Try it yourself. Open ChatGPT and ask for the kind of piece you sell, with the kind of constraint a real shopper would use. "Best rose gold engagement ring under three thousand dollars." "Fourteen karat gold hoops that won't tarnish." "Where can I buy an ethical engagement ring."
Most jewelers who run that test for the first time get an uncomfortable result. Four or five names come back, none of them theirs, and several belong to brands they consider smaller or less established. The instinct is to assume the assistant is biased toward big spenders. It isn't. There is no ad auction in that answer. What there is instead is a scoring process, and most jewelry catalogs fail it before the question is even asked.
Assistants never see your website
This is the part that takes a while to sit with. When an AI assistant answers a shopping question, it is not browsing your site the way a customer would. It is not looking at your photography, your hero video, your typography, or the care you put into your collection pages. It reads data — your product feed and the structured markup embedded in your pages — and scores it.
That means the thing you have invested most in is invisible to the thing increasingly deciding what gets recommended. A beautiful site with thin product data loses to an ugly site with complete product data, every time.
The five failures we find most often
1. There is no price in the data
Plenty of jewelry sites show price only after JavaScript runs, or only on the product page and never on collection pages. To an assistant that arrives, reads the served HTML and moves on, that product has no price. Price is a primary input for any shopping recommendation. A product without one isn't ranked poorly — it is dropped from consideration entirely.
2. Titles are internal style numbers
"Style 1042." "Classic Ring." "Charm." These are perfectly sensible names inside a business that has used them for twenty years, and they are worth nothing to a machine. Compare that to a title that reads "14K Yellow Gold Oval Lab-Grown Diamond Halo Engagement Ring" — an assistant extracts metal, karat, stone type, stone origin, shape, setting and product type in a single pass. From "Style 1042" it extracts nothing at all.
3. Attributes exist as prose, not as data
Your free lifetime resizing, your lifetime warranty, your GIA certification, your two-week custom turnaround, your nickel-free construction — these are exactly what high-intent shoppers ask assistants about. On almost every jewelry site we look at, they live in a paragraph on a policy page. An assistant cannot filter on a paragraph. If it isn't a field, it may as well not exist.
4. The catalog contradicts itself
This one does more damage than missing data. When a product's displayed title says one stone, its web address says another, and its page title says a third, an assistant doesn't average them. Inconsistency reads as unreliability, and unreliable sources get downgraded across the board — not just for that product.
5. The crawlers are blocked and nobody knows
A surprising number of jewelry sites are invisible because a security or performance layer is turning AI crawlers away by default. The merchant never chose this and has no idea it is happening. It takes ten minutes to check and is the first thing worth checking.
Why fixing this is genuinely urgent for jewelry specifically
Jewelry has the worst product data in retail, and not because jewelers are careless. The category is structurally hostile to clean data. Custom and made-to-order pieces have no universal barcode. A single ring can vary by metal, karat, centre stone, carat weight and finger size at once, which exceeds the variant limits most platforms impose. Loose stone inventories run to tens of thousands of rows with prices that move daily.
That difficulty cuts both ways. It is why most of your competitors are also invisible — and why the ones who fix it first will be the default answer for a long time. Reviews, citations and cross-source consistency compound. They are not a switch you flip the month you decide to care.
What to do this week
Run ten real shopping prompts against your own brand and three named competitors, across ChatGPT, Gemini, Perplexity and Claude. Write down who appears, in what order, and which sources get cited. That last column is the most useful thing you will learn all quarter — it tells you exactly where the assistants go for answers in your category, and whether your brand exists there at all.
Then check whether your prices render without JavaScript, and whether AI crawlers are permitted to reach you. Those two answers usually explain most of the gap.