Opens in a new tabSkip to content
Agent LighthouseAgent Lighthouse

    Searches the text of every published page. The evidence sources themselves are not in this index — search all of them on the trusted sources page.

    GitHub ↗
    Browse checks and page contents
    answer-readiness/review-signals

    Review/testimonial signals

    What it checks

    Social proof an agent can actually read.

    State Result
    JSON-LD Review/AggregateRating carrying a rating value or a non-zero reviewCount/ratingCount, on a homepage or product page; or a <blockquote> with attribution on any scanned page pass
    review UI with nothing machine-readable behind it — client-injected widget markup, or visible “N reviews” text — on a homepage or product page warn, priority medium
    none of the above fail, priority medium

    Attribution means a cite attribute naming a document, or a non-empty <cite>, <footer>, or <figcaption> in a wrapping <figure>.

    The review vocabulary is read only from homepage and product pages: that is the population Google’s review rich results and OpenAI’s review_count/star_rating cover. The quotation branch is not scoped that way, because its GEO measurement is about generative-answer citation on content generally.

    An empty <blockquote> is ignored. A node stating reviewCount/ratingCount of 0 is not social proof, and neither is the vocabulary without a value behind it — "aggregateRating": {}, "aggregateRating": true and a bare {"@type":"Review"} all count for nothing. An unattributed pull-quote is reported but sets no status.

    Why it matters

    Machine-readable review data is read by named consumers. Google Search parses schema.org Review and AggregateRating to render review rich results, and OpenAI ingests review_count and star_rating to build ChatGPT product results. A product page that exposes ratings only as pixels is therefore invisible to both, while one that exposes them structurally is not.

    Evidence

    • Google documents parsing the markup and the feature it drives: “When Google finds valid reviews or ratings markup, we may show a rich snippet that includes stars and other summary info from reviews or ratings”. The feature is supported on Book, Course, Event, Local business, Movie, Product, Recipe, Software App and further schema.org types — developers.google.com/…/review-snippet (verified 2026-08-21)
    • OpenAI’s commerce specification carries first-class review fields for ChatGPT product results — review_count (“Number of product reviews”), star_rating (“Average review score”), store_review_count, store_star_rating, reviews — introduced as: “Supply aggregated review statistics and frequently asked questions. User-generated insights strengthen credibility and help shoppers make informed decisions.” — developers.openai.com/…/feed (verified 2026-08-21)
    • AggregateRating is core, ratified schema.org vocabulary: “The average rating based on multiple ratings or reviews”. It carries ratingValue, reviewCount and ratingCount, deployed on 1M–10M domains per the Google July 2026 web index sample shown on the type page — schema.org/AggregateRating (verified 2026-08-21)

    Limits

    The OpenAI spec “does not address how OpenAI/ChatGPT obtains product data outside of feed submissions”. It says nothing about crawling merchant pages, or about reading on-page schema.org. The ChatGPT consumer path therefore does not directly validate the on-page signal (developers.openai.com/…/feed, verified 2026-08-21). Google states that for AI Overviews and AI Mode “There’s also no special schema.org structured data that you need to add” (developers.google.com/…/ai-features, verified 2026-08-21). Google also constrains what presence can mean.

    It prohibits “fake or undisclosed incentivized reviews on your page or in your structured data markup”, and requires that “Ratings must be sourced directly from users”. The existence of review markup is not itself evidence of social proof (developers.google.com/…/review-snippet, verified 2026-08-21). No published measurement links review markup to generative-answer citation rates, and nothing in any source supports counting an unattributed blockquote as a review signal.

    How it scores

    Consumption of the review vocabulary is documented on both a search surface and an AI surface. But the ChatGPT path runs through a submitted product feed, not the on-page markup this audit inspects. And no study measures a citation delta for review markup.

    Sources