How Kojable Integrated DataForSEO for Fan-Out Queries, Keyword Intelligence and AI Answer Alignment

Keyword demand in one market should not silently drive a plan for another. Kojable creates an immutable snapshot containing country code, language code, location code, location name, configuration or fallback provenance and a stable key: dataforseo:google:<location-code>:<language-code>.

Metric records can also carry provider, endpoint family, discovery-run identity, fetch timestamp, location and language provenance and a request fingerprint. The system distinguishes intended-market data, documented fallbacks, legacy values with unknown context and values that should not be reused after a market change.

Kojable’s shared DataForSEO API transport adds credential validation, configurable timeouts, bounded retry, safe error serialisation, latency and response-size measurement, attempt counts and endpoint-family telemetry without logging credentials or raw sensitive headers.

The LLM Mentions client validates both the root response and task envelope. Pagination guards reject repeated tokens or offsets that do not advance. AI-volume batch isolation records a failed batch while allowing remaining batches to continue.

DataForSEO describes AI Search Volume as a proprietary directional estimate with trend information, not a raw count from private ChatGPT, Perplexity or Google logs.

Kojable retains AI demand alongside traditional search volume, keyword difficulty, CPC, competition, trend direction, trend velocity, SERP weakness, user intent, cluster depth and business relevance. At cluster level, AI volume can be aggregated while a fast-win heuristic considers difficulty, SERP weakness, trend, depth and whether AI or traditional demand exists.

Kojable builds context from company name, organisation description, products and services, target audiences, approved and excluded themes, seed keywords, approved cluster pillars, primary market and workspace knowledge.

A deterministic pass identifies high-confidence mismatches. Unresolved candidates can enter a model-assisted classifier using stable candidate IDs and exactly one decision per candidate. Unknown IDs, duplicates, missing decisions and malformed confidence are rejected. Lexical overlap alone is insufficient and uncertainty is retained.

The current precision-first automatic rejection threshold is 0.80. Only non-authoritative candidates with a specific high-confidence irrelevance reason are removed. Invalid or unavailable output retains unresolved candidates as uncertain. This is a Kojable default, not a universal AEO recommendation.

Kojable preserves authoritative commercial, transactional and navigational intent. When new or niche phrases lack useful intent metadata, deterministic signals can correct it.

Transactional examples include buy, purchase, subscribe, free trial and download. Commercial examples include pricing, consultant, software, platform, provider, comparison, best, alternative and demo.

A candidate can be discovered from several seeds. Kojable can retain the strongest metric record while merging every seed relationship. Multi-seed provenance may identify a useful bridge between topic areas.

Accepted metadata can record source seed, stable source-seed ID, all contributing seeds, market, deterministic route, model decision, confidence and whether a fallback was used.

Lexically related phrases can carry different intent: AI visibility software, how to improve AI visibility, AI visibility pricing and AI visibility platform alternatives. Pure lexical collapse can erase those differences; keeping every variant can create competing content.

Kojable considers authoritative-seed status, opportunity score, intent clarity, search volume, canonical phrase length and introduced intent or modifier evidence. Rejections retain comparison, overlap evidence and reason code for audit.

Current defaults retain 50 candidates per seed after normalisation, allow 150 selected candidates before clustering, cap clustering inputs at 120, retain 20 rejection samples and write 25 lineage records per batch.

Preparation applies language policy, removes drop candidates, forms duplicate groups, elects a representative, retains bounded alternates, assigns stable IDs and applies capacity limits. These are operational safeguards, not universal recommendations.

Kojable separately checks zero members, duplicate topics and identities, cross-cluster duplicates, missing or multiple pillars, pillars outside membership, ineligible pillars, intent mismatch, weak pillar similarity, low peer cohesion, disconnected groups, outliers, source-seed conflict, excessive size and incomplete quality evidence.

The states are PASS, REVIEW_REQUIRED, BLOCKED and UNASSESSED. Current membership-quality-v1 thresholds include 0.65 dominant intent, 0.20 average pillar similarity, 0.34 peer overlap, 0.70 connected members and mandatory review above 40 members.

A deterministic SHA-256 identity derives from pillar, members, intent, seeds, state and diagnostics. Changed membership produces a changed evaluation, preventing old approval from applying to a new cluster.

Users can review exact members and the pillar, compare intent and demand, exclude topics, select a better pillar, acknowledge an eligible warning, defer or reject and approve only exact selected clusters.

The correction service is token-protected and idempotent, validates unchanged review state and blocks structural changes once campaign, sequence or calendar work exists.

Kojable evaluates topic relevance, company or owned-domain appearance, cited URLs, owned, competitor or third-party status and evidence category. Citation-domain categories are owned, competitor, publication, authority, community, marketplace and other.

The metrics layer calculates Citation Share of Voice, citation coverage, answer mention rate, AI-opportunity-weighted coverage, AI search opportunity, distinct owned URL count, leading citation domain and platform or model differences. See the related DataForSEO LLM Mentions case study.

AI-opportunity weighting matters because five low-demand citations can represent less strategic coverage than two high-demand citations. Kojable uses AI Search Volume rather than treating every observation as equal.