ChatGPT Shopping switched to product feeds overnight. Check your feed this week
Profound's data shows ChatGPT Shopping picks moved to feed-integrated merchants on 10 July, and fewer merchants now win more. What D2C brands should check.
The short answer
On 10 July, ChatGPT Shopping started drawing most of its product picks from integrated product feeds. Profound's data shows feed-integrated retrieval jumping from 8.26% to 61.54% of picks, with the top ten merchants' share nearly doubling. If you sell online, confirm you have a feed route into ChatGPT, then audit the fields it reads.
Key takeaways
- In Profound's data from 687 customers (1 July to 24 August 2026), feed-integrated retrieval rose from 8.26% to 61.54% of ChatGPT Shopping picks on 10 July.
- Over the same change, the top ten merchants' share of picks rose from 22.5% to 41.8%, and the number of unique merchants fell from 13,524 to 10,607.
- Profound estimates that Shopify accounts for about a third of feed-integrated retrieval, the main route for smaller merchants.
- Being in a feed is now the entry ticket; price, stock, identifiers and variants decide what an engine can match.
- Our view: treat the feed as a product page that a machine reads, with an owner and a weekly check.
In this article
What changed on 10 July
ChatGPT Shopping changed where it gets products. Profound, which sells AI visibility tracking, analysed shopping prompts tracked by 687 of its customers between 1 July and 24 August 2026.1 In its daily data, the share of picks retrieved through integrated product feeds rose from 8.26% to 61.54% on 10 July.1
The field narrowed at the same time. Comparing 7–9 July with 10–12 July, the top ten merchants' share of picks rose from 22.5% to 41.8%, and the number of unique merchants fell from 13,524 to 10,607.1 Fewer sellers now win more of the recommendations.
Profound ties the change to a ChatGPT model update. Its summary is blunt: merchants that are not integrated are left out of most product recommendations in Shopping.1
ChatGPT Shopping picks, before and after 10 July 2026
How far the data goes
This is one vendor's customer base, not the whole of ChatGPT. Profound's customers choose which prompts to track, and they skew towards brands that already invest in AI visibility. Profound narrowed its comparison to 517 customers whose visibility swung by a third or more.1
That selection makes the size of the jump less certain than the direction. A move from under a tenth to over half, on a single day, is hard to explain by sampling alone. Our view: the direction is reliable enough to act on, and the exact percentages are not a benchmark for your own category.
The study also says nothing about sales. It counts which products ChatGPT picks, not what shoppers buy. For conversion evidence on US retail sites, see our note on Adobe's July data.
Why feeds, and why now
A feed gives an engine structured, current facts: price, stock, identifiers, variants and images. Crawled pages give it prose that may be stale, blocked or rendered in a browser it does not run. When a product question needs today's price in a given size, the structured source wins.
The commercial route has moved too. In March 2026 OpenAI moved Instant Checkout out of ChatGPT's product results and into merchants' own apps, with the Agentic Commerce Protocol continuing underneath.2 Discovery stays inside ChatGPT; checkout moves to the merchant. That makes the feed the main place a brand controls what ChatGPT knows about each product.
Which integration route fits your store?
Profound describes two routes.1 Large retailers integrate their own feeds directly. Smaller merchants mostly arrive through Shopify, which Profound estimates accounts for about 35% of all feed-integrated retrieval.1
- On Shopify: check that your products are eligible and that the sales channel settings expose them, then check what the feed actually contains.
- On another platform: OpenAI publishes a product feed reference with required and recommended fields.3 Build to it, or use a feed tool that already supports it.
- On a marketplace only: your listing data is the marketplace's feed, not yours, so you compete on their terms.
Five feed fields to audit this week
Profound expects the contest to move from being integrated to the quality of the fields inside the feed.1 Our view: start with the fields where a wrong value loses the pick outright. OpenAI's feed reference lists the fields an engine can read.3
- Price and sale price. The feed price must match the product page on the day. Update it when a sale starts or ends.
- Availability. An out-of-stock item that shows as in stock costs you a shopper and the engine's trust in your feed.
- Identifiers. GTINs, brand and manufacturer part numbers let the engine match your listing to the product a shopper names.
- Variants. Group sizes and colours under one product so a question about a size finds the right item.
- Title and description. Write factual descriptions with the attributes people ask about: material, dimensions, compatibility and care.
Shipping, returns and ratings come next. They answer the follow-up questions shoppers ask once a product is on the shortlist.
Then test. Ask ChatGPT the product questions your customers ask, in the markets you sell to, and note whether your items appear and with which details. Our playbook on why ChatGPT does not recommend your products covers the non-feed causes, from reviews to retailer listings.
What this means for smaller brands
Concentration is the risk. When the top ten merchants nearly double their share in a day, smaller brands lose picks they never knew they had. The fix is not expensive, but it has an owner: someone who checks the feed weekly, as they would check the product pages.
Our view: if your team can name who owns the product feed, you are ahead of many stores. If nobody can, assign it before the next model update changes the rules again.
