- Profound, a company that tracks AI search, studied more than 1.7 million ChatGPT Shopping prompts in July 2026.
- On July 10 the share of product picks coming from store feeds jumped from about 8% to about 62%. By early September it was around 65%.
- In the days after the change, 450 of the 687 stores Profound tracks lost at least a third of their ChatGPT Shopping visibility. 67 gained.
- Picks also got more concentrated: the top 10 stores went from 22.5% to 41.8% of all references.
- Shopify and Etsy catalogs are connected automatically. Most other stores are on a waitlist. OpenAI has not confirmed the change.
What changed on July 10
Until this summer, when someone asked ChatGPT to help them buy something, most of the products it showed came from web pages it found by searching. That changed almost overnight. Profound tracked ChatGPT Shopping prompts through July and saw the share of picks pulled from integrated product feeds go from 8.26% to 61.54% on July 10, 2026. By September 3 it was around 65%.
A product feed is the structured file a store sends to a shopping platform: titles, descriptions, prices, stock, images and attributes. In other words, ChatGPT stopped reading your website to decide what to recommend, and started reading your catalog.
The effect on stores was immediate. Of the 687 customers in Profound's sample, 450 lost at least a third of their ChatGPT Shopping visibility when you compare July 7 to 9 with July 10 to 12. Only 67 gained by the same margin. At the same time the number of different stores ChatGPT mentioned fell by more than 20%, and the top 10 stores went from 22.5% to 41.8% of all references.
ChatGPT stopped reading your website to decide what to recommend. It started reading your catalog.
Why this hits jewelry harder than most categories
Jewelry is bought on details. A shopper asking for "gold hoops that won't tarnish" or "a lab-grown solitaire under $2,000" is asking about the metal, the karat, the plating, the stone and the price. Those are exactly the fields a feed carries, and exactly the fields many jewelry catalogs leave half empty or bury in a paragraph of brand copy.
On a product page, a beautiful photo and a good story can do the work. In a feed, there is no photo shoot and no story. There is "14k solid gold" in a material field, or there is nothing. If the field is blank, an AI that is matching a shopper's question to structured data has no reason to pick you.
Who is already connected
- Shopify stores are connected through the Shopify catalog. You do not need to do anything to be in, but your catalog data is what ChatGPT sees. About 35% of July's feed picks were Shopify-related.
- Etsy catalogs are connected too.
- Everyone else is mostly on a waitlist for direct feed access. Some feed providers, including Salesforce and Stripe, connect through the Agentic Commerce Protocol, so check whether yours already does before you join the waitlist. OpenAI reportedly plans a self-serve feed platform later this year.
What to do this month
- Look at your catalog the way a machine does. Export your products and read the raw fields, not the product page. Is the metal stated in its own field? The karat? Solid or plated? Stone type, and lab-grown or natural? Size and length?
- Rename products in shopper words. "Thick gold hoop earrings, 14k solid gold, 20mm" says more to a matching system than "The Luna" or an internal style code.
- Answer the questions people actually ask in the description. Will it tarnish? Is it safe for sensitive ears? Can I wear it in the shower? Profound's own advice is the same: fill in the attributes and answer fit, compatibility and material questions in plain words.
- Track it over time, not once. AI answers change from day to day. One check tells you little. The same questions asked every few weeks tell you whether your changes worked.
What we do not know yet
This is one vendor's data, based on prompts it runs, not on real shoppers. OpenAI has not confirmed the July 10 change or tied it to a model release. OpenAI's stated position is that product results are chosen independently by ChatGPT and are not ads. So treat the exact numbers with care. The direction, though, is hard to argue with: for shopping questions, structured product data now matters more than it did in June.
For years the advice to jewelry brands was to make the product page beautiful. That is still true for the shopper who lands on it. But the shopper who asks ChatGPT first may never see that page unless your catalog earns the recommendation.
Does AI name your jewelry?
Tell us your brand or your store. We will show you what ChatGPT and Claude say when a shopper asks where to buy what you sell, and who they name instead of you.
Talk to us