Signals Database

Signals Database

The structured catalog of the signals that shape which store, product, or offer an AI shopping agent selects. Every record is evidence-typed (spec-fact, reported, or hypothesis) with a confidence level and, where a platform documents its behavior, honest per-platform status.

The whole database is also published in machine-legible form: /signals.json (structured) and /signals.md (Markdown catalog).

39 signals, sorted by funnel stage (discovery, matching, comparison, choice), then impact.
SignalAffectsPlatforms documentedEvidenceConfidenceImpactLast verified
Agent infrastructure reachabilityDiscovery3/5 ChatGPT, Gemini, ClaudeHypothesisMedium3/3
Content equivalence (no cloaking)Discovery2/5 Gemini, CopilotSpec-factVery high3/3
Crawl directives (robots.txt + sitemap)Discovery4/5 ChatGPT, Gemini, Perplexity, ClaudeSpec-factVery high3/3
Market, language and currency eligibilityDiscovery3/5 ChatGPT, Gemini, CopilotSpec-factVery high3/3
Agent-readiness monitoring and ownershipDiscovery1/5 GeminiHypothesisLow2/3
llms.txt priority orderingDiscoveryNone documentedSpec-factVery high2/3
Machine-readable page representationDiscoveryNone documentedHypothesisLow2/3
Origin serve-ability for agentsDiscovery3/5 ChatGPT, Gemini, ClaudeHypothesisLow2/3
Track AI-answer presence, not just referralsDiscovery1/5 GeminiHypothesisLow2/3
Conversational product attributesMatching1/5 GeminiSpec-factVery high3/3
Offer markup completeness and truthfulnessMatching2/5 ChatGPT, GeminiSpec-factVery high3/3
Server-rendered, crawlable contentMatching2/5 Gemini, ClaudeHypothesisLow3/3
Answering agent tool callsMatching3/5 ChatGPT, Gemini, ClaudeSpec-factHigh2/3
Product identity and canonical URLMatching3/5 ChatGPT, Gemini, CopilotSpec-factVery high2/3
Validation hygiene (use existing validators)Matching2/5 Gemini, CopilotSpec-factVery high2/3
Variants and titles use the words shoppers typeMatching3/5 ChatGPT, Gemini, ClaudeHypothesisLow2/3
Competitive price awareness with parityComparison2/5 ChatGPT, GeminiReportedHigh3/3
Delivery ETA is computable per destinationComparison4/5 ChatGPT, Gemini, Copilot, ClaudeHypothesisLow3/3
Descriptions answer relayed buyer questionsComparison2/5 ChatGPT, GeminiHypothesisLow3/3
Facts per token, complete against peersComparison2/5 ChatGPT, GeminiHypothesisLow3/3
Offer cost legibilityComparison2/5 ChatGPT, GeminiHypothesisLow3/3
Loyalty benefits visible to the agentComparison1/5 GeminiSpec-factVery high2/3
Query-conditional rewrites are a targeted experimentComparison1/5 ChatGPTReportedHigh2/3
Return terms are fully legibleComparison4/5 ChatGPT, Gemini, Copilot, ClaudeHypothesisLow2/3
ACP feed and checkout readinessChoice1/5 ChatGPTSpec-factVery high3/3
Agent-payment rail readinessChoice3/5 ChatGPT, Gemini, ClaudeReportedHigh3/3
Audit position and ordering effects per engineChoice1/5 ChatGPTReportedMedium3/3
Brand familiarity versus a small rating edgeChoiceNone documentedReportedMedium3/3
Defined agent win-rate measurementChoiceNone documentedHypothesisLow3/3
Earn endorsement, never self-label sponsoredChoiceNone documentedReportedHigh3/3
Maker or primary seller statusChoice1/5 ChatGPTSpec-factVery high3/3
Offer consistency and freshness across surfacesChoice3/5 ChatGPT, Gemini, CopilotHypothesisLow3/3
Re-test after every major model updateChoice1/5 ChatGPTReportedHigh3/3
Review depth: count, recency, distributionChoice2/5 ChatGPT, GeminiHypothesisLow3/3
Guest checkout completabilityChoice1/5 ChatGPTHypothesisLow2/3
Personalization context you cannot seeChoice1/5 ChatGPTSpec-factVery high2/3
Price sensitivity varies by modelChoiceNone documentedReportedHigh2/3
Seller identity reachable in two hopsChoice2/5 ChatGPT, CopilotHypothesisLow2/3
UCP profile and order-management readinessChoice2/5 Gemini, CopilotSpec-factVery high2/3

Platform coverage matrix

Signals as rows, shopping surfaces as columns. A cell is filled only where a platform documents its own behavior (69 of 195); everything else readsunknown, which is the honest default and not a gap waiting to be filled in.

The shape of what is filled says more than the count. Google and OpenAI document feed and structured-data requirements in detail, so the signals about being eligible are well covered. Almost nothing is documented about how any engine ranks one eligible product against another. Microsoft documents a feed, Anthropic publishes a reference implementation that contains no ranking logic, and Perplexity documents the product data its Merchant Program expects and little else. The blank half of this table is the subject of this site.

This view is generated from the signal records; fixing a cell means fixing a signal, never this table.

Per-platform status for each signal. Unknown is the honest default where a platform has not documented its behavior.
SignalChatGPTGeminiPerplexityCopilotClaude
Agent infrastructure reachabilityPartialPartialUnknownUnknownPartial
Content equivalence (no cloaking)UnknownSupportedUnknownReportedUnknown
Crawl directives (robots.txt + sitemap)SupportedSupportedSupportedUnknownSupported
Market, language and currency eligibilitySupportedPartialUnknownSupportedUnknown
Agent-readiness monitoring and ownershipUnknownPartialUnknownUnknownUnknown
llms.txt priority orderingUnknownUnknownUnknownUnknownUnknown
Machine-readable page representationUnknownUnknownUnknownUnknownUnknown
Origin serve-ability for agentsPartialPartialUnknownUnknownPartial
Track AI-answer presence, not just referralsUnknownPartialUnknownUnknownUnknown
Conversational product attributesUnknownSupportedUnknownUnknownUnknown
Offer markup completeness and truthfulnessPartialSupportedUnknownUnknownUnknown
Server-rendered, crawlable contentUnknownSupportedUnknownUnknownSupported
Answering agent tool callsPartialPartialUnknownUnknownPartial
Product identity and canonical URLSupportedSupportedUnknownSupportedUnknown
Validation hygiene (use existing validators)UnknownSupportedUnknownPartialUnknown
Variants and titles use the words shoppers typePartialSupportedUnknownUnknownPartial
Competitive price awareness with paritySupportedPartialUnknownUnknownUnknown
Delivery ETA is computable per destinationPartialSupportedUnknownPartialPartial
Descriptions answer relayed buyer questionsPartialPartialUnknownUnknownUnknown
Facts per token, complete against peersPartialPartialUnknownUnknownUnknown
Offer cost legibilityPartialPartialUnknownUnknownUnknown
Loyalty benefits visible to the agentUnknownSupportedUnknownUnknownUnknown
Query-conditional rewrites are a targeted experimentPartialUnknownUnknownUnknownUnknown
Return terms are fully legiblePartialSupportedUnknownPartialPartial
ACP feed and checkout readinessSupportedUnknownUnknownUnknownUnknown
Agent-payment rail readinessPartialPartialUnknownUnknownUnsupported
Audit position and ordering effects per engineReportedUnknownUnknownUnknownUnknown
Brand familiarity versus a small rating edgeUnknownUnknownUnknownUnknownUnknown
Defined agent win-rate measurementUnknownUnknownUnknownUnknownUnknown
Earn endorsement, never self-label sponsoredUnknownUnknownUnknownUnknownUnknown
Maker or primary seller statusSupportedUnknownUnknownUnknownUnknown
Offer consistency and freshness across surfacesPartialSupportedUnknownPartialUnknown
Re-test after every major model updatePartialUnknownUnknownUnknownUnknown
Review depth: count, recency, distributionPartialSupportedUnknownUnknownUnknown
Guest checkout completabilitySupportedUnknownUnknownUnknownUnknown
Personalization context you cannot seeSupportedUnknownUnknownUnknownUnknown
Price sensitivity varies by modelUnknownUnknownUnknownUnknownUnknown
Seller identity reachable in two hopsPartialUnknownUnknownPartialUnknown
UCP profile and order-management readinessUnknownSupportedUnknownSupportedUnknown

[ newsletter ]