How we score

AI shopping engines (ChatGPT search, Perplexity, Google AI Overviews, Claude) can only recommend a product they can fetch, understand and trust. The score measures how well a single product page supports that: whether crawlers are allowed in, whether the facts are machine-readable, whether the copy answers what shoppers ask, and whether everything agrees with itself.

It measures readiness, not outcomes. A score of 78 means the page earned 78 of 100 points on these factors, not a 78% chance of being recommended.

Checks that don't apply to a page (for example review markup on a page with no reviews) are marked N/A, and the rest of that category is scaled up so the page isn't penalized for them.

• Excellent: 90–100
• Strong: 75–89
• Developing: 60–74
• Weak: 40–59
• Critical: 0–39

A few problems make everything else irrelevant, so they cap the score no matter how good the rest of the page is:

• The page is marked noindex, or robots.txt blocks every major AI search crawler: capped at 35.
• No product markup, visible price or buy button, so the page doesn't present a purchasable offer: capped at 50.

Most checks are deterministic: we parse the HTML, structured data and robots.txt. The checks marked AI-graded judge the copy itself, such as whether it says who the product is for or answers typical buyer questions. For those, an AI model reads the page and must quote the exact text that supports each grade. We verify every quote against the page, and a grade whose quotes aren't found is reduced.

If the AI model isn't available, keyword rules grade those checks instead, and the report says so and shows lower confidence.

If bot protection blocks our request, the page doesn't load, or the site's robots.txt asks our crawler to stay out, the page is marked not scored rather than given a low score. You can paste the page source instead. More about our crawler.

Each report shows a confidence figure separate from the score. It drops when we had to work from less: pasted source instead of a live fetch, no JavaScript rendering for a page that needed it, missing product markup, or keyword grading instead of the AI model.

• Off-site reputation: reviews on other sites, Reddit threads, "best of" articles, press and videos strongly influence which products AI engines recommend. A URL audit can't see them; the visibility probe shows which sources engines actually cite.
• Product feeds: Google Merchant Center and ChatGPT merchant feeds can outweigh on-page markup for shopping results. Connect a feed (coming later) to audit it.
• Firewall rules: a CDN or WAF can block AI crawlers by user agent even when robots.txt allows them. We can only see robots.txt.
• Price, shipping speed, seller reputation and the shopper's location also affect recommendations and are outside this score.

The category weights are a considered starting point, and we're calibrating them against expert reviews and against what AI engines actually answer. When the scoring changes, the version number changes, and every report shows the version it was scored with, so scores from different versions aren't silently compared.