ChatGPT Shopping now runs on product feeds. Which part of the answer does a feed win?

KEY TAKEAWAYS

A ChatGPT Shopping answer has two layers. Which product is named, and whose offer carries the link. Feeds mostly decide the second.
The 65% feed share is a share of tracked offers. It does not mean brands without a feed are unseen 65% of the time.
If your product is named but the link goes to a retailer, fix the feed. If your product is not named, a feed alone will not fix it.
Measure both layers separately, on purchase-ready prompts, and re-check monthly. Carousels reshuffle heavily between runs.

On 10 July 2026 the source behind ChatGPT's shopping results changed in a day. Merchant feeds went from a minority of tracked offers to most of them. The coverage since then has treated this as one switch: have a feed and you are visible, lack one and you are not. The data supports something narrower and more useful. A feed mostly decides who sells a product in the answer. It has much less to do with which products get named.

What changed on 10 July?

Profound, an AI visibility vendor, labels every product offer it sees in ChatGPT Shopping by where it came from. Either the offer was found through web search and a product page, or it came from a merchant feed connected to OpenAI.

In June, product pages still supplied 88.29% of offer instances in its data. Feed share had grown from 4.3% to about 20% since late 2025.
On 10 July, the feed share of tracked recommendations rose from 8.26% to 61.54%. That daily series drew on a 50% sample of 1,757,723 prompt runs in July.
By 3 September, feeds accounted for about 65% of tracked recommendations, according to Search Engine Journal's report of the study.

The field narrowed at the same time. Comparing 7 to 9 July with 10 to 12 July, the top 10 merchants' share of references rose from 22.5% to 41.8%. Unique merchants fell from 13,524 to 10,607. Of 687 Profound customers, 450 lost at least a third of their Shopping visibility and 67 gained that much.

Profound credits OpenAI's GPT-5.6 release on 9 July. OpenAI has not confirmed this, and Search Engine Journal found no shopping change in OpenAI's release notes for 9 or 10 July. The cause is unknown. The new state is what matters.

Why does the feed debate keep talking past itself?

Because the studies measure different parts of the answer.

A ChatGPT Shopping answer has two layers. The product layer is which products appear in the carousel. The offer layer is which seller's link sits under each product, and in what order. OpenAI's own help page describes them separately. Products are chosen using the query, context, and factors including price, reviews and ease of use. Merchants are ranked on availability, price, quality, and whether they are the maker or primary seller.

Profound's data classifies offers. It shows feeds taking over the offer layer. In its June data, about 99.9% of feed-sourced citations appeared as the first offer.

The opposing evidence measures the product layer. cloro collected 3,312 answers to product prompts across six engines, ChatGPT among them, and counted the domains each answer cited. YouTube and Reddit were each cited in 19% of answers and RTINGS in 16%. Brand-owned sites barely appeared. That points to third-party reviews and community content shaping which products get named. BotRank, another vendor, argues the same way: a feed buys eligibility, not placement.

Both sides are right about their own layer. One detail settles where reviews live. OpenAI's feed spec says raw review entries and question-and-answer lists "are not part of this discovery contract". The feed can carry a review count and an average star rating. The review summaries ChatGPT writes come from public websites, according to its help page.

	Product layer	Offer layer
The question	Is your product in the carousel?	Is your site the first link under it?
What OpenAI names	Query, context, price, reviews, ease of use	Availability, price, quality, maker or primary seller
Main evidence	cloro cited domains, OpenAI help page	Profound offer sourcing, July shift
What moves it	Third-party reviews, editorial and video coverage, product page quality	Feed access, complete fields, price and stock accuracy
Does a feed fix it?	Partly, through rating fields only	Yes, most directly

Does the 65% mean brands without a feed are invisible?

No. ALM Corp's summary of the study says brands without a feed are unseen for about 65% of recommendations. That reads a share of offers as a loss of reach.

The 65% is the share of tracked recommendations sourced from feeds. A brand without its own feed can still be named if a retailer that stocks it has one. Profound's March data on 22.5 million buy offers put Walmart in the first offer slot 8.78% of the time and Target 4.93%. A brand sold through those stores can sit in the carousel with no feed of its own. What it loses is the buy link.

That loss is still real. In a study commissioned by ReFiBuy and run by Clickstream Solutions, 40 US shoppers completed 224 tasks. Across the 104 ChatGPT tasks in the headline result, 84% of final choices came from product cards, and 76% of offer choices went to the first offer card. Kevin Indig's Growth Memo study of 48 shoppers in Google's AI Mode found the same pull: 74% chose the first-ranked item. Neither study randomised position or tracked purchases. Both point one way: the top slot takes most of the choice.

How do you tell which layer you are losing?

Read your own carousels. The symptom tells you the layer.

What you see	Layer	First move
Your product appears, a retailer holds the first offer	Offer	Get a feed in. Check price and stock match your site.
Your product appears, your site is a lower offer	Offer	Check the fields that differ from the first offer: price, availability, brand and seller name.
Your product rarely appears for unbranded prompts	Product	Earn reviews and coverage in the sources the answers cite. A feed will not do this alone.
Your product appears only when the prompt names your brand	Product	Same as above. Brand prompts trigger fewer carousels and crowd one brand in.
You are on Shopify or Etsy and still absent	Product, or a data gap	Check Shopify's Agentic Storefronts listing insights before assuming a feed problem.
You sell outside the US	Neither yet	Shopping is live for US users only.

The prompt matters as much as the reading. AEOsome's study of 8,520 fashion queries found carousels on 66.6% of unbranded queries against 21.1% of branded ones. Queries with five product attributes triggered carousels 87.3% of the time. Test the attribute-rich, unbranded prompts a buyer would type.

A worked example (hypothetical)

A cookware brand sells on its own site, built on a platform other than Shopify, and through Walmart and Target. It runs 20 purchase-ready prompts five times each, 100 answers in all.

Its pans appear in a carousel in 40 of the 100 answers.
In those 40, Walmart holds the first offer 22 times, Target 14 times, and the brand's own site 4 times.

The product layer is working: the brand is named in 40% of answers. The offer layer is not. Using the ReFiBuy figure as a rough guide, about three in four choices from those carousels go to whoever holds the first offer. Here that is a retailer 36 times out of 40. The brand still sells, but at wholesale margin, and the customer record goes to the retailer. OpenAI's maker or primary seller factor should favour the brand's own offer once it has a feed. Until then, it has nothing in the offer layer to rank.

Now change one number. If the pans appeared in 5 of 100 answers, a feed would mostly decide who sells those 5. The work would be on the product layer first.

What should you do this month?

Pick 20 prompts. Purchase-ready, unbranded, with real attributes such as size, budget and use. Include the comparisons a buyer would type.
Run each prompt several times. Profound's March data found about 95% of product titles appeared in fewer than 30% of runs of the same prompt. Single readings mislead. How to measure AI visibility when the answers keep changing covers sample sizes.
Record two things per answer. Is your product in the carousel? Who holds the first offer under it?
Sort the misses by layer using the table above.
Fix the offer layer through a feed. Shopify stores are included through Shopify Catalog, and Etsy catalogs are already connected. Everyone else applies on OpenAI's merchant page and joins a waitlist. The spec asks for a full catalog snapshot at least daily, by SFTP, with Parquet preferred. Brand and seller name are required. GTIN, review count and star rating are optional and worth filling. Products dropped from a snapshot stay for up to 14 days.
Check consistency. Price, stock and variants in the feed should match the product page. Profound's June data showed product-page offers carried availability and delivery details far more often than feed offers. A feed that drops them leaves the shopper less to go on.
Fix the product layer outside the feed. Product pages still feed the web side, so keep them crawlable and complete. Our readiness audit for ChatGPT's domain-scoped searches covers your own pages. The larger lever is review and editorial coverage in the sites the answers already cite.

Re-run the same prompts monthly and compare trends.

Where this fits in digital authority

The offer layer is plumbing. Once a feed is in and accurate, it holds. The product layer is earned, and it maps onto the press coverage and AI visibility pillars of digital authority. The July shift made the plumbing urgent for merchants without Shopify or Etsy. It did not change what makes ChatGPT name a product in the first place.

If the gap is coverage in the review and editorial sources your category's answers cite, that is what the placement marketplace is for. Questions about this piece can go to esh@ethosengine.io.

Questions

Do I need a product feed to appear in ChatGPT Shopping?

Not to be named. Your product can appear through a retailer that has a feed. You need your own feed, or Shopify or Etsy, to compete for the buy link. Tracked data from Profound shows feed-sourced offers rising from 8.26% to 61.54% of recommendations on 10 July 2026.

How do merchants who are not on Shopify or Etsy get a feed into ChatGPT?

They apply through OpenAI's merchant page. OpenAI says applicants join a waitlist and a self-service portal is planned for later this year. The spec asks for a full catalog snapshot, refreshed at least daily, sent by SFTP. Shopping is live for US users only.

Does a product feed improve where ChatGPT ranks my product?

OpenAI says optional fields enrich relevance and trust, but publishes no weights. Its help page ranks merchants by availability, price, quality and whether they are the maker or primary seller. Product choice draws on price, reviews and other factors. Raw reviews are not part of the feed.

How much does the first product card matter?

A lot, on the evidence so far. In a 40 person study by ReFiBuy and Clickstream Solutions, 84% of ChatGPT choices came from product cards and 76% of offer choices went to the first offer. The study observed choices rather than purchases and did not randomise position.

Sources

Profound, ChatGPT 5.6 has transformed Shopping mode tryprofound.com
Profound, What ChatGPT actually looks for and how to get your products cited higher in AI Shopping tryprofound.com
Profound, Which retailers does ChatGPT actually send shoppers to? tryprofound.com
Search Engine Journal, ChatGPT Shopping Results Lean Hard On Product Feeds searchenginejournal.com
OpenAI Help Center, Shopping with ChatGPT Search help.openai.com
OpenAI for Developers, Product Feed Spec developers.openai.com
OpenAI for Developers, Commerce documentation (full text) developers.openai.com
ChatGPT, Power product discovery in ChatGPT chatgpt.com
Shopify, Agentic Storefronts shopify.com
Shopify Developers, A new home for Agentic Storefronts in your admin shopify.dev
Newswire, In ChatGPT, 84% of Buying Decisions Came From Product Cards, ReFiBuy Study Finds newswire.com
Growth Memo, How consumers navigate high-stakes purchases in AI Mode growth-memo.com
cloro, AI Shopping: What ChatGPT Recommends (Data Study) cloro.dev
AEOsome, 8,520 ChatGPT Shopping Queries: How to Get Recommended aeosome.com
BotRank, ChatGPT shopping now favors product feeds over the open web botrank.ai
ALM Corp, ChatGPT Shopping Now Pulls Most Product Picks From Merchant Feeds almcorp.com

Esh

Founder, Ethos Engine. Builds the Digital Authority Score and runs the placement marketplace. esh@ethosengine.io

NEXT STEP

See where your brand stands.

Your Digital Authority Score, who AI recommends for your target questions, and the placements that close each gap. Your first report is free.

Get your free report →