"How do we get listed in ChatGPT shopping?" is the question we are asked most often, and it usually arrives attached to the wrong assumption — that there is a submission form, or an SEO trick, or an ad product that buys the slot. There is none of those. There is a dependency chain, and most brands are broken at a layer well below the one they are working on.
This is the playbook, built from a hands-on test of 28 shopping prompts and 97 buy links on 21 October 2026, plus three external datasets. It is ordered deliberately: each layer is worthless until the one beneath it holds.
Key takeaways
- Five layers, in order: product page → catalogue feed → source presence → attribution → paid alignment. Break any one and everything upstream stops converting.
- The highest-return single action is source presence. A brand page cited in the answer gets the buy button 34% of the time; one that is not gets 6%.
- Catalogue feeds now supply 61.54% of shopping recommendations, up from 8.26% in July 2026. No feed, no eligibility.
- ChatGPT rewrites the shopper's question before searching. The terms it generated in our test added a year, a content type and an audience qualifier — "best sleep earbuds stay in ears overnight side sleepers 2026 reviews".
- Source display has narrowed to one site name per paragraph. Being a source is no longer the goal; being the first source is.
- You cannot buy a recommendation. Ads and recommendations are different systems on the same screen.
L1 — Make the product page machine-readable
This is unglamorous and it is where most failures actually live.
- Put specifications, price, stock status and use case in the server-rendered HTML. Not behind a client-side script. Test it by disabling JavaScript and viewing source, or by fetching the page with
curland reading what comes back. If the price only appears after hydration, assume it does not exist. - Ship product structured data —
Product,Offer,AggregateRating— with price and availability fields that match what the page actually says. A schema that disagrees with the visible page is worse than no schema. - Write the "who is this for" copy. In our testing the search terms ChatGPT generated carried audience qualifiers: side sleepers, college students. If your page has no language describing who a product suits and in what situation, it cannot be matched on that basis. This is the cheapest high-leverage copy change on the list.
L2 — Get your products into a catalogue
A crawlable page is no longer sufficient. Since 10 July 2026, the majority of shopping recommendations have come from merchant catalogues rather than from web pages.
- Confirm you are actually in one, and reconcile the SKU count in the catalogue against the SKUs you sell. "We assumed our platform handled it" is the most common gap we see.
- If you have no feed infrastructure, route through Shopify, which accounts for roughly 35% of all catalogue retrieval.
- Align price and stock across every storefront in your group, and decide which store should receive ChatGPT traffic. We watched one product's buy button land on a sibling brand's store, carrying
utm_campaign=openai_catalog— the catalogue picked, not the brand. - Treat coverage as a standing metric. In Profound's sample, 450 of 687 customers lost a third or more of their shopping visibility through the channel switch.
Detail: the ChatGPT product feed.
L3 — Get your pages into the answer's sources
This is the layer with the best return on effort, and the one most commerce teams do not own.
When a brand's product page is among the sources cited in an answer, the buy button points at the brand's own store 34% of the time. When it is not, 6%. On a stricter definition the split is 39% versus 7% — same direction. There is no landing-page optimisation that recovers a five-fold gap at the top of the funnel.
What to do with that:
- Cover the content types the model actually retrieves: reviews, comparisons, ranked lists and question-and-answer pages. Not brochure copy.
- Cover the query shape the model actually generates, which in our observation was year + category + audience or situation + "reviews". Optimise for the rewritten query, not the shopper's spoken question. The method for finding those is in how to find grounding queries and the mechanism is in query fan-out.
- Aim for first position among sources, not inclusion. Source display has collapsed to one visible site name per paragraph, with the rest behind a tap — roughly half of all sources are now invisible by default. See why ChatGPT now shows one source per paragraph.
- Make the recommended product reachable from wherever the link lands. Nine of our 97 buy links opened a brand homepage and eight opened a category page. If the shopper lands one or two clicks away from the product they were just shown, you have paid for the visibility and donated the conversion.
Mechanics: how AI search picks sources and how to get cited by ChatGPT.
L4 — Rebuild your attribution before you judge any of this
You cannot manage what you are measuring wrong, and in October 2026 almost everyone was measuring it wrong for a week.
- Annotate 8 October and 15 October on every traffic report and exclude the span between them. During that window ChatGPT's links dropped
utm_sourceand the visits were logged as direct traffic. One observer measured ChatGPT-attributed visits falling ~31,000/day while direct rose ~26,000/day across the same pages. - Segment GA4 on
source = chatgpt.com, and splitmedium = feedout separately. Feed-sourced and page-sourced visits respond to different work. - Do not use referrer. Button-type entry points are opened with
noreferrerand send no referring domain. - Track being recommended, not just being clicked. Clicks record the bottom of the funnel only. Whether your brand appears at all — your Share of Model across ChatGPT, Google AI Mode, Gemini, Google AI Overviews and Microsoft Copilot — is the metric that tells you whether L1 to L3 worked.
Detail: utm_source=chatgpt.com explained and tracking AI traffic in GA4.
L5 — Plan paid and organic on one board
ChatGPT Ads reached 89 countries on 21 October 2026 and shows only on the Free and Go tiers — which is also where the bulk of casual, signed-out shopping questions come from. Organic visibility and paid reach now compete for the same attention in the same interface for the first time.
But advertising does not buy a recommendation. Product recommendations and buy links are still allocated by catalogue data and source citation. Plan them separately: paid solves reach, catalogue and content solve being recommended. And note that ad targeting works on conversation-topic hints rather than keywords, so creative and landing pages should be organised around real conversational situations, not a keyword tree inherited from search.
Detail: ChatGPT Ads now reach 89 countries.
The self-check
Run this before you commission anything.
# | Check | How to verify | Priority |
|---|---|---|---|
1 | Specs, price and stock readable in server HTML | Disable JavaScript and view source, or | High |
2 | Product feed submitted and covering your lead SKUs | Reconcile catalogue SKU count against SKUs on sale | High |
3 | Price and stock consistent across group storefronts | Compare quotes store by store, SKU by SKU | High |
4 | Your pages appear in the sources for your target questions | Run the question, open the source marker, log every entry | High |
5 | GA4 splits ChatGPT traffic by source and medium | Check the acquisition report's custom dimensions | Medium |
6 | 8–15 Oct annotated and excluded from comparisons | Check report annotations and period-over-period logic | Medium |
7 | Recommended products reachable in one step from the landing page | Walk the path from each buy link's destination | Medium |
8 | Target question tested both signed out and signed in | Ask plainly, ask for official links, ask for product cards | Medium |
FAQ
Is there a way to submit my brand to ChatGPT shopping? There is no brand submission form. There is a product catalogue route — see the ChatGPT product feed — and there is the editorial route of becoming a cited source. Both matter; neither is a form.
How long does it take to appear? We have no reliable figure and will not invent one. Catalogue changes propagate on the platform's schedule; source presence follows the pace at which the content ecosystem around your category updates. Measure it rather than estimating it.
Can I pay to be listed in ChatGPT shopping? No. ChatGPT Ads buys ad placements on the Free and Go tiers. Product recommendations and buy links are allocated separately.
We are already ranked well in Google. Does that carry over? Partly. Good technical pages and strong third-party coverage help with source presence. They do nothing for the catalogue layer, which is now the majority channel for shopping recommendations.
What is the single highest-return action? Getting your product pages into the answer's sources for your priority questions. It moves buy-button ownership from 6% to 34%.
How do I know if any of this worked? Track Share of Model across the five engines, split your ChatGPT traffic by utm_medium, and re-run your priority questions on a schedule. GEOly's AI shopping monitoring is built for exactly this loop, and a GEO audit will tell you which of the five layers is currently broken.
Sources
The five-layer framework, the 97-link destination counts, the multi-store routing observation and the query-rewriting observation are GEOly Research: 28 shopping prompts in a signed-out Chrome window, 21 October 2026. Buy-button destination percentages are from WISLR (4 brands, 250 prompts, 10–13 October 2026). Catalogue share and merchant concentration are from Profound (687 customers, 280k–1.76M prompt runs, "ChatGPT 5.6 has transformed Shopping mode", 3 September 2026). The UTM outage figures are from Search Engine Roundtable, 14 October 2026, reporting Adobe's Cedric Huesler.
