Most of the attention in October 2026 went to how ChatGPT shopping answers look. The more consequential change happened three months earlier and had no visible interface at all: ChatGPT largely stopped sourcing its product recommendations from the web and started sourcing them from merchant catalogues.
If your products are not in a catalogue ChatGPT can read, the quality of your product page is increasingly beside the point. You are competing for a slot that is no longer allocated by crawling.
Key takeaways
- On 10 July 2026 — the day after ChatGPT 5.6 shipped — the share of shopping recommendations sourced from merchant catalogues jumped from 8.26% to 61.54% (Profound, 687 customers).
- Concentration rose with it: the top 10 merchants went from 22.5% to 41.8% of recommendation share, and the number of unique merchants reaching recommendations fell from 13,524 to 10,607 — a drop of over 20%.
- Shopify accounts for roughly 35% of all catalogue retrieval, which makes Shopify Catalog the lowest-effort route in for brands without feed infrastructure.
- You can detect feed-sourced traffic yourself: links originating from a catalogue carry
utm_medium=feed. In our 21 October test, 17 of 97 buy links did. - A feed gets you considered. It does not get you cited — those are separate mechanisms, and the second one decides whether the buy button points at your store.
The switch, in one chart
Profound ran this across 687 customers and between 280,000 and 1.76 million prompt runs. The discontinuity is sharp: ChatGPT 5.6 shipped on 9 July 2026, and the next day the catalogue share of shopping recommendations went from single digits to roughly three in five.
Read the right-hand box carefully, because it is the part that affects you. The channel switch did not just change where recommendations come from — it shrank the field. Twenty percent fewer merchants make it into recommendations at all, and the top ten take nearly twice the share they used to. In Profound's sample, 450 customers saw their shopping visibility fall by a third or more; only 67 gained by a comparable margin.
That is the signature of a catalogue-driven system. Crawling is permissive — anyone with a page is in the index. A catalogue is a gate: you are in it or you are not.
Who is on the other side of that gate
Evertune's study of 11 categories and 21,000 shopping prompts predates the October revamp, so do not read these as current absolutes. What it establishes is structural, and it has not changed: each category surfaced more than 2,000 retailers, yet the top 5–10 took 18%–47% of all buy links. Roughly 60% of the time, the first buy link also had the lowest price.
Put the two studies together and the picture is consistent with what we measured by hand in October: of 97 buy links, 51 opened a retailer product page and 20 opened a brand product page. Retailers have had product feeds plumbed into every channel for fifteen years. When a system switches from crawling to catalogues, that infrastructure advantage converts directly into recommendation share.
How to tell whether a feed is already working for you
You do not need a vendor to answer this. You need your own analytics.
Signal | Where to look | What it means |
|---|---|---|
| Landing page URL / GA4 medium dimension | The buy link came from a merchant catalogue, not a crawled page |
| Landing page URL | Same origin; we observed it on a catalogue-sourced buy button |
| GA4 source + medium | The placement came from your page, not your catalogue |
Traffic landing on the wrong store in your group | Destination hostname | Catalogue data is choosing a storefront you did not intend |
Split those in reporting and keep them split. Feed-sourced and page-sourced visits respond to completely different work, and averaging them hides which one is moving. The GA4 setup is covered in utm_source=chatgpt.com explained and how to track AI-referred traffic in GA4.
What to do, in order
1. Confirm you are actually in a catalogue. The most common failure is assuming you are. Check how your product data reaches OpenAI today, and count the SKUs that arrive against the SKUs you sell.
2. If you have no feed infrastructure, use Shopify. Shopify accounts for about 35% of all catalogue retrieval, product data flows through automatically, and it is the cheapest path in for a brand without a feed team. Shopify has told merchants that products can be discovered in ChatGPT conversations while the transaction completes in the merchant's own store, through Shopify checkout and payments. OpenAI, separately, has indicated that Instant Checkout is moving toward Apps. The practical reading for now: ChatGPT is a discovery and referral surface, not a checkout.
3. Fix the data before you expand it. The fields that decide whether a product can be matched to a question are the mundane ones — accurate price, honest stock status, complete attributes, and the use-case language a shopper would actually type. In our testing, the search terms ChatGPT generated routinely carried audience qualifiers such as "side sleepers". If your product data has no vocabulary for who a product is for, it cannot be matched on that basis.
4. Reconcile multi-store groups. We watched two Soundcore earbud models get routed across four domains in the same corporate group, including one button that landed on an Anker SOLIX store carrying utm_campaign=openai_catalog. If you run several storefronts, decide which one should receive ChatGPT traffic, align price and stock across all of them, and submit accordingly — otherwise the catalogue decides for you.
5. Monitor coverage as a standing metric. Profound's sample showed 450 of 687 customers losing a third or more of their shopping visibility through the channel switch. Catalogue gaps were a significant part of that. Coverage is not a one-off project.
What a feed does not do
This is the part that gets skipped, and it is why feed-only strategies plateau.
A catalogue determines whether your product can be considered. It does not determine whether your page is cited — and citation is what decides where the buy button points. When a brand's product page appears among an answer's sources, the button points at the brand's own store 34% of the time; when it does not, 6%.
So the two tracks are:
- Catalogue work → eligibility for the shelf. Owned by commerce and data teams.
- Source work → whether the shelf label has your name on it. Owned by content and PR teams, and covered in how AI search picks sources and how to get cited by ChatGPT.
Doing one without the other produces a predictable failure mode: your products appear, and every buy link goes to Walmart.
FAQ
What is a ChatGPT product feed? A merchant product catalogue submitted to OpenAI so that product data — price, availability, attributes — can be used directly in shopping recommendations instead of being crawled from a web page. Links sourced this way carry utm_medium=feed.
How do I submit a product feed to ChatGPT? Either through a commerce platform that forwards catalogue data on your behalf — Shopify is the most widely used route, at roughly 35% of catalogue retrieval — or through OpenAI's own merchant channels. Check OpenAI's current merchant documentation for the live specification before building anything, as the requirements have changed repeatedly through 2026.
Does having a product feed guarantee I will be recommended? No. A feed makes you eligible. Which products get recommended still depends on matching, price competitiveness and the sources the answer draws on. Roughly 60% of the time the first buy link also carries the lowest price.
Is Shopify the only way in? No, but it is the lowest-effort way for brands without feed infrastructure, and it carries about a third of all catalogue retrieval.
Can I use my Google Merchant Center feed? The underlying product data is the same material, and most feed-management tooling can target multiple destinations. Confirm field requirements against OpenAI's current merchant documentation rather than assuming parity with Google's specification.
Will ChatGPT handle the checkout? Not in the current arrangement. Shopify's guidance to merchants is that discovery happens in the conversation and the transaction completes in the merchant's own store. See agentic commerce and the Universal Commerce Protocol for where this is heading.
Sources
Catalogue share, merchant concentration and unique-merchant counts: Profound, "ChatGPT 5.6 has transformed Shopping mode", 3 September 2026; 687 customers, 280k–1.76M prompt runs. Retailer concentration and the price-leadership figure: Evertune Research, 11 categories and 21,000 shopping prompts, February–March 2026 — predates the October revamp and is used here for direction, not absolutes.
The UTM parameter counts and the multi-store routing observation are from GEOly Research: 28 shopping prompts run in a signed-out Chrome window on 21 October 2026, with every buy link clicked and logged. Full method in how ChatGPT shopping results work after GPT-6.
