not establish malformation or a broken destination. The earlier "malformed" label is not supported by the corrected OpenAI analysis , which records valid syntax. The only genuine limitations are that the URL, stock and price were not verified, and that the answer gives no explicit tax or shipping basis. External factual verification was not performed.
Interpretation for this shopper. This is a preference difference between two different product picks that both meet the stated budget. It is not a factual contradiction, because the answers make no incompatible narrator assertions about the same product and condition. Their purchase guidance also differs: OpenAI gives a named retailer, stock status and a citation URL, while Meta gives a brand address and one variant status. Neither answer verifies its claims. The apparent markup-malformation concern about the Ulta link is not supported and should not be kept as a source defect.
Shared and exclusive — brand
| Saved overlap category | Qualifying records |
|---|---|
| Shared | None / not applicable |
| Meta Muse Spark only | Thrive Causemetics |
| OpenAI only | Tarte Cosmetics |
Shared and exclusive — exact product
| Saved overlap category | Qualifying records |
|---|---|
| Shared | None / not applicable |
| Meta Muse Spark only | thrive causemetics | liquid lash extensions tubing mascara | variant unstated |
| OpenAI only | tarte cosmetics | tartelette xl tubing mascara | xl |
Q05 Gloss & slight plumping (Unbranded discovery)
Unbranded discovery
I'm looking for a juicy, glossy, slightly plumping lip product under $35 USD, available in the United States. Which three brands and products are worth trying? The budget is before sales tax and shipping.
OpenAI
gpt-6-luna
The answer recommends three specific lip-gloss products (Milani Keep It Full Nourishing Lip Plumper, Fenty Beauty Gloss Bomb Heat, Too Faced Lip Injection Lip Gloss) with listed prices under $35, framed as juicy/glossy/mildly plumping options, and singles out Milani and Too Faced as the closest match for 'slightly plumping.' It states all three prices are under the pre-tax, pre-shipping budget, but never explicitly confirms U.S. availability or gives stock/purchase/merchant information; citation links are provenance only.
Qualifying narrator selections in the saved table: Fenty Beauty — Gloss Bomb Heat; Milani — Keep It Full Nourishing Lip Plumper; Too Faced — Lip Injection Lip Gloss
Tarte recommended: No in this saved answer.
A shiny, sheer gloss with a **slight cooling sensation** and a naturally enhanced look—my pick if you want a gentler, budget-friendly option.
Excerpt from the saved OpenAI answer. Product claims are unverified.
Meta Muse Spark
muse-spark-1.3-contributor
The answer lists three distinct lip-plumping products (Tarte Maracuja Juicy Lip Plump, e.l.f. Pout Clout Lip Plumping Pen, Maybelline Lifter Gloss) each pitched as primary picks meeting the shopper's juicy/glossy/slightly-plumping, under-$35, US-available brief, with per-product 'Best if' framing that functions as conditional narrower recommendations rather than universal endorsements. Purchase-relevant facts (price, stock status, merchant, shipping messaging) are decomposed separately from marketing/benefit claims. One internal description tension (e.l.f. pen called both 'clear' and having a 'sheer cherry color') is flagged as a potential, unresolved consistency issue; all other product benefit/caveat claims are retained or split from quoted marketing copy without external verification.
Qualifying narrator selections in the saved table: Maybelline — Lifter Gloss Plumping Lip Gloss with Hyaluronic Acid; Tarte Cosmetics — Maracuja Juicy Lip Plump; e.l.f. — Pout Clout Lip Plumping Pen
Tarte recommended: Yes in this saved answer.
### 1. tarte - Maracuja Juicy Lip Plump
Excerpt from the saved Meta Muse Spark answer. Product claims are unverified.
Recorded comparison
Selections and preferences. The two answers recommend entirely different product sets, with no overlap in brands or products. Meta (narrator) gives three co-equal picks: tarte Maracuja Juicy Lip Plump ($27.00, Sephora U.S.), e.l.f. Pout Clout Lip Plumping Pen ($9, elfcosmetics.com) and Maybelline Lifter Gloss with Hyaluronic Acid ($8.99 online at Target). OpenAI (narrator) gives Milani Keep It Full Nourishing Lip Plumper ($11.99), Fenty Beauty Gloss Bomb Heat ($26) and Too Faced Lip Injection Lip Gloss ($28), and singles out Milani or Too Faced as the best start for "slightly plumping". It frames Fenty as an alternative for a classic gloss feel.
Alternatives and conditions. Meta attaches a "Best if" condition to each pick: tarte for color, treatment and plump in one pen; e.l.f. for the most budget-friendly, sheer, cooling tingle rather than an extreme plumper; Maybelline for high shine and hyaluronic acid hydration without a strong tingle. OpenAI conditions Milani on a gentler, budget-friendly option and Fenty on a classic gloss feel. It names Milani and Too Faced for slight plumping, and Too Faced is described as entry-level with cooling rather than burning sensation. Both answers use the same shopper constraints of juicy, glossy, slight plumping, under $35 and pre-tax/shipping. Neither names the other's products as alternatives.
Reasons, caveats and attribution. Meta relies heavily on quoted brand/retailer copy (hydration, hyaluronic acid, finish) and gives merchant, stock status and shipping messaging. Its e.l.f. description is internally unresolved: it is called a "clear" gloss and also "a sheer cherry color", with no shade named. OpenAI relies on its own descriptions plus citation links, and gives only listed prices. It does not state U.S. availability, stock or merchant, and the citation URLs are provenance only. OpenAI adds a safety caveat that plumping sensations vary and to stop use if irritated. Meta covers U.S. availability explicitly ("listed in the U.S."). Value differs: Meta's picks run $8.99 to $27.00, OpenAI's $11.99 to $28. No external verification was done, and no factual conflict arises because no product is shared.
Interpretation for this shopper. This is a difference in which products are recommended and how the answers frame intensity and purchase support, not a factual contradiction. Meta provides richer purchase guidance and explicit U.S. coverage. OpenAI provides explicit guidance on which picks fit "slightly plumping" best, plus a safety note. Differences in recommended product sets cannot be settled from the saved answers alone.
Shared and exclusive — brand
| Saved overlap category | Qualifying records |
|---|---|
| Shared | None / not applicable |
| Meta Muse Spark only | Maybelline; Tarte Cosmetics; e.l.f. |
| OpenAI only | Fenty Beauty; Milani; Too Faced |
Shared and exclusive — exact product
| Saved overlap category | Qualifying records |
|---|---|
| Shared | None / not applicable |
| Meta Muse Spark only | e.l.f. | pout clout lip plumping pen | variant unstated; maybelline | lifter gloss plumping lip gloss with hyaluronic acid | variant unstated; tarte cosmetics | maracuja juicy lip plump | variant unstated |
| OpenAI only | fenty beauty | gloss bomb heat | variant unstated; milani | keep it full nourishing lip plumper | variant unstated; too faced | lip injection lip gloss | variant unstated |
Q06 Shape Tape versus alternatives (Branded · consideration)
Branded · consideration
Is Tarte Shape Tape still worth buying, or should I choose NARS Radiant Creamy Concealer or e.l.f. Hydrating Camo Concealer? What are the trade-offs for dry under-eyes?
OpenAI
gpt-6-luna
The answer gives a conditional three-way comparison for dry under-eyes: it leans toward NARS Radiant Creamy for a natural/radiant finish or e.l.f. Hydrating Camo for budget/full coverage, while cautioning against the *original* Tarte Shape Tape for dry/fine-line concerns unless the shopper prioritizes full coverage and long wear on under-eyes that aren't very dry. It separately flags Tarte Shape Tape Creamy as the more dry-skin-appropriate Tarte formula, distinct from the original. Price points ($8 for e.l.f., $32 for Tarte original) are stated as brand/retailer-listed figures with no stock, merchant, or promotion details attached. A general application tip closes the answer without being tied to one product.
Qualifying narrator selections in the saved table: NARS — Radiant Creamy Concealer; Tarte Cosmetics — Shape Tape Concealer · variant: Original; Tarte Cosmetics — Shape Tape Creamy Concealer · variant: Creamy; e.l.f. — Hydrating Camo Concealer
I’d pick NARS Radiant Creamy for the more polished, luminous look
The best starting point if you want a more natural, radiant finish.
Excerpt from the saved OpenAI answer. Product claims are unverified.
Meta Muse Spark
muse-spark-1.3-contributor
The answer evaluates three concealers (Tarte Shape Tape original, with a passing mention of the Shape Tape Medium Coverage Radiant variant, NARS Radiant Creamy Concealer, and e.l.f. Hydrating Camo Concealer) specifically against the shopper's three explicit asks: whether Shape Tape is still worth buying, how it compares to NARS and e.l.f., and the dry-under-eye trade-offs of each. It gives conditional guidance for each product rather than a single universal winner, cites prices for all three, and closes with a direct 'if truly dry, skip Shape Tape for everyday, get NARS or e.l.f.' recommendation. Purchase facts are limited to approximate prices; no stock, merchant, or fulfilment information is given.
Qualifying narrator selections in the saved table: NARS — Radiant Creamy Concealer; Tarte Cosmetics — Shape Tape Concealer · variant: Original; e.l.f. — Hydrating Camo Concealer
**Still worth buying if:** you want heavy coverage, brightening + a polished, set look that lasts all day, and you are willing to prep.
Excerpt from the saved Meta Muse Spark answer. Product claims are unverified.
Recorded comparison
Selections and preferences. Both answers steer a dry-under-eye shopper away from original Tarte Shape Tape as an everyday pick and toward NARS Radiant Creamy or e.l.f. Hydrating Camo. Meta (narrator) calls NARS the 'safest for dry, comfortable under-eyes' and gives it as the daily-comfort choice. It offers e.l.f. as the budget full-coverage option and keeps Shape Tape as conditionally worth it for heavy, long-wear, matte needs with prep. OpenAI (narrator) leads with NARS for a polished, luminous finish, or e.l.f. for full coverage on a budget. It says original Shape Tape is still worth it only if full coverage and long wear matter most and the under-eyes are not very dry.
Alternatives and conditions. Meta's alternatives: the Shape Tape Medium Coverage Radiant version, framed in an unattributed blockquote and positioned as great for dry skin. Its conditions are NARS for fine lines, dryness and mature skin, with less one-swipe coverage; e.l.f. for a tight budget with a thin layer, quick blending and light setting; Shape Tape for 12-16hr sweat-proof glam if the shopper accepts matte and preps. OpenAI's alternative is Tarte Shape Tape Creamy, a separate formula marketed for dry or mature skin. Its choice rule is NARS if finish matters most, e.l.f. if price and full coverage matter more. The in-family dry-skin options differ: Meta names the Radiant version, OpenAI names the Creamy version. These are different named variants, not a contradiction.
Reasons, caveats and attribution. Meta gives much more detail: wear hours, review quotes, mixed e.l.f. customer reports on dryness, oxidation and creasing, prices of about $32, $36 and $8, and a saving of about $28. Much of this comes from unnamed listings or reviewers, and several voices are unclear. OpenAI is shorter. It attributes the 24-hour hydration claim to NARS and the matte, combination/oily positioning to Tarte, and notes that the hydration claim does not guarantee no creasing. It prices only e.l.f. ($8 'currently') and Tarte original ($32). Neither answer gives stock, merchant or fulfilment guidance. External facts were not verified.
Interpretation for this shopper. The answers largely agree on the direction: avoid original Shape Tape for very dry under-eyes, and pick NARS for comfort or finish and e.l.f. for budget. The differences are in emphasis and depth. Meta ranks NARS as safest for dryness and reports mixed e.l.f. dryness experiences. OpenAI is more cautious about NARS's hydration claim and recommends Shape Tape Creamy where Meta recommends the Radiant variant. No narrator assertions conflict about the same product and condition.
Shared and exclusive — brand
| Saved overlap category | Qualifying records |
|---|---|
| Shared | NARS; Tarte Cosmetics; e.l.f. |
| Meta Muse Spark only | None |
| OpenAI only | None |
Shared and exclusive — exact product
| Saved overlap category | Qualifying records |
|---|---|
| Shared | e.l.f. | hydrating camo concealer | variant unstated; nars | radiant creamy concealer | variant unstated; tarte cosmetics | shape tape concealer | original |
| Meta Muse Spark only | None |
| OpenAI only | tarte cosmetics | shape tape creamy concealer | creamy |
Q07 Shape Tape variants & shade (Branded · product evaluation)
Branded · product evaluation
Tarte has several Shape Tape concealers. Which version suits oily, blemish-prone skin versus dry under-eyes, and how should I choose my shade?
OpenAI
gpt-6-luna
The answer gives a direct, narrator-voiced recommendation matching Tarte Shape Tape Original to oily/blemish-prone skin and Shape Tape Creamy to dry under-eyes, briefly mentions Shape Tape Radiant as an alternative for those wanting lighter/radiant coverage, and provides general shade-matching guidance (undertone coding, blemish vs. under-eye matching, use of shade finder/try-on). No prices, stock, merchant, promotion, or fulfilment facts are present; cited URLs are product-page citations only, not purchase links or stock evidence.
Qualifying narrator selections in the saved table: Tarte Cosmetics — Shape Tape Concealer · variant: Original; Tarte Cosmetics — Shape Tape Creamy Concealer
Original Shape Tape**—full coverage, natural-matte finish, and Tarte’s pick for oily or combination skin.
Excerpt from the saved OpenAI answer. Product claims are unverified.
Meta Muse Spark
muse-spark-1.3-contributor
The answer gives three shade/variant recommendations for the Tarte Shape Tape concealer line (Original for oily/blemish-prone skin, Ultra Creamy/Creamy for dry under-eyes, and Radiant as a lighter alternative for dry skin), plus general shade-matching guidance (undertone letters, depth placement, and Tarte's shade-finder tools). All reasons are drawn from brand/retailer descriptions and unattributed 'reviewer' paraphrase; no prices, stock, or merchant/purchase links are given anywhere in the answer.
Qualifying narrator selections in the saved table: Tarte Cosmetics — Original Shape Tape Concealer · variant: Original; Tarte Cosmetics — Shape Tape Creamy Concealer · variant: Ultra Creamy / Creamy; Tarte Cosmetics — Shape Tape Radiant Concealer · variant: Radiant
Choose the **Original Shape Tape Concealer.**
Excerpt from the saved Meta Muse Spark answer. Product claims are unverified.
Recorded comparison
Selections and preferences. Both answers make the same core picks. Original Shape Tape is for oily, blemish-prone skin. The Creamy version (Meta: "Ultra Creamy / Creamy") is for dry under-eyes. Radiant is a lighter alternative for dry skin. Meta names Original, Ultra Creamy/Creamy and Radiant as primary, primary and alternative. OpenAI names the same three, with Radiant offered as a preference-based alternative for lighter, radiant coverage. Both also suggest using Original on spots and Creamy under the eyes (Meta , OpenAI).
Alternatives and conditions. Meta adds a conditional split between Radiant (no-makeup, fresh coverage) and Ultra Creamy (full coverage with moisture) . It also flags Original as potentially heavy on dry, crepey or textured under-eyes . OpenAI does not give that warning about Original, and treats Radiant only briefly. Neither answer gives price, stock, merchant or fulfilment details.
Reasons, caveats and attribution. Meta supports its picks mostly with quoted brand, retailer and reviewer text. These quotes are not always attributed to a named source, and the "Many people own both" line is an unsupported generalisation. OpenAI gives shorter narrator-voiced reasons and cites Tarte product pages. It adds a caveat that Original covers the appearance of blemishes and is not an acne treatment, which Meta lacks. On shade, both say to match undertone and depth. Both say to go about one shade lighter under the eyes for brightening, and to match skin tone closely on blemishes. Both point to Tarte's shade finder or virtual try-on. Meta adds the full letter system (including G), a 1-2 shades lighter/deeper rule, a before/after photo tip and a natural-light jawline swatch. OpenAI gives a shorter B/N/S/H key and asks for the shopper's foundation details. No external verification was performed.
Interpretation for this shopper. The answers agree on the substantive recommendations and shade method. They differ mainly in depth, caveats and presentation: Meta gives more detail and a Radiant-versus-Creamy conditional, while OpenAI is more concise and adds the acne caveat. These are coverage differences, not conflicting claims.
Shared and exclusive — brand
| Saved overlap category | Qualifying records |
|---|---|
| Shared | Tarte Cosmetics |
| Meta Muse Spark only | None |
| OpenAI only | None |
Shared and exclusive — exact product
| Saved overlap category | Qualifying records |
|---|---|
| Shared | tarte cosmetics | shape tape concealer | original |
| Meta Muse Spark only | tarte cosmetics | shape tape creamy concealer | ultra creamy / creamy; tarte cosmetics | shape tape radiant concealer | radiant |
| OpenAI only | tarte cosmetics | shape tape creamy concealer | variant unstated |
Q08 Maracuja value & reviewer opinions (Branded · product evaluation)
Branded · product evaluation
I've seen Tarte Maracuja Juicy Lip recommended on social media. Is it worth the hype, and what do reviewers say compared with similar lip products?
OpenAI
gpt-6-luna
The answer gives a hedged overall verdict on Tarte Maracuja Juicy Lip Balm Gloss (worth trying for glossy tint, not for treatment/wear), reports reviewer sentiment (positive: comfortable shine, popularity, shade range; negative: packaging, value, limited plumping/wear) partly spilling over to the separate Lip Plump formula, and compares the Balm Gloss against Summer Fridays Lip Butter Balm and Clinique Almost Lipstick using Allure's comparison framing, closing with conditional purchase guidance.
Qualifying narrator selections in the saved table: Clinique — Almost Lipstick; Summer Fridays — Lip Butter Balm; Tarte Cosmetics — Maracuja Juicy Lip Balm Gloss
it’s worth a try if you want a cushy, glossy tint—not if you’re expecting a miracle lip treatment or long wear
Excerpt from the saved OpenAI answer. Product claims are unverified.
Meta Muse Spark
muse-spark-1.3-contributor
The answer gives a qualified verdict on Tarte Maracuja Juicy Lip (good for shine/comfort, weak for dramatic plump or long wear), relays brand marketing claims, aggregates reviewer praise and complaints for the original Balm and the Juicy Shift variant, and runs a three-way comparison table/verdict (Tarte vs LANEIGE Lip Glowy Balm vs NYX Butter Gloss) plus looser mentions of Summer Fridays (Lip Butter Balm and Dream Lip Oil), e.l.f. Glow Reviver Lip Oil, Makeup by Mario plumping serum and LANEIGE Lip Sleeping Mask. Several passages are quoted from a third-party comparison site or individual reviewers rather than being the narrator's own verdict, and some reviewer experiences on plumping, brand preference, and e.l.f.'s price conflict with each other.
Qualifying narrator selections in the saved table: LANEIGE — LANEIGE Lip Glowy Balm; NYX — NYX Butter Gloss; Tarte Cosmetics — Maracuja Juicy Lip Balm (original/standard)
Want **high shine that still feels like lip care**, not a burning traditional plumper
Excerpt from the saved Meta Muse Spark answer. Product claims are unverified.
Recorded comparison
Selections and preferences. Meta's narrator gives a conditional verdict on the original Maracuja Juicy Lip Balm: yes for shine and comfort, no for dramatic plump or long wear. It recommends buying on sale . Its closing text names NYX Butter Gloss as the pick under $6 and LANEIGE Lip Glowy Balm for everyday tinted hydration with a larger review base . Other items are quoted comparison or reviewer material: the comparison source's 'best overall' and 'best splurge' verdicts, the Summer Fridays reviewer split, e.l.f. and Makeup by Mario. OpenAI's narrator gives a hedged verdict on the Balm Gloss: worth a try for a cushy, glossy tint, not for miracle treatment or long wear . It suggests trying a mini or waiting for a sale . It names Summer Fridays and Clinique as better for a sheer everyday tint . It says Tarte is the pick for a glossier look and more color payoff (relayed from Allure).
Alternatives and conditions. Meta's alternatives are LANEIGE (everyday tinted hydration), NYX (budget and shade range), Summer Fridays Lip Butter Balm and Dream Lip Oil, e.l.f. and Makeup by Mario. Most of these are tied to comparison-source or reviewer voices. OpenAI's alternatives are Summer Fridays Lip Butter Balm and Clinique Almost Lipstick, both from one Allure comparison, plus an unnamed 'simpler balm' for basic hydration. The two answers share only Summer Fridays. Their alternative sets otherwise differ: LANEIGE, NYX and e.l.f. appear only in Meta, Clinique only in OpenAI. Conditions differ too. Meta stresses sale price, plump expectations and packaging. OpenAI stresses formula separation (Balm Gloss vs Lip Plump), value at full price and the sheer-tint alternative.
Reasons, caveats and attribution. Both answers rest on a similar logic: the product is glossy and comfortable but not dramatically plumping or long-wearing. Meta shows more reviewer conflict. Reviewers disagree on plumping, one reviewer gives Summer Fridays and Tarte parity while another prefers Summer Fridays, and the e.l.f. price appears as both $9 and $8. These are source-answer limitations or attributed experiences, not narrator contradictions. OpenAI adds a derived ~$300/oz value critique and Ulta rating data, and it flags that ratings cannot predict texture or shade fit. It uses a single reviewed.com citation for both the Lip Plump and the balm, which leaves the support for each formula unclear. Meta's price and rating figures come from the answer's Sephora and Amazon sources, in an 'Oct. 2026 check' framing. OpenAI's come from Tarte, Ulta and Allure. The two answers use different merchants and price points. External facts were not verified.
Interpretation for this shopper. The answers largely agree on the core verdict: good shine, comfort and color, but not dramatic plump or long wear, and best bought on sale or tried cautiously. They differ in which alternatives they surface (LANEIGE, NYX and e.l.f. in Meta; Clinique in OpenAI) and in how much they emphasise value, packaging and the Lip Plump distinction. These are differences in emphasis and preference, not factual contradictions. No same-product, same-condition narrator assertions conflict.
Shared and exclusive — brand
| Saved overlap category | Qualifying records |
|---|---|
| Shared | Tarte Cosmetics |
| Meta Muse Spark only | LANEIGE; NYX |
| OpenAI only | Clinique; Summer Fridays |
Shared and exclusive — exact product
| Saved overlap category | Qualifying records |
|---|---|
| Shared | None / not applicable |
| Meta Muse Spark only | laneige | laneige lip glowy balm | variant unstated; nyx | nyx butter gloss | variant unstated; tarte cosmetics | maracuja juicy lip balm (original/standard) | variant unstated |
| OpenAI only | tarte cosmetics | maracuja juicy lip balm gloss | variant unstated |
Q09 Cruelty-free & vegan guidance (Branded · policy explanation)
Branded · policy explanation
I only buy cruelty-free makeup and prefer vegan formulas. Is Tarte entirely vegan, and how can I verify whether a specific concealer or mascara is vegan?
OpenAI
gpt-6-luna
The answer tells the shopper that Tarte is not entirely vegan, distinguishes cruelty-free from vegan, gives three verification steps, and then cites three specific products (Shape Tape Concealer, Shape Tape Creamy Concealer, tartelette Tubing Mascara) that Tarte currently labels vegan and cruelty-free, attributing that status to Tarte's own labeling rather than independent verification.
Qualifying narrator selections in the saved table: No new qualifying product selection; this answer is informational or facilitates an already-selected purchase.
Tarte currently labels **Shape Tape Concealer**, **Shape Tape Creamy Concealer**, and **tartelette Tubing Mascara** vegan and cruelty-free.
Excerpt from the saved OpenAI answer. Product claims are unverified.
Meta Muse Spark
muse-spark-1.3-contributor
The answer tells the shopper that Tarte is cruelty-free but not entirely vegan, explains the cruelty-free/vegan distinction, and gives a three-step method (check the Vegan-Friendly collection, look for a 'Vegan' label on the product page/packaging, and read the ingredient list for beeswax/carmine/lanolin) for verifying whether a specific concealer or mascara is vegan. It uses shape tape concealer and tartelette tubing mascara (including an XL variant) as positive examples found in the vegan collection/labeled vegan, and cites tarte Maneater Waterproof and Lights, Camera, Splashes mascaras as non-vegan due to beeswax. No prices, stock, merchants, or promotions are given.
Qualifying narrator selections in the saved table: No new qualifying product selection; this answer is informational or facilitates an already-selected purchase.
Independent cruelty-free guides list it as **certified cruelty-free by PETA**, noting **it does not test finished products or ingredients on animals**.
Excerpt from the saved Meta Muse Spark answer. Product claims are unverified.
Recorded comparison
Selections and preferences. Neither answer recommends a product to buy. Both give the same core answer: Tarte is cruelty-free but not entirely vegan. Meta relays this through unnamed guides and Tarte's wording, and codes the brand as a conditional recommendation. OpenAI states it in the narrator's own voice (mentioned, null stance). Both name the same concealer and mascara as vegan-labeled examples. Meta names Shape Tape concealer, tartelette tubing mascara and tartelette XL tubing mascara (mentioned). OpenAI names Shape Tape Concealer, Shape Tape Creamy Concealer and tartelette Tubing Mascara (mentioned). Meta also gives two non-vegan examples, Maneater Waterproof and Lights, Camera, Splashes (rejected_option). Those two are attributed only to unnamed 'research cited'.
Alternatives and conditions. The conditions are verification routes, not product preferences. Meta: (1) check Tarte's Vegan-Friendly collection and do not assume vegan if the product is absent; (2) look for the Vegan label or logo on the page and packaging; (3) read the ingredient list by shade for beeswax, carmine, lanolin and similar; (4) apply a rule to treat the product as non-vegan if it is unlabeled, not in the collection, or lists animal-derived ingredients. OpenAI: (1) check the vegan label or the Vegan Friendly collection; (2) check the exact shade and formula, comparing packaging against the product page; (3) if unclear, ask Tarte to confirm, since an ingredient list alone may not show ingredient sources. Meta gives a conservative default and an ingredient-screening list. OpenAI adds a packaging-versus-page check and a brand-contact step. Neither gives prices, merchants or stock.
Reasons, caveats and attribution. Both separate cruelty-free (no animal testing) from vegan (no animal-derived ingredients). Meta adds that formulas can differ by shade and lists animal-derived ingredients. OpenAI cautions that ingredient lists may not reveal ingredient sources. Meta's sourcing is mostly unnamed guides and 'research cited', and its beeswax claims for two mascaras are unresolved. Its XL mascara label quote omits 'cruelty-free', a quoted-wording gap rather than a conflict. OpenAI's examples rest on Tarte's own labeling, and its links use two literal hostnames (tartecosmetica.com and tartecosmetics.com) with unverified affiliation. No external fact verification was done.
Interpretation for this shopper. For the review focus, preference_difference is not the accurate primary classification. Both answers agree on the core claim that Tarte is cruelty-free but not entirely vegan. Neither endorses a product, and the named examples are informational. What differs is verification guidance. Meta stresses the collection, the label and an ingredient-screening rule with a default of non-vegan. OpenAI stresses the label, a packaging-versus-page comparison and contacting Tarte. These methods are complementary and not incompatible. The verification-method differences are real but are differences in guidance and emphasis, not conflicting preferences or factual claims, so agreement on the core claim is the better primary label. Meta's extra non-vegan examples rest on unnamed sourcing and are not contradicted by OpenAI.
Shared and exclusive — brand
| Saved overlap category | Qualifying records |
|---|---|
| Shared | None / not applicable |
| Meta Muse Spark only | None |
| OpenAI only | None |
Shared and exclusive — exact product
| Saved overlap category | Qualifying records |
|---|---|
| Shared | None / not applicable |
| Meta Muse Spark only | None |
| OpenAI only | None |
Q10 Purchase facilitation (Branded · purchase facilitation)
Branded · purchase facilitation
I'm in the United States and ready to buy Tarte Shape Tape concealer and Tartelette tubing mascara. What are the current prices in USD ($), total cost, US shipping threshold and most direct place to purchase? Stop before checkout.
OpenAI
gpt-6-luna
The answer identifies Tarte's official U.S. site as the most direct purchase location for two already-selected products (Shape Tape Concealer and Tartelette Tubing Mascara), states individual prices ($32, $28), a $60 subtotal, a $40 free-shipping threshold that the order meets, a conditional 15%-off first-purchase promotion that could reduce the subtotal to $51, and confirms it stopped before checkout. No stock/availability status is asserted; purchase facilitation (links) is provided for products the shopper already decided to buy, not new endorsement.
Qualifying narrator selections in the saved table: No new qualifying product selection; this answer is informational or facilitates an already-selected purchase.
Shop the [Shape Tape Concealer]
Excerpt from the saved OpenAI answer. Product claims are unverified.
Meta Muse Spark
muse-spark-1.3-contributor
The answer gives current USD list prices for Tarte Shape Tape full-coverage concealer ($32.00) and Tartelette tubing mascara ($28.00), sums them to a $60.00 pre-tax/shipping subtotal, states a $40 US standard-shipping free-shipping threshold and confirms the order qualifies, and directs the shopper to buy both items directly from tartecosmetics.com/tarte.com while explicitly stopping before checkout. No stock/availability statement, discount, or promotion period is asserted; the narrator is facilitating a purchase the shopper has already decided to make rather than making a comparative product recommendation.
Qualifying narrator selections in the saved table: No new qualifying product selection; this answer is informational or facilitates an already-selected purchase.
In the US, buy direct from Tarte — **tartecosmetics.com / tarte.com**:
Excerpt from the saved Meta Muse Spark answer. Product claims are unverified.
Recorded comparison
Selections and preferences. Both answers make the same core recommendation. The shopper has already chosen both products, so each answer facilitates the purchase rather than endorsing a product. Each tells the shopper to buy directly from Tarte's US site. Both give $32 for Shape Tape concealer and $28 for Tartelette tubing mascara, a $60 pre-tax subtotal, and a $40 US free-standard-shipping threshold that the order meets. Both stop before checkout.
Alternatives and conditions. Neither answer names a third-party retailer or an alternative product. The Meta answer adds nothing conditional. It says $60 is before tax and shipping and quotes the $40 threshold including HI, AK and PR. The OpenAI answer adds a date stamp (November 3, 2026) and a conditional 15% first-purchase discount for eligible sign-ups. If that offer applied to both items, the subtotal would be $51 before tax. That figure is explicitly conditional and does not replace the $60 subtotal. The OpenAI answer also gives no promotion period.
Reasons, caveats and attribution. URL review: the literal product URLs are identical across the two answers. Concealer: https://tartecosmetics.com/products/shape-tape-concealer. Mascara: https://tartecosmetics.com/products/tartelette-tubing-mascara. There is no slug difference, so the earlier claim about a slight slug difference was unsupported and is not carried forward. What differs is presentation. Meta lists them as bare URLs under 'Direct product pages'. OpenAI wraps them in nested Markdown, [text](([tartecosmetics.com](url))), which its analysis records as malformed. OpenAI also cites collections/all, faq and promotional-disclaimers pages as support. Meta gives no citations and states the price and shipping figures directly. Meta labels prices explicitly as USD, while OpenAI uses $ and the currency comes from the panel context. Both answers assert 'most direct' without comparing other retailers. Nothing was externally verified, and the OpenAI discount eligibility and date provenance remain unresolved source-answer uncertainties.
Interpretation for this shopper. The answers agree on the destination, prices, subtotal, shipping threshold and eligibility. Differences are in presentation, citation, date stamp and the added conditional promotion. These are not preference differences or factual contradictions. The link difference is Markdown formatting only, not URL identity.
Shared and exclusive — brand
| Saved overlap category | Qualifying records |
|---|---|
| Shared | None / not applicable |
| Meta Muse Spark only | None |
| OpenAI only | None |
Shared and exclusive — exact product
| Saved overlap category | Qualifying records |
|---|---|
| Shared | None / not applicable |
| Meta Muse Spark only | None |
| OpenAI only | None |
What the answers say about benefits and caveats
All twenty answers. Counts represent individual coded statements about benefits or caveats, separated by who makes the claim. They do not measure customer numbers, independent confirmation or recommendation quality. Underlying values are retained in the accompanying tables.
Accessible reason-count source table
| Provider | Panel | Kind | Criterion | Voice | Atomic claims |
|---|---|---|---|---|---|
| Meta Muse Spark | branded | benefit | availability | model | 1 |
| Meta Muse Spark | branded | benefit | coverage | model | 1 |
| Meta Muse Spark | branded | benefit | coverage | unclear | 1 |
| Meta Muse Spark | branded | benefit | ethics | brand | 3 |
| Meta Muse Spark | branded | benefit | ethics | reviewer | 2 |
| Meta Muse Spark | branded | benefit | finish | brand | 4 |
| Meta Muse Spark | branded | benefit | finish | comparison | 4 |
| Meta Muse Spark | branded | benefit | formulation | brand | 7 |
| Meta Muse Spark | branded | benefit | formulation | model | 1 |
| Meta Muse Spark | branded | benefit | formulation | reviewer | 1 |
| Meta Muse Spark | branded | benefit | formulation | unclear | 1 |
| Meta Muse Spark | branded | benefit | hydration | brand | 3 |
| Meta Muse Spark | branded | benefit | hydration | model | 1 |
| Meta Muse Spark | branded | benefit | reviewer_opinion | comparison | 3 |
| Meta Muse Spark | branded | benefit | reviewer_opinion | retailer | 2 |
| Meta Muse Spark | branded | benefit | reviewer_opinion | reviewer | 3 |
| Meta Muse Spark | branded | benefit | reviewer_opinion | unclear | 1 |
| Meta Muse Spark | branded | benefit | suitability | brand | 3 |
| Meta Muse Spark | branded | benefit | suitability | reviewer | 4 |
| Meta Muse Spark | branded | benefit | suitability | unclear | 1 |
| Meta Muse Spark | branded | benefit | value | comparison | 1 |
| Meta Muse Spark | branded | benefit | value | model | 1 |
| Meta Muse Spark | branded | benefit | value | reviewer | 1 |
| Meta Muse Spark | branded | benefit | wear | brand | 3 |
| Meta Muse Spark | branded | benefit | wear | model | 1 |
| Meta Muse Spark | branded | benefit | wear | unclear | 1 |
| Meta Muse Spark | branded | caveat | formulation | model | 5 |
| Meta Muse Spark | branded | caveat | formulation | reviewer | 1 |
| Meta Muse Spark | branded | caveat | formulation | unclear | 2 |
| Meta Muse Spark | branded | caveat | hydration | reviewer | 1 |
| Meta Muse Spark | branded | caveat | packaging | reviewer | 1 |
| Meta Muse Spark | branded | caveat | packaging | unclear | 1 |
| Meta Muse Spark | branded | caveat | shade | reviewer | 2 |
| Meta Muse Spark | branded | caveat | suitability | model | 1 |
| Meta Muse Spark | branded | caveat | suitability | reviewer | 2 |
| Meta Muse Spark | branded | caveat | suitability | unclear | 2 |
| Meta Muse Spark | branded | caveat | value | reviewer | 1 |
| Meta Muse Spark | branded | caveat | wear | brand | 1 |
| Meta Muse Spark | branded | caveat | wear | reviewer | 1 |
| Meta Muse Spark | branded | caveat | wear | unclear | 1 |
| Meta Muse Spark | branded | drawback | coverage | model | 1 |
| Meta Muse Spark | branded | drawback | formulation | reviewer | 1 |
| Meta Muse Spark | branded | drawback | suitability | model | 1 |
| Meta Muse Spark | branded | drawback | wear | reviewer | 1 |
| Meta Muse Spark | branded | reason | availability | model | 2 |
| Meta Muse Spark | branded | reason | formulation | comparison | 1 |
| Meta Muse Spark | branded | reason | hydration | comparison | 1 |
| Meta Muse Spark | branded | reason | other | brand | 1 |
| Meta Muse Spark | branded | reason | other | model | 2 |
| Meta Muse Spark | branded | reason | other | reviewer | 1 |
| Meta Muse Spark | branded | reason | other | unclear | 1 |
| Meta Muse Spark | branded | reason | packaging | reviewer | 1 |
| Meta Muse Spark | branded | reason | reviewer_opinion | comparison | 1 |
| Meta Muse Spark | branded | reason | shade | brand | 3 |
| Meta Muse Spark | branded | reason | shade | model | 3 |
| Meta Muse Spark | branded | reason | suitability | comparison | 3 |
| Meta Muse Spark | branded | reason | suitability | model | 4 |
| Meta Muse Spark | branded | reason | value | comparison | 2 |
| Meta Muse Spark | unbranded | benefit | coverage | brand | 5 |
| Meta Muse Spark | unbranded | benefit | coverage | model | 6 |
| Meta Muse Spark | unbranded | benefit | coverage | unclear | 1 |
| Meta Muse Spark | unbranded | benefit | finish | brand | 6 |
| Meta Muse Spark | unbranded | benefit | finish | model | 1 |
| Meta Muse Spark | unbranded | benefit | formulation | brand | 3 |
| Meta Muse Spark | unbranded | benefit | formulation | model | 2 |
| Meta Muse Spark | unbranded | benefit | formulation | unclear | 1 |
| Meta Muse Spark | unbranded | benefit | hydration | brand | 2 |
| Meta Muse Spark | unbranded | benefit | hydration | model | 3 |
| Meta Muse Spark | unbranded | benefit | other | brand | 1 |
| Meta Muse Spark | unbranded | benefit | removal | unclear | 1 |
| Meta Muse Spark | unbranded | benefit | reviewer_opinion | reviewer | 1 |
| Meta Muse Spark | unbranded | benefit | reviewer_opinion | unclear | 2 |
| Meta Muse Spark | unbranded | benefit | shade | brand | 3 |
| Meta Muse Spark | unbranded | benefit | suitability | brand | 1 |
| Meta Muse Spark | unbranded | benefit | suitability | model | 3 |
| Meta Muse Spark | unbranded | benefit | suitability | unclear | 1 |
| Meta Muse Spark | unbranded | benefit | value | model | 2 |
| Meta Muse Spark | unbranded | benefit | wear | brand | 4 |
| Meta Muse Spark | unbranded | benefit | wear | model | 3 |
| Meta Muse Spark | unbranded | benefit | wear | unclear | 1 |
| Meta Muse Spark | unbranded | caveat | coverage | model | 2 |
| Meta Muse Spark | unbranded | caveat | formulation | model | 1 |
| Meta Muse Spark | unbranded | caveat | other | model | 2 |
| Meta Muse Spark | unbranded | caveat | reviewer_opinion | unclear | 1 |
| Meta Muse Spark | unbranded | caveat | shade | model | 1 |
| Meta Muse Spark | unbranded | caveat | suitability | model | 3 |
| Meta Muse Spark | unbranded | caveat | wear | model | 3 |
| Meta Muse Spark | unbranded | caveat | wear | reviewer | 1 |
| Meta Muse Spark | unbranded | reason | availability | brand | 1 |
| Meta Muse Spark | unbranded | reason | formulation | model | 1 |
| Meta Muse Spark | unbranded | reason | formulation | unclear | 2 |
| Meta Muse Spark | unbranded | reason | other | model | 4 |
| Meta Muse Spark | unbranded | reason | removal | unclear | 1 |
| Meta Muse Spark | unbranded | reason | suitability | model | 1 |
| Meta Muse Spark | unbranded | reason | value | model | 2 |
| Meta Muse Spark | unbranded | reason | wear | model | 4 |
| OpenAI | branded | benefit | coverage | model | 3 |
| OpenAI | branded | benefit | coverage | reviewer | 1 |
| OpenAI | branded | benefit | ethics | brand | 3 |
| OpenAI | branded | benefit | finish | comparison | 2 |
| OpenAI | branded | benefit | finish | model | 3 |
| OpenAI | branded | benefit | formulation | model | 1 |
| OpenAI | branded | benefit | hydration | brand | 1 |
| OpenAI | branded | benefit | hydration | model | 1 |
| OpenAI | branded | benefit | hydration | reviewer | 1 |
| OpenAI | branded | benefit | other | model | 2 |
| OpenAI | branded | benefit | reviewer_opinion | retailer | 1 |
| OpenAI | branded | benefit | shade | model | 1 |
| OpenAI | branded | benefit | suitability | brand | 1 |
| OpenAI | branded | benefit | suitability | comparison | 1 |
| OpenAI | branded | benefit | suitability | model | 1 |
| OpenAI | branded | benefit | value | model | 1 |
| OpenAI | branded | caveat | coverage | model | 2 |
| OpenAI | branded | caveat | formulation | brand | 1 |
| OpenAI | branded | caveat | other | model | 4 |
| OpenAI | branded | caveat | reviewer_opinion | model | 2 |
| OpenAI | branded | caveat | suitability | brand | 1 |
| OpenAI | branded | caveat | suitability | model | 3 |
| OpenAI | branded | caveat | value | model | 1 |
| OpenAI | branded | caveat | value | retailer | 1 |
| OpenAI | branded | caveat | wear | model | 2 |
| OpenAI | branded | caveat | wear | reviewer | 1 |
| OpenAI | branded | drawback | ethics | model | 1 |
| OpenAI | branded | drawback | finish | model | 1 |
| OpenAI | branded | drawback | formulation | reviewer | 1 |
| OpenAI | branded | drawback | packaging | reviewer | 1 |
| OpenAI | branded | drawback | value | model | 1 |
| OpenAI | branded | reason | other | brand | 1 |
| OpenAI | branded | reason | shade | model | 4 |
| OpenAI | branded | reason | suitability | model | 3 |
| OpenAI | branded | reason | value | model | 1 |
| OpenAI | unbranded | benefit | coverage | brand | 2 |
| OpenAI | unbranded | benefit | coverage | model | 5 |
| OpenAI | unbranded | benefit | finish | brand | 1 |
| OpenAI | unbranded | benefit | finish | comparison | 1 |
| OpenAI | unbranded | benefit | finish | model | 3 |
| OpenAI | unbranded | benefit | formulation | model | 3 |
| OpenAI | unbranded | benefit | formulation | retailer | 1 |
| OpenAI | unbranded | benefit | removal | model | 1 |
| OpenAI | unbranded | benefit | removal | reviewer | 1 |
| OpenAI | unbranded | benefit | suitability | model | 1 |
| OpenAI | unbranded | benefit | value | model | 4 |
| OpenAI | unbranded | benefit | wear | brand | 1 |
| OpenAI | unbranded | benefit | wear | model | 1 |
| OpenAI | unbranded | benefit | wear | reviewer | 1 |
| OpenAI | unbranded | caveat | coverage | model | 1 |
| OpenAI | unbranded | caveat | finish | model | 2 |
| OpenAI | unbranded | caveat | other | model | 2 |
| OpenAI | unbranded | caveat | suitability | model | 1 |
| OpenAI | unbranded | caveat | suitability | unclear | 1 |
| OpenAI | unbranded | caveat | value | model | 1 |
| OpenAI | unbranded | caveat | wear | model | 1 |
| OpenAI | unbranded | reason | coverage | brand | 1 |
| OpenAI | unbranded | reason | coverage | model | 1 |
| OpenAI | unbranded | reason | formulation | brand | 1 |
| OpenAI | unbranded | reason | formulation | editorial | 1 |
| OpenAI | unbranded | reason | hydration | model | 1 |
| OpenAI | unbranded | reason | other | model | 3 |
| OpenAI | unbranded | reason | reviewer_opinion | model | 1 |
| OpenAI | unbranded | reason | reviewer_opinion | reviewer | 1 |
| OpenAI | unbranded | reason | suitability | model | 8 |
| OpenAI | unbranded | reason | value | model | 2 |
| OpenAI | unbranded | reason | wear | model | 1 |
Historical usefulness figure and rubric
Saved communication-usefulness assessments
Frozen historical figure, twenty answers. Categorical judgments about communication, not a product-quality or composite usefulness score. The inherited budget-wording limitation in Q04 remains visible. Underlying values are retained in the accompanying tables.
See the 170 saved communication assessments.
| Question | Provider | Dimension | Status | Note |
|---|---|---|---|---|
| Q01 | Meta Muse Spark | Not stated | addressed | Full-coverage framing is stated at the outset and reiterated per product's coverage bullet. |
| Q01 | Meta Muse Spark | Not stated | addressed | Dark circles are addressed explicitly for each of the three products. |
| Q01 | Meta Muse Spark | Not stated | addressed | Addressed for all three, though with a caveat for Tarte limiting the claim to thin application and excluding dry/mature skin overapplication. |
| Q01 | Meta Muse Spark | Not stated | addressed | Three distinct brands (NARS, Tarte, Maybelline) are provided as required. |
| Q01 | Meta Muse Spark | Not stated | addressed | Both online retailers and Manhattan store addresses are given. |
| Q01 | Meta Muse Spark | request_coverage | addressed | The opening sentence directly frames the response around all stated shopper requirements. |
| Q01 | Meta Muse Spark | choice_explanation | addressed | Each product includes explicit coverage and 'why it won't look cakey' explanations supporting the choice. |
| Q01 | Meta Muse Spark | internal_consistency | partial | The Tarte tip caveats dry/mature skin and over-application, whereas the non-cakey line is a general statement; this is a conditional caveat, not a same-condition contradiction. Opening blanket claim is narrowed, hence partial. |
| Q01 | Meta Muse Spark | purchase_guidance | addressed | Purchase guidance includes both online retailers and specific physical store addresses relevant to the shopper's New York location. |
| Q01 | OpenAI | Not stated | addressed | All three products are explicitly described with full or buildable medium-to-full coverage. |
| Q01 | OpenAI | Not stated | partial | Only the Tarte product explicitly ties to dark circles; the other two are general full-coverage concealers without an explicit dark-circle claim. |
| Q01 | OpenAI | Not stated | addressed | Addressed via framing and explicit application technique to reduce cakey appearance. |
| Q01 | OpenAI | Not stated | addressed | Three distinct brands (Tarte, Bobbi Brown, Huda Beauty) are provided as requested. |
| Q01 | OpenAI | Not stated | addressed | Purchase guidance given via Sephora pricing per item and a New York store list, though it stops short of confirming specific in-stock status. |
| Q01 | OpenAI | request_coverage | addressed | All core elements of the shopper's request (coverage, dark circles for at least one item, cakey avoidance, three brands, buying locations) are addressed to varying completeness. |
| Q01 | OpenAI | choice_explanation | addressed | Each product includes a brief rationale (finish type, wear, targeted use) explaining why it was chosen. |
| Q01 | OpenAI | internal_consistency | addressed | No internal contradictions identified; differing finishes (radiant, natural, matte) are presented as preference-based options rather than conflicting claims about the same condition. |
| Q01 | OpenAI | purchase_guidance | partial | Purchase guidance (pricing, retailer, NY store list, pickup option) is given, but it explicitly disclaims confirmed stock/shade availability at any specific store, so guidance is general rather than a firm purchase confirmation. |
| Q02 | Meta Muse Spark | Not stated | addressed | The 45-year-old context is directly invoked as the framing for the whole answer and recurs via 'mature skin' references. |
| Q02 | Meta Muse Spark | Not stated | addressed | Dry under-eyes is addressed via hydrating-ingredient rationale and product-specific caveats. |
| Q02 | Meta Muse Spark | Not stated | addressed | Anti-settling claims and caveats are supplied for all three products. |
| Q02 | Meta Muse Spark | Not stated | addressed | Coverage level is explicitly described for each of the three products. |
| Q02 | Meta Muse Spark | Not stated | addressed | Three named products are compared with a dedicated 'Which should you pick?' summary. |
| Q02 | Meta Muse Spark | request_coverage | addressed | All explicit prompt elements (age context, dryness, anti-settling, coverage, three-way comparison) are covered. |
| Q02 | Meta Muse Spark | choice_explanation | addressed | A dedicated section explains the rationale for choosing each product under different sub-conditions. |
| Q02 | Meta Muse Spark | internal_consistency | partial | The NARS caveat contains a tension between 'looks better over time' and a cited review reporting creasing after 4-6 hours on dry skin, which is not reconciled within the text. |
| Q02 | Meta Muse Spark | purchase_guidance | missing | No price, stock/availability, merchant, promotion, or fulfilment information is present anywhere in the answer. |
| Q02 | OpenAI | Not stated | addressed | The shopper's stated age is explicitly acknowledged and used to frame the recommendation approach. |
| Q02 | OpenAI | Not stated | addressed | Dry under-eyes is directly addressed via formula guidance and the Kosas primary pick. |
| Q02 | OpenAI | Not stated | partial | The 'no settling into lines' requirement is addressed with a caveat and technique tips rather than a guaranteed solution. |
| Q02 | OpenAI | Not stated | addressed | Coverage levels are compared and a product is specifically positioned for coverage priority. |
| Q02 | OpenAI | Not stated | addressed | Three named products are compared in a structured table as requested. |
| Q02 | OpenAI | request_coverage | addressed | All stated shopper requirements (age context, dry under-eyes, anti-creasing, coverage, three-way comparison) receive some level of response. |
| Q02 | OpenAI | choice_explanation | addressed | Each product's table row and the bottom line give a distinct rationale for when it would be chosen. |
| Q02 | OpenAI | internal_consistency | addressed | No outright contradictions found; caveats (e.g., NARS discomfort on very dry skin) are consistent with, not contradictory to, the comparative positioning. |
| Q02 | OpenAI | purchase_guidance | partial | Only list prices and a general price-sourcing note are given; no stock, merchant, or purchase-link information is present in the substantive answer itself (citations are provenance only). |
| Q03 | Meta Muse Spark | Not stated | addressed | Wear-time is explicitly stated and compared for all three products. |
| Q03 | Meta Muse Spark | Not stated | addressed | Combination/oily skin suitability is addressed for all three products. |
| Q03 | Meta Muse Spark | Not stated | addressed | All three listed prices fall under $45 and this is explicitly stated. |
| Q03 | Meta Muse Spark | Not stated | addressed | US market context stated for pricing basis; no separate US-specific availability claim beyond this. |
| Q03 | Meta Muse Spark | Not stated | addressed | A dedicated comparison table and per-product sections cover coverage, finish and value across all three brands. |
| Q03 | Meta Muse Spark | Not stated | addressed | The before-tax-and-shipping basis is explicitly restated. |
| Q03 | Meta Muse Spark | request_coverage | addressed | The answer covers wear, skin type, budget, location, and the three-way comparison requested by the prompt. |
| Q03 | Meta Muse Spark | choice_explanation | addressed | The verdict paragraph explicitly explains why each of the three products was chosen for a distinct priority (wear, matte+SPF, value). |
| Q03 | Meta Muse Spark | internal_consistency | partial | One sentence introduces an unexplained '24H' L'Oréal reference alongside the otherwise consistent '32 Hour' naming used throughout, creating an internal naming inconsistency that is not resolved elsewhere in the answer. |
| Q03 | Meta Muse Spark | purchase_guidance | addressed | Per-ounce value calculations and explicit price figures are given for all three products, supporting purchase comparison, though no stock/merchant/fulfilment guidance is provided. |
| Q03 | OpenAI | Not stated | addressed | Long-wear requirement addressed via brand wear-time claims for the primary pick, with a general caveat about 'up-to' claims. |
| Q03 | OpenAI | Not stated | addressed | Combination/oily skin suitability is explicitly addressed for the primary pick and referenced for the Revlon alternative. |
| Q03 | OpenAI | Not stated | addressed | All three listed prices fall well under $45, satisfying the budget ceiling, though the $45 figure itself is not restated. |
| Q03 | OpenAI | Not stated | partial | US-market availability is implied by USD pricing and US brand-site citations but never explicitly asserted in the answer text. |
| Q03 | OpenAI | Not stated | addressed | A direct three-brand comparison table covers coverage, finish and value as requested. |
| Q03 | OpenAI | Not stated | addressed | Answer explicitly states the pre-tax/pre-shipping basis matching the shopper's stated condition. |
| Q03 | OpenAI | request_coverage | partial | Most explicit requirements (wear, skin type, budget, three-brand comparison, tax/shipping basis) are addressed; US-market applicability is only implicitly supported. |
| Q03 | OpenAI | choice_explanation | addressed | The closing quick guide explains the rationale for each brand choice relative to specific conditions. |
| Q03 | OpenAI | internal_consistency | addressed | No incompatible factual assertions were identified; comparative statements between products are consistent with the stated finish descriptions. |
| Q03 | OpenAI | purchase_guidance | missing | The answer provides only citation URLs to brand product pages, not stock status, merchant, or purchase facilitation details. |
| Q04 | Meta Muse Spark | Not stated | addressed | The answer explicitly identifies the product as using a tubing formula, directly addressing this requirement. |
| Q04 | Meta Muse Spark | Not stated | addressed | Price of $26.00 explicitly stated as under the $35 budget. |
| Q04 | Meta Muse Spark | Not stated | partial | The US availability section gives a corporate address and one variant's 'Available' label. It does not state US-wide stock, a store or a purchase link, so coverage is partial. |
| Q04 | Meta Muse Spark | Not stated | addressed | Answer directly cites product-page language claiming resistance to smudging, flaking, and clumping all day, addressing the requirement (claim itself unverified). |
| Q04 | Meta Muse Spark | Not stated | addressed | Answer cites quoted claims and instructions indicating gentle warm-water removal without harsh rubbing, addressing this requirement. |
| Q04 | Meta Muse Spark | Not stated | addressed | The answer explicitly frames the price comparison as being before tax and shipping, matching the shopper's clarified budget condition. |
| Q04 | Meta Muse Spark | request_coverage | partial | Most explicit requirements (tubing, price, smudge, removal) are addressed via quoted claims, but the US-availability requirement is only partially supported, so overall request coverage is partial. |
| Q04 | Meta Muse Spark | choice_explanation | addressed | The answer gives explicit bullet-point reasons (price, formula type, wear claims, removal ease) for the recommendation, addressing the choice-explanation dimension. |
| Q04 | Meta Muse Spark | internal_consistency | addressed | No internal contradictions were identified across the price, formula, wear, and removal claims presented. |
| Q04 | Meta Muse Spark | purchase_guidance | partial | The answer names the product and cites a brand address and one variant's status, but provides no direct purchase link, checkout destination, or clear stock confirmation, so purchase guidance is only partially complete. |
| Q04 | OpenAI | Not stated | addressed | The recommended product is explicitly identified as a tubing mascara. |
| Q04 | OpenAI | Not stated | addressed | Price explicitly stated as $28, under the $35 budget, though basis relative to tax/shipping is not explicitly confirmed. |
| Q04 | OpenAI | Not stated | partial | US availability is supported only by a single US retailer (Ulta) citation with in-stock status; no broader statement about US-wide availability is made. |
| Q04 | OpenAI | Not stated | addressed | Smudge-resistance need is addressed via a reviewer-sourced (Allure) claim relayed by the narrator; not independently verified. |
| Q04 | OpenAI | Not stated | addressed | Removal-without-rubbing need is addressed via explicit narrator instructions and a supporting reviewer note. |
| Q04 | OpenAI | Not stated | missing | The answer never explicitly states whether the $28 figure excludes sales tax and shipping, despite the shopper's explicit stipulation. |
| Q04 | OpenAI | request_coverage | partial | Most explicit requirements are addressed, but the pre-tax/pre-shipping budget condition and broad US availability are only partially substantiated. |
| Q04 | OpenAI | choice_explanation | addressed | Narrator explains the choice by linking it to the shopper's stated smudge-resistance and easy-removal priorities, citing an external reviewer source. |
| Q04 | OpenAI | internal_consistency | addressed | No internal contradictions identified; the narrator's caveat about variable results is consistent with, not contradictory to, the overall recommendation. |
| Q04 | OpenAI | purchase_guidance | addressed | Clear, direct purchase instruction naming the product, price, retailer, and stock status. |
| Q05 | Meta Muse Spark | Not stated | addressed | Juicy/glossy language appears both in the intro and in each product's rationale section. |
| Q05 | Meta Muse Spark | Not stated | addressed | Intro explicitly frames all picks as 'slightly' plumping; per-product caveats reinforce a gentle (non-extreme) plumping intensity. |
| Q05 | Meta Muse Spark | Not stated | addressed | All three listed prices are under $35 and this is explicitly confirmed again in the closing line. |
| Q05 | Meta Muse Spark | Not stated | addressed | US availability is both generally stated and backed by specific US retailer names (Sephora, elfcosmetics.com, Target). |
| Q05 | Meta Muse Spark | Not stated | addressed | Three distinct brands/products are listed with headings 1–3. |
| Q05 | Meta Muse Spark | Not stated | addressed | The before-tax/shipping condition is restated both in the intro and the closing summary line. |
| Q05 | Meta Muse Spark | request_coverage | addressed | The answer structure maps directly onto each stated requirement (juicy/glossy, slightly plumping, budget, US availability, three products). |
| Q05 | Meta Muse Spark | choice_explanation | addressed | Each product includes an explicit 'Why it fits your brief' rationale plus a distinguishing 'Best if' framing; this is the first of three identical occurrences, one per product. |
| Q05 | Meta Muse Spark | internal_consistency | partial | The e.l.f. product is described both as a 'moisturizing clear' gloss and as having a 'sheer cherry color' without clarifying whether these refer to the same or different variants, creating an unresolved internal tension; this undermines full internal consistency. |
| Q05 | Meta Muse Spark | purchase_guidance | addressed | Purchase guidance includes prices, stock status, merchant names, and shipping messaging per product, supporting actionable purchase decisions. |
| Q05 | OpenAI | Not stated | addressed | Each product row uses glossy/juicy descriptive language. |
| Q05 | OpenAI | Not stated | addressed | Plumping-related sensations are described for all three, and a best match is explicitly named. |
| Q05 | OpenAI | Not stated | addressed | All three listed prices fall under $35, and this is explicitly stated. |
| Q05 | OpenAI | Not stated | missing | No explicit statement in the substantive answer confirms U.S. availability; citation URLs to brand sites are provenance only, not an availability assertion. |
| Q05 | OpenAI | Not stated | addressed | Exactly three brand/product pairs are given. |
| Q05 | OpenAI | Not stated | addressed | The pre-tax, pre-shipping budget constraint is explicitly echoed. |
| Q05 | OpenAI | request_coverage | partial | Most explicit prompt requirements (juicy/glossy, plumping, budget, three products, tax/shipping framing) are covered, but explicit U.S.-availability confirmation is absent. |
| Q05 | OpenAI | choice_explanation | addressed | Each product has a dedicated rationale column explaining why it's recommended. |
| Q05 | OpenAI | internal_consistency | addressed | No contradictory claims about the same product/condition were found; the plumping guidance is consistent with the individual product descriptions. |
| Q05 | OpenAI | purchase_guidance | missing | No stock status, merchant, purchase link, or fulfilment terms are given; only citation URLs, which are provenance only. |
| Q06 | Meta Muse Spark | Not stated | addressed | The answer gives a conditional yes with explicit scenarios rather than leaving the question unanswered. |
| Q06 | Meta Muse Spark | Not stated | addressed | Both alternatives are explicitly compared and each given its own conditional recommendation. |
| Q06 | Meta Muse Spark | Not stated | addressed | Dry-under-eye trade-offs are addressed per product with dedicated sections. |
| Q06 | Meta Muse Spark | request_coverage | addressed | All three explicit asks (Shape Tape worth-it, NARS vs e.l.f., dry trade-offs) are covered in structured sections and a closing quick-verdict summary. |
| Q06 | Meta Muse Spark | choice_explanation | addressed | Each product's recommendation is tied to explicit conditions explaining why a shopper would pick it. |
| Q06 | Meta Muse Spark | internal_consistency | partial | The e.l.f. section juxtaposes a 'Best Overall for Dry Skin' claim with later reviewer statements that it is 'not the most moisturizing' and can crease/oxidize; the answer labels these as mixed customer experience rather than resolving the tension, which is reasonable but leaves some internal tension unexplained. |
| Q06 | Meta Muse Spark | purchase_guidance | partial | Approximate prices are given for all three products, but no stock status, specific retailer, promotion, or fulfilment/delivery information is provided, so purchase guidance is incomplete. |
| Q06 | OpenAI | Not stated | addressed | The 'worth buying' question is answered with explicit conditions rather than a flat yes/no. |
| Q06 | OpenAI | Not stated | addressed | Both named alternatives are directly compared and recommended conditionally. |
| Q06 | OpenAI | Not stated | addressed | A dedicated comparison table and surrounding text address dry under-eye trade-offs for all three concealers. |
| Q06 | OpenAI | request_coverage | addressed | All three explicit prompt elements (Tarte worth-buying, NARS-vs-elf choice, dry under-eye trade-offs) are covered. |
| Q06 | OpenAI | choice_explanation | addressed | Choices are explained by tying them to specific shopper priorities (finish vs. price/coverage vs. wear). |
| Q06 | OpenAI | internal_consistency | addressed | No internal factual contradictions were found; the one notable tension is the narrator flagging an unverified brand claim, coded separately as an evidence limitation rather than an inconsistency. |
| Q06 | OpenAI | purchase_guidance | partial | Price points are given for two products, but no stock status, merchant/retailer, fulfilment, or promotion details are provided; citation links are provenance only, not purchase links. |
| Q07 | Meta Muse Spark | Not stated | addressed | The oily/blemish-prone skin requirement is addressed with a named product and specific reasons (full coverage, matte, long-wear). |
| Q07 | Meta Muse Spark | Not stated | addressed | The dry under-eyes requirement is addressed with a named primary product plus an alternative (Radiant) and relevant caveats. |
| Q07 | Meta Muse Spark | Not stated | addressed | Shade selection is addressed via undertone-letter system, depth rules, and brand tools (quiz, virtual try-on, photo gallery, natural-light swatch test). |
| Q07 | Meta Muse Spark | request_coverage | addressed | All three explicit prompt asks (oily/blemish-prone pick, dry under-eye pick, shade-choosing method) are each given a dedicated section. |
| Q07 | Meta Muse Spark | choice_explanation | addressed | Reasons are supplied for each recommendation (coverage/matte/wear for Original; hydration/ingredients for Ultra Creamy; lightness/finish for Radiant), mostly as quoted brand/retailer/reviewer claims rather than independently verified facts. |
| Q07 | Meta Muse Spark | internal_consistency | partial | Overall consistent, but the brand/retailer claim ('best for oily or combination skin') and the reviewer paraphrase ('works best for normal to oily skin types') describe Original's ideal skin type slightly differently, and legacy claim anchors for p2 overlapped identically across two IDs, requiring manual disambiguation. |
| Q07 | Meta Muse Spark | purchase_guidance | missing | No price, stock status, merchant, promotion, or delivery/pickup information appears anywhere in the answer; only non-purchase brand tools (shade quiz, virtual try-on, photo gallery) are referenced. |
| Q07 | OpenAI | Not stated | addressed | The oily/blemish-prone skin requirement is directly addressed with a specific product recommendation. |
| Q07 | OpenAI | Not stated | addressed | The dry under-eye requirement is directly addressed with a specific product recommendation and an alternative. |
| Q07 | OpenAI | Not stated | addressed | A structured shade-selection process is provided covering undertone coding, blemish vs. under-eye matching, and tool usage. |
| Q07 | OpenAI | request_coverage | addressed | All three explicit elements of the shopper's request (oily/blemish skin pick, dry under-eye pick, shade guidance) are covered. |
| Q07 | OpenAI | choice_explanation | addressed | Reasons are given for each pick (matte finish/oil control framing for Original; moisturizing/hydration framing for Creamy), though sourced to brand-page citations rather than independently substantiated. |
| Q07 | OpenAI | internal_consistency | addressed | No internal contradictions observed; the mixed-area suggestion is consistent with the two primary recommendations rather than conflicting with them. |
| Q07 | OpenAI | purchase_guidance | missing | No price, stock/availability, merchant, promotion, or fulfilment information is provided anywhere in the answer; the four citations are product-page URLs only, which per policy do not constitute purchase links or availability assertions. |
| Q08 | Meta Muse Spark | Not stated | addressed | Narrator gives an explicit conditional 'worth the hype' verdict and a full section elaborating it. |
| Q08 | Meta Muse Spark | Not stated | addressed | Dedicated 'love' and 'criticize' sections with multiple quoted reviewer statements address this requirement. |
| Q08 | Meta Muse Spark | Not stated | addressed | Comparison table and verdict plus additional looser mentions (Summer Fridays, e.l.f., Makeup by Mario) address this requirement. |
| Q08 | Meta Muse Spark | request_coverage | addressed | All three explicit shopper needs are covered with dedicated sections. |
| Q08 | Meta Muse Spark | choice_explanation | partial | The narrator explains criteria generally ('choose it if...') but defers a tailored choice since the shopper's specific priority wasn't stated, leaving the explanation partially conditional rather than a single resolved recommendation. |
| Q08 | Meta Muse Spark | internal_consistency | partial | Internal consistency is complicated by conflicting reviewer accounts on plumping and brand preference (Summer Fridays vs Tarte), and a price discrepancy for e.l.f. ($9 vs $8); these are presented as differing attributed experiences rather than resolved, so coverage is partial rather than fully consistent. |
| Q08 | Meta Muse Spark | purchase_guidance | partial | Price and a discount condition are mentioned, but no stock status, merchant-specific purchase link, or fulfilment detail is given; purchase guidance is limited to price/sale framing. |
| Q08 | OpenAI | Not stated | addressed | The 'worth the hype' requirement is answered with a conditional verdict rather than a flat yes/no. |
| Q08 | OpenAI | Not stated | addressed | Reviewer sentiment is reported in dedicated sections, though it mixes Balm Gloss and Lip Plump sourcing in places. |
| Q08 | OpenAI | Not stated | addressed | Comparison to similar products is addressed via Summer Fridays and Clinique, both sourced from a single Allure gallery citation. |
| Q08 | OpenAI | request_coverage | addressed | All three explicit parts of the shopper's request (hype, reviewers, comparison) are covered. |
| Q08 | OpenAI | choice_explanation | addressed | Narrator explains the basis for its verdict (finish/color vs. lack of transformative/long-wear quality) and ties conditions to recommendations. |
| Q08 | OpenAI | internal_consistency | partial | The same citation is used to support claims about two distinct formulas (Balm Gloss and Lip Plump) within one sentence, creating some ambiguity about which evidence supports which product, though no outright contradiction is present. |
| Q08 | OpenAI | purchase_guidance | addressed | Provides price, a value caveat, and conditional purchase guidance (mini/sale, or alternatives) relevant to a buying decision. |
| Q09 | Meta Muse Spark | Not stated | addressed | Cruelty-free requirement is confirmed for the brand as a whole, with sourcing to unnamed third-party guides. |
| Q09 | Meta Muse Spark | Not stated | addressed | The entirely-vegan question is answered directly and unambiguously at the top of the response. |
| Q09 | Meta Muse Spark | Not stated | addressed | A concrete three-step verification process is provided with worked examples for specific concealer and mascara products. |
| Q09 | Meta Muse Spark | request_coverage | addressed | The answer covers both the brand-wide vegan/cruelty-free question and the specific-product verification request. |
| Q09 | Meta Muse Spark | choice_explanation | partial | The answer explains general reasoning behind the verification steps (why checking the collection, label, and ingredients matters) but does not recommend one specific concealer or mascara choice for the shopper to buy; it is explanatory rather than a single-choice justification. |
| Q09 | Meta Muse Spark | internal_consistency | partial | Mostly internally consistent, but the Tartelette XL label quote omits 'cruelty-free' wording present in the other two product labels, a minor unexplained inconsistency given the brand-wide cruelty-free claim. |
| Q09 | Meta Muse Spark | purchase_guidance | partial | Only a general pointer to the website's vegan-friendly section is given; no prices, stock status, merchant specifics, or direct purchase links are provided. |
| Q09 | OpenAI | Not stated | addressed | The cruelty-free requirement is addressed at brand level and for the three named example products, all sourced to Tarte's own statements, not independently verified. |
| Q09 | OpenAI | Not stated | addressed | Directly answers whether Tarte is entirely vegan. |
| Q09 | OpenAI | Not stated | addressed | Gives a stepwise verification method and concrete examples for a concealer and a mascara. |
| Q09 | OpenAI | request_coverage | addressed | Both parts of the shopper's compound question (overall vegan status, verification method) are covered. |
| Q09 | OpenAI | choice_explanation | partial | Explanation for why these examples are vegan/cruelty-free rests solely on citing Tarte's own labeling, without further detail on what makes them qualify; thin but present. |
| Q09 | OpenAI | internal_consistency | addressed | No internal contradictions observed; the cruelty-free/vegan distinction is maintained consistently throughout. |
| Q09 | OpenAI | purchase_guidance | missing | No price, stock, merchant, or purchase-link information is given; the answer provides verification guidance only, not purchase facilitation. |
| Q10 | Meta Muse Spark | Not stated | addressed | Current USD prices for both named products are explicitly provided. |
| Q10 | Meta Muse Spark | Not stated | addressed | Total cost is computed and clearly scoped as excluding tax/shipping. |
| Q10 | Meta Muse Spark | Not stated | addressed | US shipping threshold is stated as a quoted figure. |
| Q10 | Meta Muse Spark | Not stated | addressed | Direct brand site is named, though no explicit comparison to other retailers is shown to justify 'most direct' beyond naming the brand's own store. |
| Q10 | Meta Muse Spark | Not stated | addressed | Explicit instruction to halt before payment/checkout is given. |
| Q10 | Meta Muse Spark | request_coverage | addressed | All five explicit shopper requirements (prices, total, threshold, direct place, stop-before-checkout) are addressed in the answer structure. This judgment corresponds to the request_coverage dimension. |
| Q10 | Meta Muse Spark | choice_explanation | partial | This judgment corresponds to the choice_explanation dimension. The answer names the brand site as 'most direct' but does not explain why it is more direct than other retailers (e.g., Sephora, Ulta) that may also sell these products. |
| Q10 | Meta Muse Spark | internal_consistency | addressed | This judgment corresponds to the internal_consistency dimension. The arithmetic ($32+$28=$60) and the threshold comparison ($60>$40) are internally consistent. |
| Q10 | Meta Muse Spark | purchase_guidance | addressed | This judgment corresponds to the purchase_guidance dimension. Clear, explicit guidance to add to cart and stop before completing purchase is given. |
| Q10 | OpenAI | Not stated | addressed | Both current prices are provided in $ consistent with the US/USD context. |
| Q10 | OpenAI | Not stated | addressed | Total cost given as $60 before tax, with a conditional alternative if an unconfirmed promo applies. |
| Q10 | OpenAI | Not stated | addressed | US shipping threshold explicitly stated and applied to the order. |
| Q10 | OpenAI | Not stated | addressed | Direct purchase channel identified, though without comparative reasoning against other retailers. |
| Q10 | OpenAI | Not stated | addressed | Explicit confirmation of stopping before checkout. |
| Q10 | OpenAI | request_coverage | addressed | All five explicit requirements (prices, total, threshold, place, stop) appear to be covered in the answer. |
| Q10 | OpenAI | choice_explanation | partial | The 'most direct' claim is asserted but not explained via comparison to other retailers, so the reasoning behind the choice is thin. |
| Q10 | OpenAI | internal_consistency | addressed | Arithmetic is internally consistent ($32+$28=$60; $60 minus 15% ≈ $51), and the conditional alternate total is clearly flagged as conditional rather than contradicting the base figure. |
| Q10 | OpenAI | purchase_guidance | addressed | Provides product-page links as purchase facilitation while explicitly stopping short of checkout, aligning with the shopper's instruction. |
03 / Citation and source visibility
Which websites receive citations when OpenAI and Meta Muse Spark answer shopping questions?
Lead measure: domain reach. A domain cited in one answer counts once for that answer, even when it has many annotations. The denominator is five citation-eligible answers in each provider/panel. All twenty answers are citation eligible; missing annotations contribute zero observed reach, not proof of no search.
| Provider | Panel | tartecosmetics.com reach |
|---|---|---|
| OpenAI | unbranded | 1/5 eligible answers |
| Meta Muse Spark | unbranded | 0/5 eligible answers |
| OpenAI | branded | 5/5 eligible answers |
| Meta Muse Spark | branded | 3/5 eligible answers |
In the branded panel, tartecosmetics.com reaches every OpenAI answer and three Meta answers. In unbranded answers it appears in one OpenAI answer and none of the retained Meta annotations. A brand can be recommended without its own domain being cited; neither presence nor absence establishes a causal connection to the recommendation.
Domain reach across eligible answers
The twelve domains cited across the most answers, split by provider and question group. Each group contains five answers. The complete thirty-row table below includes ties and domains omitted from the chart. Underlying values are retained in the accompanying tables.
See unbranded domain reach and branded domain reach for all 30 saved rows.
Unbranded sources
| Provider | Domain | Eligible-answer reach | Occurrences | Distinct URLs | Saved classification | Answers |
|---|---|---|---|---|---|---|
| Meta Muse Spark | lorealparisusa.com | 1/5 | 19 | 1 | other-brand-owned | Q03 |
| Meta Muse Spark | maybelline.com | 1/5 | 19 | 1 | other-brand-owned | Q03 |
| Meta Muse Spark | revlon.com | 1/5 | 15 | 1 | other-brand-owned | Q03 |
| Meta Muse Spark | thrivecausemetics.com | 1/5 | 8 | 1 | other-brand-owned | Q04 |
| Meta Muse Spark | elfcosmetics.com | 1/5 | 7 | 1 | other-brand-owned | Q05 |
| Meta Muse Spark | sephora.com | 1/5 | 7 | 1 | retailer | Q05 |
| Meta Muse Spark | target.com | 1/5 | 7 | 1 | retailer | Q05 |
| OpenAI | sephora.com | 2/5 | 6 | 6 | retailer | Q01 · Q02 |
| OpenAI | maybelline.com | 1/5 | 2 | 1 | other-brand-owned | Q03 |
| OpenAI | allure.com | 1/5 | 1 | 1 | editorial/media | Q04 |
| OpenAI | fentybeauty.com | 1/5 | 1 | 1 | other-brand-owned | Q05 |
| OpenAI | kosas.com | 1/5 | 1 | 1 | other-brand-owned | Q02 |
| OpenAI | lorealparisusa.com | 1/5 | 1 | 1 | other-brand-owned | Q03 |
| OpenAI | milanicosmetics.com | 1/5 | 1 | 1 | other-brand-owned | Q05 |
| OpenAI | revlon.com | 1/5 | 1 | 1 | other-brand-owned | Q03 |
| OpenAI | tartecosmetics.com | 1/5 | 1 | 1 | brand-owned | Q04 |
| OpenAI | toofaced.com | 1/5 | 1 | 1 | other-brand-owned | Q05 |
| OpenAI | ulta.com | 1/5 | 1 | 1 | retailer | Q04 |
Branded sources
| Provider | Domain | Eligible-answer reach | Occurrences | Distinct URLs | Saved classification | Answers |
|---|---|---|---|---|---|---|
| Meta Muse Spark | tartecosmetics.com | 3/5 | 21 | 5 | brand-owned | Q07 · Q09 · Q10 |
| Meta Muse Spark | marcyvanity.com | 1/5 | 25 | 1 | other/unknown | Q08 |
| Meta Muse Spark | lemon8-app.com | 1/5 | 13 | 2 | creator/social | Q08 |
| Meta Muse Spark | sephora.com | 1/5 | 10 | 1 | retailer | Q08 |
| Meta Muse Spark | havenlycosmetics.com | 1/5 | 4 | 1 | other/unknown | Q07 |
| OpenAI | tartecosmetics.com | 5/5 | 17 | 9 | brand-owned | Q06 · Q07 · Q08 · Q09 · Q10 |
| OpenAI | allure.com | 1/5 | 2 | 1 | editorial/media | Q08 |
| OpenAI | reviewed.com | 1/5 | 2 | 1 | editorial/media | Q08 |
| OpenAI | ulta.com | 1/5 | 2 | 1 | retailer | Q08 |
| OpenAI | elfcosmetics.com | 1/5 | 1 | 1 | other-brand-owned | Q06 |
| OpenAI | narscosmetics.com | 1/5 | 1 | 1 | other-brand-owned | Q06 |
| OpenAI | tartecosmetica.com | 1/5 | 1 | 1 | other/unknown | Q09 |
Retailers, manufacturers, publishers and unknown sources
Retailer domains include Sephora, Ulta and Target. Other-brand-owned sources include Kosas, Maybelline, L’Oréal Paris, Revlon, Thrive, e.l.f., Milani, Fenty, Too Faced and NARS. Saved editorial/media classifications include Allure and Reviewed; Lemon8 is creator/social. havenlycosmetics.com, marcyvanity.com and the literal tartecosmetica.com hostname remain other/unknown. The similar hostname is not silently treated as Tarte-owned. These classifications come from the saved study, not a new website audit.
Source-type composition of retained citation occurrences
Questions naming Tarte and questions without the brand are shown separately. Counts include repeated citations. Sources that could not be classified remain other/unknown; source type does not establish credibility. Underlying values are retained in the accompanying tables.
See the 15 saved source-type rows.
Accessible source-type table
| Provider | Panel | Source type | Occurrences | Distinct URLs | Eligible-answer reach |
|---|---|---|---|---|---|
| Meta Muse Spark | branded | brand-owned | 21 | 5 | 3/5 |
| Meta Muse Spark | branded | creator/social | 13 | 2 | 1/5 |
| Meta Muse Spark | branded | other/unknown | 29 | 2 | 2/5 |
| Meta Muse Spark | branded | retailer | 10 | 1 | 1/5 |
| Meta Muse Spark | unbranded | other-brand-owned | 68 | 5 | 3/5 |
| Meta Muse Spark | unbranded | retailer | 14 | 2 | 1/5 |
| OpenAI | branded | brand-owned | 17 | 9 | 5/5 |
| OpenAI | branded | editorial/media | 4 | 2 | 1/5 |
| OpenAI | branded | other-brand-owned | 2 | 2 | 1/5 |
| OpenAI | branded | other/unknown | 1 | 1 | 1/5 |
| OpenAI | branded | retailer | 2 | 1 | 1/5 |
| OpenAI | unbranded | brand-owned | 1 | 1 | 1/5 |
| OpenAI | unbranded | editorial/media | 1 | 1 | 1/5 |
| OpenAI | unbranded | other-brand-owned | 8 | 7 | 3/5 |
| OpenAI | unbranded | retailer | 7 | 7 | 3/5 |
Annotation volume, URL breadth and concentration
Meta’s unbranded panel has 82 occurrences across seven URLs, compared with OpenAI’s 17 across sixteen URLs. In branded answers, the corresponding counts are 73 across ten URLs and 26 across fifteen. Repeated annotations can therefore raise volume without broadening the source set. Provider annotation conventions are not standardised, so these volumes do not rank evidence quality or model preference.
Citation occurrences versus distinct cited URLs
Four groups of five answers, split by provider and whether the question names Tarte. Repeated citations and the number of different URLs measure different things. Neither establishes which sources influenced the answer. Underlying values are retained in the accompanying tables.
See the citation and URL tally and the original retrieval summary.
| Provider | Panel | Largest URL occurrence count / panel occurrences | URLs tied at that maximum | Distinct cited URLs |
|---|---|---|---|---|
| OpenAI | unbranded | 2/17 | 1 | 16 |
| OpenAI | branded | 5/26 | 1 | 15 |
| Meta Muse Spark | unbranded | 19/82 | 2 | 7 |
| Meta Muse Spark | branded | 25/73 | 1 | 10 |
Retrieval is a separate observable stream
| Provider | Panel | Saved answers | Citation eligible | No annotations | Retrieval eligible | Citations | Distinct cited URLs | Retrieved records | Distinct retrieved URLs | Citation occurrences matching retrieved URL |
|---|---|---|---|---|---|---|---|---|---|---|
| Meta Muse Spark | branded | 5 | 5 | 1 | 5 | 73 | 10 | 298 | 269 | 71 |
| Meta Muse Spark | unbranded | 5 | 5 | 2 | 4 | 82 | 7 | 331 | 314 | 82 |
| OpenAI | branded | 5 | 5 | 0 | 5 | 26 | 15 | 286 | 223 | 24 |
| OpenAI | unbranded | 5 | 5 | 0 | 5 | 17 | 16 | 311 | 252 | 17 |
Capture limits. Meta Q01, Q02 and Q06 have no saved citation annotations despite retained retrieval records. Meta Q04 has two failed search calls and is excluded from retrieval-eligible comparisons, while all 48 saved source records and eight citations remain visible descriptively. Retrieval eligibility is 19/20 overall: OpenAI 10/10 and Meta 9/10. Complete model retrieval is not observable. Inline/body citation style can differ from provider annotations; Q10’s four Meta annotations are retained even where historical prose says “no citations”.
Discussion: what this evidence can inform
The saved evidence supports questions about which shopper conditions and product identities appear together, rather than an overall brand or model score. Discovery frequency alone hides different product choices; source volume alone hides repetition.
Investigations suggested by the existing report
These examples suggest what a brand could investigate next. They are not proven fixes or forecasts.
Dry under-eyes: recognition differs from discovery
Observation. Neither saved unbranded answer for a woman aged 45+ with dry under-eyes selects Tarte among its three products. Both later recommend Shape Tape Creamy for dry under-eyes when directly asked about Shape Tape variants. This is an observed difference between prompted product knowledge and unsolicited selection.
Possible business relevance. Recognition of a variant does not guarantee discovery for a relevant shopper need. Kosas and NARS appear in both answers; Dior appears in OpenAI and Maybelline in Meta.
Asset to examine. Shape Tape Creamy product information and the Original / Creamy / Radiant comparison, subject to a separate suitability review.
Investigation / conditional hypothesis, as supplied. First establish suitability for mature, dry under-eyes independently. Then examine whether existing product and retailer descriptions explain the relevant conditions clearly. Test a factual clarification only if that review finds a gap.
Proposed future retest. A proposed factual clarification may improve accurate conditional identification. In a separately authorized future collection, repeat the same US/USD prompts and prespecify the share of answers selecting the exact Creamy variant, plus the accuracy of reasons and caveats. Track repeated responses; this panel is not a stable baseline rate.
Omission does not establish that Tarte deserved inclusion. The branded and unbranded prompts differ, so this is not a causal experiment. Product suitability and content effects are unverified.
Q02 · Q02 · Q02 · Q02 · Q02 · Q02 · Q07 · Q07Lip Plump: different competitive sets for a similar need
Observation. For glossy, slightly plumping lips, Meta recommends Maracuja Juicy Lip Plump for an all-in-one color, treatment and plump proposition. OpenAI discusses Milani for a gentler budget option, Fenty for a classic gloss feel and Too Faced conditionally when plumping sensation matters. The answers communicate different trade-offs.
Possible business relevance. The observable comparison is about gloss feel, sensation, hydration and value, rather than a universal winner. These are useful dimensions for understanding the competing propositions presented to shoppers.
Asset to examine. Maracuja Juicy Lip Plump variant information, distinct from Juicy Lip Balm and Balm Gloss.
Investigation / conditional hypothesis, as supplied. Audit the factual support for the communicated combined-benefit proposition and the strength of the plumping sensation. If warranted, clarify the variant comparison without claiming a content issue caused the omission.
Proposed future retest. In a future prespecified repeat panel, measure exact Lip Plump selection, Balm/Plump confusion and whether slight versus intense sensation is described accurately. Compare the same shopper need before and after any separately approved clarification.
One answer per provider does not establish persistent exclusion. Price alone does not explain the different sets. Benefits and sensation are saved model claims, not independently established facts.
Q05 · Q05 · Q05 · Q05 · Q05 · Q05Tubing mascara: product choice and buying evidence diverge
Observation. OpenAI recommends Tartelette XL and states $28 at Ulta, in stock, with warm-water removal. Meta recommends Thrive Liquid Lash Extensions at $26. Its US availability section combines a corporate address with one shade marked Available, which does not establish retail stock or a purchase destination.
Possible business relevance. The saved answers differ in both selection and the usefulness of purchase guidance. An address, shade, price and stock claim answer different shopper questions.
Asset to examine. Tartelette XL versus Original product identity, removal instructions, and retailer purchase details; named Ulta evidence is observable but unverified.
Investigation / conditional hypothesis, as supplied. Independently verify the XL identity, removal guidance and cited purchase information before deciding whether any brand-controlled asset needs correction. Treat the corporate-address issue as an observation about Meta’s answer.
Proposed future retest. In a separately collected repeat panel, measure exact XL identification and the proportion of purchase answers with a product destination, explicit merchant, shade-specific stock scope and independently checked availability. Keep those measures separate.
Live stock, prices, performance and destinations were not checked. The ordinary Ulta citation syntax and tracking parameter do not establish a malformed link. Citation presence does not explain product selection.
Q04 · Q04 · Q04 · Q04 · Q04 · Q04 · Q04Original Shape Tape: coverage and comfort are separate choices
Observation. Both answers retain conditional support for Original Shape Tape where full coverage and wear matter, while steering very dry everyday under-eyes away. OpenAI offers Creamy conditionally; both discuss NARS for comfort and e.l.f. for value/full coverage. Meta’s Radiant alternative is in an unattributed quotation, so its speaker remains unclear.
Possible business relevance. A broad positive or negative brand label would hide the actual shopper trade-off. Variant choice and expectation setting are more actionable than treating dry-skin caveats as an answer contradiction.
Asset to examine. Original / Creamy / Radiant comparison and use-condition guidance, including application/preparation claims where supported.
Investigation / conditional hypothesis, as supplied. Review suitability and wear evidence by need, then examine whether variant descriptions communicate the established limits. Preserve supported caveats rather than trying to remove them from future answers.
Proposed future retest. Prespecify two future shopper conditions: dryness-first daily wear and coverage/wear-first use. Measure accurate variant choice, retained caveats and attribution in each condition; do not combine them into one sentiment or quality score.
The same product can legitimately have different stances under different conditions. Reviewer disagreement and unclear quotation voice do not establish incompatible narrator facts. Product claims remain unverified.
Q06 · Q06 · Q06 · Q06 · Q06 · Q06 · Q06 · Q06 · Q06What remains uncertain
Product claims, current prices, availability and source ownership have not been independently verified. The study cannot explain why a product or source was selected. The methodology describes the remaining source questions.
Methodology and limitations
Collection: 3 November 2026, 09:52–10:09 UTC. United States; en-US; USD. Ten exact shopper prompts, five unbranded and five branded; one selected response per provider/question. The recorded model identifiers are OpenAI gpt-6-luna and Meta muse-spark-1.3-contributor. These identify the configurations recorded during collection; equivalence to the consumer ChatGPT or Meta AI experience has not been verified.
Response selection and collection limits
The ten OpenAI answers completed in the initial collection. The ten Meta answers were selected from their first successful responses in a later recovery collection, following failures in the initial collection. Provider collection was not simultaneous. This is not a repeated-response sample or a model reliability benchmark.
| Provider | Usable answers | Citation occurrences | Retrieved-source records |
|---|---|---|---|
| OpenAI | 10 | 43 | 597 |
| Meta Muse Spark | 10 | 155 | 629 |
| Combined | 20 | 198 | 1,226 |
Collection settings and analysis method
Answers used native-auto web-search grounding, medium search context and one repetition. Progress narration was excluded from substantive analysis. Citation scope follows the saved answers; differences in capture and annotation format limit provider comparisons.
The automated analysis used Claude Sonnet 5 and 5.5 through Claude Platform on AWS to classify statements in the saved answers. This publication preserves that analysis and its values. No models were rerun and no recommendations were reclassified for this publication.
A product counts as a recommendation when the answer itself presents it as a positive or conditional choice for the shopper’s actual need. Alternatives can qualify; passing mentions, quoted opinions, rejected products and instructions to buy an already-selected item do not. Preferences must be stated, rather than inferred from list order. Product overlap follows the report’s recorded names and conservative treatment of variants; catalogue equivalence has not been independently verified. In the detailed tables, ‘narrator’ means the answer’s own recommendation rather than quoted material. ‘Retrieval eligible’ identifies answers with usable search records; it does not mean that every retrieved source was used.
How to interpret the detailed comparisons
The published analytic tables and original figures retain the report’s values. Some narrative comparisons differ from eligibility flags or annotation counts; the qualifications next to Q01, Q03, Q07 and Q10 explain these differences. Historical shopper-fit statuses describe communication coverage, not separate constraint-satisfaction judgments. The Q04 budget-wording assessment must not be read as proof that the listed price exceeds budget. No composite score or model ranking is created.
The reasons for choosing these topics, budget levels and collection times were not documented. Earlier prompts mixed markets; those earlier answers were excluded after the questions were revised for the United States and USD. These twenty selected answers cannot establish representative market behaviour, repeatability, statistical significance, causal content effects, conversion outcomes or revenue impact.
Kojable reviewed this report. Tarte was not involved in the research or review. This report review does not establish a separate human check of the automated recommendation classifications. The source analysis records 36 open questions about its sources; product claims, prices, stock, promotion eligibility and other external facts remain unverified.
All exact prompts and intent labels
| ID | Purpose | Original shopper question |
|---|---|---|
| Q01 | Unbranded discovery | I'm shopping in New York and need a full-coverage concealer for dark circles that won't look cakey. Which three products from different brands should I consider, and where can I buy them? |
| Q02 | Unbranded discovery | I'm 45 with dry under-eyes and fine lines. What concealer gives good coverage without settling into lines? Compare three options. |
| Q03 | Unbranded discovery | What's the best long-wearing foundation for combination or oily skin under $45 USD in the United States? Compare coverage, finish and value across three brands. The budget is before sales tax and shipping. |
| Q04 | Unbranded discovery | I want a tubing mascara under $35 USD, available in the United States, that won't smudge during a long day and removes without harsh rubbing. What should I buy? The budget is before sales tax and shipping. |
| Q05 | Unbranded discovery | I'm looking for a juicy, glossy, slightly plumping lip product under $35 USD, available in the United States. Which three brands and products are worth trying? The budget is before sales tax and shipping. |
| Q06 | Branded · consideration | Is Tarte Shape Tape still worth buying, or should I choose NARS Radiant Creamy Concealer or e.l.f. Hydrating Camo Concealer? What are the trade-offs for dry under-eyes? |
| Q07 | Branded · product evaluation | Tarte has several Shape Tape concealers. Which version suits oily, blemish-prone skin versus dry under-eyes, and how should I choose my shade? |
| Q08 | Branded · product evaluation | I've seen Tarte Maracuja Juicy Lip recommended on social media. Is it worth the hype, and what do reviewers say compared with similar lip products? |
| Q09 | Branded · policy explanation | I only buy cruelty-free makeup and prefer vegan formulas. Is Tarte entirely vegan, and how can I verify whether a specific concealer or mascara is vegan? |
| Q10 | Branded · purchase facilitation | I'm in the United States and ready to buy Tarte Shape Tape concealer and Tartelette tubing mascara. What are the current prices in USD ($), total cost, US shipping threshold and most direct place to purchase? Stop before checkout. |
Technical appendix B — Belgian Boys first-public replication evidence
This appendix documents the 5 November 2026 data collection, not a prior public publication. The original Belgian Boys research HTML was unpublished when this consolidated report was prepared. The separate 20-answer pilot is excluded from all values below.
B1. Belgian Boys original question text
The ten questions below are transcribed from the user-supplied, unpublished Belgian Boys HTML research record. All had three fresh saved responses per provider in the replication collection. Their exact text should still be reconciled with the original API request logs during final release QA.
| Question ID | Question text from the saved research record | Panel |
|---|---|---|
U01 | “I have two school-age kids and need breakfasts I can prepare in under five minutes. What store-bought options should I consider?” | Unbranded |
U02 | “What are the best ready-made mini pancakes for busy weekday mornings? Compare three products available in US grocery stores.” | Unbranded |
U03 | “I want Belgian-style waffles without making batter. Which ready-made products should I buy?” | Unbranded |
U04 | “What are the best ready-made crêpes I can buy in the US for an easy weekend brunch?” | Unbranded |
U05 | “What are some good ready-to-eat sweet snacks for kids that can be served cold or warmed quickly?” | Unbranded |
B01 | “Are Belgian Boys Bite-Sized Pancakes worth buying compared with other ready-made mini pancakes?” | Branded |
B02 | “What's the difference between Belgian Boys Bite-Sized Waffles and Original Belgian Waffle? Which would you recommend for busy mornings?” | Branded |
B03 | “How do Belgian Boys Original Dairy Crêpes compare with other ready-made crêpes?” | Branded |
B04 | “Which Belgian Boys products are dairy-free or gluten-free? What should I verify on the packaging?” | Branded |
B05 | “I want to buy Belgian Boys Bite-Sized Pancakes in New York City, ZIP 10001. Where can I purchase them, and what prices and stock can you actually verify?” | Branded |
B2. Every unbranded provider-by-question numerator and denominator
Each denominator is three fresh assessable answers to the same exact question/provider. Counting applies once per target brand per answer; passing, quoted or rejected mentions do not qualify.
| Question | OpenAI | Meta Muse Spark |
|---|---|---|
U01 | 0/3 | 0/3 |
U02 | 1/3 | 0/3 |
U03 | 2/3 | 3/3 |
U04 | 3/3 | 3/3 |
U05 | 0/3 | 0/3 |
| Total: five questions × three saved answers/provider | 6/15 | 6/15 |
The OpenAI sum is 0+1+2+3+0=6; the Meta sum is 0+0+3+3+0=6. The role of U02 is conditional in its one OpenAI inclusion. Those totals were checked against the original question-summary table; no independent full-record replay is claimed here.
Recommendation-role detail from those same saved summary tables: In U03, 1/3 OpenAI and 2/3 Meta answers included a conditional Belgian Boys recommendation. In U04, 0/3 were coded as conditional for either provider. Presence and conditional-role fields are separate: do not add them together as if they were mutually exclusive answer counts.
B3. Separate branded comparison evidence
For B02, Original Belgian Waffle was the primary recommendation in three of three OpenAI and three of three Meta saved responses to that brand-named format comparison. Bite-Sized Waffles was described as a conditional choice for a different situation. These six branded answers do not enter the unbranded 6/15 metric. Saved model product-use claims remain unverified.
B4. Replication, coding and review
Belgian Boys' 60 answers were collected between 13:24:09 and 13:43:01 UTC on 5 November 2026, with ten purposively chosen questions, two providers and three fresh responses/provider/question. The original 20-answer pilot was analysed separately. Automated analysis used Claude Sonnet 5.5; Piush spot-checked 20/60 replication answers and their coding, covering all questions, both providers and all three repetition positions, with no corrections. This was a coding spot-check, not independent testing of ingredients, allergens, stock, price or commercial claims.
Technical appendix C — Metric dictionary and claim boundaries
| Item | Definition used in this report |
|---|---|
| Saved answer | An assessable stored model response to an exact question in the recorded collection; not a unique human shopper, web impression or sale. |
| Target-brand recommendation inclusion | Answer narrator offers a positive or conditional target-brand product choice for the stated need; counted once per answer/brand. Eligible alternatives count. |
| Exclusions | Passing mention, quoted third-party opinion, rejected option, unprompted brand-text interpretation, and instructions to buy an already selected item. Branded prompt results are separately labelled and excluded from unbranded discovery numerators. |
| Tarte unbranded rate unit | Qualifying answer count over five selected unbranded saved answers per provider (Q01–Q05), one answer per question/provider, collected 3 Nov 2026. |
| Belgian Boys unbranded rate unit | Qualifying answer count over 15 unbranded saved answers per provider (U01–U05 × three fresh responses), collected 5 Nov 2026 in one short window; pilot excluded. |
| Primary / conditional role | Primary = an answer expressly favours an option for the stated decision; conditional = it offers an option when an additional condition applies. Both can qualify for presence but must retain their different roles. |
| Named product / variant | Product string and any variant that the saved answer communicated; not automatically a manufacturer-reconciled current SKU. |
| Within-window recurrence | How many of the three saved same-question/provider responses contained a qualifying recommendation. Not a longitudinal stability estimate or population probability. |
| Citation/source annotation | A visible saved citation or source record under provider-specific capture conventions. It does not, by itself, demonstrate source use, factual support, internal ranking order or causal influence on a product choice. |
| 80 selected answers | 20 Tarte + 60 Belgian Boys as an inventory of two differently designed first-party investigations only. There is no pooled recommendation rate. |
| Evidence grade | A: directly recorded selected-answer observation; B: short-window repeated pattern; C: bounded interpretation; D: demonstrated intervention effect (none here); E: unsupported causal or universal claim (excluded). |
Transparency limit: The public page provides the source-reported figures and question-level summaries needed to understand the findings. Exact raw API-response bundles and a claim of end-to-end independent reproducibility require additional export, transformation and rights verification before any public release. The historically cited Tarte page is being replaced by this combined research destination, not silently erased from the underlying research archive.
Questions this research addresses
What do brand-level AI shopping recommendation counts leave out?
Why can equal Tarte recommendation totals conceal different shopper needs and products?
How did Belgian Boys recommendations vary across five unbranded breakfast and snack questions?
What does a three-response same-question collection establish?
Why are branded comparisons excluded from unbranded discovery counts?
What distinguishes a primary recommendation from a conditional alternative?
Can a saved citation establish why a model recommended a product?
Do these API-answer studies represent consumer ChatGPT or Meta AI sessions?
Related terms
AI shopping recommendations · AI product recommendations · shopper-needs analysis · brand-level recommendation counts · product variants · primary recommendation · conditional recommendation · Tarte Cosmetics · Belgian Boys