"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.
