First-party experiment · November 2026 baseline

We measured Kojable's AI visibility. Here's what we're investigating.

On 13 November 2026, we put Kojable through the kind of AI-answer measurement we recommend to B2B companies. We tested 12 unbranded buyer questions three times each across five AI surfaces. Kojable is an AI answer alignment platform for B2B companies, so we wanted to see how our own work appeared when buyers described the problems we help investigate.

Published By Kojable

Kojable's AI Visibility Score was 6.67%, with each question–platform pair weighted equally. We were named in 12 of 177 assessable answers. But the more useful finding was not the score: on a question directly relevant to Answer Intelligence, AI systems sometimes cited our pages without naming Kojable in the answer.

That is the gap we are investigating. We have verified the baseline. We have not yet claimed that a content change has fixed it.

Experiment status: Baseline verified · Product-page change under consideration · Follow-up results pending

In brief

  • Starting point: 180 valid observation opportunities, 177 answers to assess, and a 6.67% equally weighted visibility score.

  • Specific gap: Kojable was named in five of 15 Answer Intelligence answers; seven cited a Kojable page, including two that did not name us.

  • Next action: Consider one focused improvement to the Answer Intelligence product page, record exactly what changes, and retest the same buyer question after publication.

What we measured on 13 November

We wanted to know whether Kojable enters answers when someone describes a business need without mentioning our company.

The panel contained 12 unbranded questions. We ran each question three times on:

  • ChatGPT

  • Gemini

  • Google AI Overview

  • Google AI Mode

  • Claude

That produced 180 planned checks under US-English settings. The questions covered alignment, AI brand audits, answer quality, inaccurate claims, citation monitoring, vendor comparisons and related B2B needs.

The collection methods were documented rather than treated as interchangeable. DataForSEO captured the ChatGPT, Gemini, Google AI Overview and Google AI Mode surfaces. Claude was measured through Anthropic on AWS using Sonnet 5 with web search enabled. For the two questions examined closely below, the recorded model labels were GPT-6 for ChatGPT and 3.5 Flash-Lite for Gemini; no model identifier was recorded for the two Google Search surfaces. These are the tested collection conditions, not a claim to cover every consumer account or product mode.

We retained all 180 valid observation opportunities. Google showed an AI Overview on 33 of 36 searches. The other three were valid searches where no Overview appeared, so there was no AI Overview answer to score. They were not collection failures.

Five Google AI Overview requests did fail during the original collection. We kept those failed records and reran only the five affected checks, recording the replacement attempts and timestamps separately. The two questions discussed below did not use replacement evidence.

For the full approach to freezing questions, conditions and retries, see our AI Visibility Tracking Baseline Guide.

Our starting AI Visibility Score was 6.67%

The headline score gives each of the 60 question–platform pairs equal weight. For each pair, we calculated the share of its three valid observations that named Kojable, then averaged those shares. The 60 pair-level rates summed to 4.0; divided by 60, the result was 6.67%. That does not mean only four pairs ever named Kojable: a pair with one mention in three repetitions contributes one-third to the sum.

There are two other numbers worth keeping separate:

Overall November 13 measures
MeasureNovember 13 resultWhat the denominator represents
AI Visibility Score6.67%Equal-weight average across 60 question–platform pairs
Kojable named in an answer12 of 177 (6.78%)Answers that actually appeared and could be assessed
A Kojable page appeared as a source15 of 180 (8.33%)Valid observations, including searches with no AI Overview

The source measure counts an observation once even if its answer cites several kojable.com pages. It does not count every URL separately.

All 12 answers that named Kojable also included a Kojable domain citation. Three additional observations cited our site without naming the company.

These are different signals. A mention tells us the company appeared in the answer. A citation tells us a source link was displayed. Neither establishes that the company was appropriately recommended, accurately described or selected because of that source.

Our AI Visibility Score explanation describes the measurement contract in more detail. Here, the score is a starting point for deciding which answers deserve attention.

The Answer Intelligence question revealed the clearest gap

One question brought the next investigation into focus:

Which Answer Intelligence tools help a B2B marketing team find inaccurate claims, missing context, and misleading competitor comparisons in AI answers?

Across 15 answers to this exact question, Kojable was named five times. Seven answers listed a Kojable page as a source, and four cited the exact Answer Intelligence product page.

The outcome varied by surface:

Answer Intelligence question results by AI surface
AI surfaceAnswers naming KojableAnswers citing kojable.com
ChatGPT3 of 33 of 3
Gemini0 of 31 of 3
Google AI Overview0 of 30 of 3
Google AI Mode1 of 32 of 3
Claude1 of 31 of 3
Total5 of 157 of 15

ChatGPT named Kojable in all three answers and cited the Answer Intelligence product page each time. Claude named us once and cited an Answer Intelligence article.

The two cases we want to examine more closely occurred elsewhere.

In one Gemini answer, the source list included Kojable's LLM Info and Growth Engine pages, but the vendor shortlist did not name Kojable. Those links appeared in the answer-level source list. We cannot say they supported a particular vendor-choice statement.

In one Google AI Mode answer, the Answer Intelligence page was linked to passages explaining general capabilities, while the answer did not identify Kojable as a provider. Another AI Mode answer did name Kojable, using a different Kojable product page as a source.

Those are observable differences between source attribution and company inclusion. They are not evidence that either system deliberately demoted Kojable, preferred a competitor or used our pages to make a particular recommendation.

Our existing guide to citations without brand mentions explores that distinction in general. Here, we have two specific, dated observations against a buyer question that names the capability we offer.

A plain-language question produced no Kojable mentions

We also tested a question that described the problem without using the Answer Intelligence label:

AI assistants keep saying our software lacks a feature it supports. Which providers can check the claim against our documentation and recommend what to update?

Kojable recorded zero mentions and zero owned-domain citations across all 15 answers.

That matters because a B2B buyer may describe the task they need completed rather than the name of a tool category. It is reasonable to investigate whether our public explanation makes that practical use case sufficiently clear.

But this is not a controlled phrase comparison. The two questions differ in scope and wording, so their results cannot isolate the effect of saying “Answer Intelligence”. Nor does the absence of a Kojable link prove that a page was not indexed or retrieved.

What we can diagnose, and what remains unknown

We know which questions produced named mentions, which displayed Kojable sources and where those signals diverged. We can inspect the retained answers and compare any material claims against current company information.

We do not know the full unseen chain by which each AI system found, selected, combined or cited information. A visible citation does not reveal an internal retrieval ranking or prove why a company was left out of a shortlist. An answer with no Kojable citation tells us even less about the first stage at which Kojable stopped being considered.

The patterns also differ between surfaces, and each surface had only three repetitions for a given question. We are treating them as a dated diagnostic signal, not a permanent characteristic of ChatGPT, Gemini, Google or Claude.

For us, the decision is practical: which gap can we substantiate, what is the smallest useful change, and what would we need to see before saying the answer improved?

The page change we are considering

Our Answer Intelligence page already explains that Kojable reviews AI-generated descriptions, omissions and competitor framing, checks relevant claims against reliable information with human review, and helps teams prioritise actions and retesting.

What it does not yet show in one place is a complete example of what the review gives the buyer.

We are considering an explicitly labelled, fictional worked example that connects:

  1. The buyer question and the AI claim worth checking.

  2. The current documentation or other reliable evidence used by a human reviewer.

  3. The public page or explanation that may need updating.

  4. The exact buyer question and claim that would be checked again.

That would make the path from claim → evidence → proposed action → retest easier to inspect. It would not imply that Kojable automatically corrects third-party AI answers, or that a published edit guarantees a changed recommendation.

This is a candidate intervention, not an implemented fix. We will review the precise wording and delivery claims before deciding whether to make it. We are also deliberately keeping the full proposed example off this baseline article so the product-page change can be documented separately.

How we will check what happens next

If the edit is approved, we will preserve the existing product-page copy, record the exact diff, and note the canonical URL, approval and deployment timestamp.

We will also log the publication of this case study as a separate change to Kojable's public information. Publishing the baseline before editing the product page is itself part of the environment we are studying. It makes a simple causal claim about any later answer movement harder, so we will not hide it.

Our first focused follow-up will use the same v2 Answer Intelligence question, three times on each of the same five surfaces: 15 planned checks, under comparable US-English conditions. The incorrect-feature question is a secondary check and will be reported separately if rerun.

We will record three outcomes:

  • Named in the answer: Does the answer name Kojable? The primary question's baseline is 5 of 15. We will separately check whether the company is presented as a relevant option for the buyer's need.

  • Exact product-page citation: Does the answer list the updated Answer Intelligence page as a source? The baseline is 4 of 15.

  • Answer quality: If named, does the answer accurately explain evidence review, practical next steps and retesting, without inventing automatic correction or guaranteed placement?

The first checkpoint is planned for 14 days after any approved product-page edit goes live, followed by an optional 30-day check. Those are observation dates, not deadlines by which an AI platform must have processed the new page. Google's own recrawl guidance notes that discovery and recrawling can take time and do not guarantee inclusion.

We will use our AEO change-verification method to report whether the target signal Moved, Held, revealed a New Gap or was Inconclusive. An answer changing after publication is evidence of a changed observation, not automatically proof that the edit caused it.

Experiment log

We will keep the November baseline intact and add dated updates as work happens.

Baseline, publication and follow-up events
EventCurrent statusDate or trigger
V2 AI visibility baselineVerified13 November 2026
Baseline case study on Kojable.comPublished14 November 2026
Answer Intelligence worked-example editUnder considerationApproval and deployment not yet recorded
First focused follow-upNot run14 days after any approved edit goes live
Second checkpointOptional30 days after that edit

The next update will include the page changed, its publication time, the measured answers and citations, and the limits of the comparison. We will report unchanged or mixed results with the same care as improvements.

Limits of this baseline

This is a deliberately selected set of 12 unbranded buyer questions, not a sample of every query B2B buyers ask. The 6.67% figure describes this panel and its weighting, not Kojable's visibility across the whole market.

Three repetitions per question and surface helped reveal variation, but they do not establish long-term stability. The five measurement surfaces used different collection paths and do not represent every possible account, personalisation setting or consumer interface. Some Google searches did not display an AI Overview, and technical collection failures were repaired with their original attempt records preserved.

Our source lists are evidence of what was displayed, not a complete view of hidden retrieval or vendor selection. The first follow-up will be a before-and-after observation with documented conditions and potential confounders. We will not present it as a causal trial.

The underlying answer, citation, scoring and repair records have been retained by Kojable. This public summary gives the essential method and denominators; it does not imply that private local files are available at a public URL.

Frequently asked questions

Why is the AI Visibility Score 6.67% when 12 of 177 answers named Kojable?

The 6.67% headline is the mean of 60 equally weighted question–platform mention rates. The 12 of 177 (6.78%) figure is the share of generated, assessable answers that named Kojable. Three valid Google searches did not show an AI Overview, so they were exposure opportunities but not assessable AI-generated answers.

What does it mean when an answer cites Kojable but does not name it?

It means the visible source list included a Kojable URL while the generated answer omitted the company name. It does not establish that the URL caused the answer, that Kojable was considered and rejected, or that the system used the page for training.

Did the experiment prove why Kojable was omitted?

No. It identified specific, inspectable gaps between brand mentions and citations. The internal discovery, retrieval and selection process remains partly or entirely unobservable from the captured answers.

Have we already updated the Answer Intelligence page?

No approved worked-example edit is recorded in this case study yet. The proposed content is still being reviewed. We will add the exact change and deployment time if it is published.

When will we first check for changes?

The first observation is planned for two weeks after an approved edit goes live. We will compare the frozen question under comparable conditions, then report the named mentions, exact citations and accuracy of the answers. A result may be improved, unchanged, mixed or inconclusive. None of those outcomes automatically establishes causation.

What comes next

We are applying the same standard to Kojable that we ask other companies to use: measure the answer, inspect the evidence, choose a justified improvement, and verify what changes.

It would be convenient to publish a story in which one edit immediately improved the score. We do not have that result, and we are not going to write it in advance.

For now, the result is a verified baseline and a specific question worth investigating. If nothing moves after the first edit, that will be part of the case study too.

Learn more about Kojable's Answer Intelligence capability and how it connects claim review, evidence and practical next steps within our broader AI answer alignment platform.

About Kojable

KojableAI answer alignment platform