AI citation-gap investigation
Compare a fixed set of buying questions, cited pages, merchant attribution, product facts, and competing products.
Who does this: A written procedure for an AI coding agent or a researcher to follow. Forward it, don't work through it by hand.
Run this investigation when AI shopping answers omit a product, cite a competitor, or state the merchant's facts incorrectly.
Inputs
- Fixed question set
- Named model or answer system
- Products and markets
- Timestamped answers
- Relevant competitors
- Merchant-owned product sources
Output
- Product-visibility and citation table
- Incorrect or unsupported statements
- Patterns in cited sources
- Merchant-owned pages or data that may need work
- Retest plan
Steps
- Group questions by buying intent.
- Record answers, citations, merchant attribution, and competing products.
- Compare cited pages with merchant-owned product facts.
- Compare absent products with the products the system selected.
- Write a testable explanation involving content, catalog data, or entity identity.
- Change one merchant-controlled source when there is a sound reason.
- Run the same questions again and record the difference.
Checks
- Whether the product appears
- Accuracy of product facts
- Merchant attribution
- Cited pages
- Competing products
- Market, model, and timestamp consistency
Limits
Related pages
Keep working on this issue
Open-source project
Promptfoo
A framework for evaluating prompts, agents, and model behavior with repeatable test cases and assertions.
Open standard
Schema.org
The shared structured-data vocabulary used to describe products, offers, organizations, reviews, and web entities.
Primary reference
Google guide to generative AI features in Search
Google's current guidance on AI Overviews, AI Mode, SEO fundamentals, measurement, and claims around AEO and GEO tactics.
Open-source project
Crawlee
A web crawling and browser-automation library for collecting repeatable evidence across catalogs and storefronts.
Beseam project
Beseam AI Discovery Scan
A live scan for observing how products are represented across AI shopping and answer surfaces at a point in time.