You do not need a vendor to find out where you stand. These six checks cost nothing but an afternoon, and between them they surface most of what an audit would tell you.
Check 1 — Ask the assistants directly
Write down ten shopping questions a real customer would ask for what you sell. Not brand questions — category questions with constraints. "Best engagement ring under three thousand dollars with a lab-grown diamond." "Fourteen karat gold hoops that won't tarnish." "Jewelry brands with free resizing."
Run all ten in ChatGPT, Gemini, Perplexity and Claude. Fresh session each time so you are not seeing your own history reflected back. Record three columns: does your brand appear, which competitors appear instead, and which sources get cited.
That third column is the most valuable thing you will produce. It tells you precisely which sites the assistants trust in your category — and therefore where your brand needs to exist.
Check 2 — Read your own page the way a machine does
Open one of your product pages and view the page source, then search it for your price. If the number is not there, an assistant arriving at that page does not see a price either. Same test for stone type, metal karat and availability.
Many jewelry sites render these only after JavaScript executes. It looks fine to you and is invisible to a large share of automated readers.
Check 3 — Confirm the crawlers are allowed in
Visit yourdomain.com/robots.txt. You are looking for whether the AI crawlers are permitted — GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and their relatives.
Then check your security or performance layer separately. Several popular services block AI crawlers by default, and this does not appear in robots.txt at all. Merchants are routinely blocking the assistants without knowing they made that decision.
Check 4 — Test your structured data
Run a few product URLs through Google's Rich Results Test. You want to see Product and Offer markup detected, with price, availability and brand populated. Ratings and return policy if you have them.
If nothing is detected, that is your first project, and it outranks everything else on this list.
Check 5 — Look for your catalog contradicting itself
Pick twenty products. For each one, compare the displayed title, the page title tag, and the words in the URL. They should describe the same object.
They frequently do not — the title names one stone, the URL names another, the title tag is a machine-generated slug. This is the failure that does the most quiet damage, because inconsistency is read as unreliability rather than as missing information.
Check 6 — Open your feed and read fifty titles
Export your product feed and read fifty titles at random. For each, ask whether a stranger could identify metal, stone, and product type from the title alone.
Anything reading "Style 1042," "Classic Ring" or simply "Charm" is unreachable by any query more specific than those words.
Scoring it honestly
Zero or one problem found: you are in unusually good shape for this category. Two or three: normal, and each is fixable without a rebuild. Four or more: your products are effectively invisible to AI shopping, and the good news is that the highest-impact fixes — titles and structured data — are template changes rather than data projects.
Whatever you find, write down the date and the results. In six months that becomes your baseline, and a baseline is the only way to prove any of this worked.