A polished product page may no longer be enough to earn a place in ChatGPT’s shopping results. New tracking data suggests the platform is sourcing a growing share of its recommendations from connected product catalogues, where missing prices, incomplete variants and stale availability can keep an item out of consideration.
The finding does not establish a universal ranking rule. It does, however, show why product data has become part of the visibility conversation for online retailers.
The 65% Figure Does Not Cover All ChatGPT Shopping
AI visibility platform Profound found that feed-integrated sources accounted for roughly 65% of the product recommendations observed in its ChatGPT Shopping tracking by early September.
That percentage comes with boundaries.
For its broader retrieval analysis, Profound examined a 10% sample of 97,725 prompt runs recorded between July 1 and August 24. Each prompt was counted once per month. Feed-integrated retrieval started the year well behind web-based retrieval, grew through the spring and overtook web search in August, according to the company’s published methodology.
A separate daily analysis sampled half of the company’s July prompt runs, covering 1,757,723 runs in total. Within that dataset, the feed-integrated share climbed from 8.26% to 61.54% on July 10.
These were prompts tracked for Profound customers, not a representative sample of every product query submitted to ChatGPT. The company describes its findings as an observational snapshot covering July and August 2026.
The distinction matters. A 65% share in monitored recommendations does not prove that 65% of all shopping responses use feeds, nor does it establish that a retailer without a direct connection cannot appear through web retrieval.
A Retrieval Shift Left Fewer Merchants in View
The change in retrieval source coincided with sharp movement in merchant visibility.
Profound began with 687 customers that tracked Shopping-triggering prompts every day and received at least one product card reference between July 7 and July 12. Comparing the three days before July 10 with the three days after, 450 customers recorded visibility declines of at least 33%. Another 67 posted gains of the same size.
The pool of merchants appearing in the results also narrowed. The top 10 merchants’ combined share increased from 22.5% to 41.8%, while the number of unique merchants referenced fell from 13,524 to 10,607.
Profound’s model linked changes in web retrieval and feed-integrated retrieval with 83% of the observed variation among the 517 merchants that experienced substantial swings. That is a strong association inside the dataset. It is not evidence that feed access alone caused every gain or loss.
The company attributed the timing to OpenAI’s GPT-5.6 release on July 9. OpenAI has not publicly confirmed that the model rollout included a related change to Shopping retrieval, so the connection remains Profound’s interpretation of the timing and data.
Still, the concentration effect gives retailers a sharper way to think about AI search visibility. The source from which ChatGPT retrieves a product can influence which pool of merchants is available before relevance, price or other factors are considered.
OpenAI Still Uses More Than Merchant Feeds
OpenAI’s own documentation describes a broader selection system than feed access alone.
The company says ChatGPT considers structured metadata from first- and third-party providers, other third-party content, previously generated model responses, safety standards and product policies. It may weigh price, reviews, availability or ease of use differently depending on the shopper’s request.
“Product results are selected independently by ChatGPT and are not ads, nor influenced by any OpenAI partnerships,” the company states in its Shopping documentation.
Relevance remains the entry point. Not every eligible product is displayed, and a feed does not guarantee placement.
Shopify merchants already have their catalogue data integrated through Shopify Catalog, according to OpenAI, without requiring each store to establish a separate connection. Other merchants can apply to provide a direct feed. ChatGPT can also retrieve information from the web, provided a site permits OAI-SearchBot to crawl it.
That leaves more than one route into ChatGPT Shopping. Profound’s findings suggest those routes did not receive equal use across its tracked prompts during the study period.
Product Feed Quality Is Becoming an Ecommerce SEO Issue
A feed connection is only the first layer. The catalogue still has to describe the product accurately enough for ChatGPT to match it with a detailed request.
OpenAI’s product feed documentation identifies pricing, availability and seller context as core structured information. It also tells merchants to keep inventory and price changes current, with feeds delivered through supported methods such as SFTP, hosted URLs or file uploads.
For marketing teams, the practical work starts with a product data readiness review. That review should confirm which catalogue connection is active, whether the full sellable range is included and whether product variants resolve to the correct landing pages. Titles, descriptions, images, identifiers, prices, stock status and seller details should agree across the feed and the website.
Catalogue gaps carry a different risk from weak sales copy. A missing size or colour is not merely less persuasive. It can prevent the relevant variant from entering the candidate set for a specific query. An old price or incorrect availability signal can also create friction after the recommendation has already appeared.
This brings ecommerce SEO closer to feed operations. Merchandising, organic search and development teams may be managing the same product facts through different systems, yet ChatGPT can surface whichever version reaches its retrieval layer.
Organic Recommendations Are Not Product Feed Ads
OpenAI now supports feeds in both commerce discovery and advertising, but the two uses should not be combined in reporting.
Organic Shopping results are independently selected product recommendations. Feed-based ads are paid placements created through Ads Manager, where advertisers upload a catalogue, define eligible products and run campaigns. OpenAI explicitly says feeds uploaded for advertising are not used to place products in organic ChatGPT conversations.
Retailers already using product feeds for paid campaigns should therefore verify whether they also have an eligible commerce connection. A successful advertising upload does not confirm that the same catalogue is available to organic Shopping retrieval.
The reverse is also true. Shopify’s catalogue integration may make a merchant’s products available to ChatGPT Shopping, but it does not guarantee a recommendation, a particular merchant ranking or advertising delivery.
Profound’s observation window ended on August 24, and OpenAI has not published a platform-wide breakdown of feed versus web retrieval. For now, the 65% figure is evidence of a substantial shift within one large tracking dataset, while the exact mix across live shopper activity remains undisclosed.
