Only what the vendor documents, dated October 8, 2026, then what our weekly checks see ChatGPT actually read, and what a business does about it.
OpenAI documents three agents, and says "each setting is independent of the others":
On shopping, OpenAI's help center says product results "are selected independently by ChatGPT and are not ads, nor influenced by any OpenAI partnerships", that "ads are separate from product results", that merchants on Shopify "are already connected through Shopify Catalog", that other merchants "can apply for direct feed access", and that merchants in a product's list are ordered by "availability, price, quality, and whether they make or primarily sell the item".
Sources: developers.openai.com/docs/bots and help.openai.com (Shopping with ChatGPT), read October 8, 2026. OpenAI does not document how it ranks the pages it reads; the rows below are what our checks observe.
Three things follow for a business's pages. First, the search crawler must be allowed, and most sites allow it without knowing; a blanket block on "AI bots" that catches OAI-SearchBot removes you from ChatGPT search answers while leaving the training question untouched. Second, ChatGPT with search on reads the live web at answer time, so a page published this week can be in next week's answer, and a page that states a fact plainly (a price, a warranty term, a city) is the page that gets quoted. Third, product cards come from catalog feeds and retailer listings, so a store whose catalog is not syndicated can be named in the text and absent from the cards beside it.
Rows from scans and checks run in September and October 2026, with the companies unnamed where they have not agreed to be named.
| Category | What happened |
|---|---|
| Software, dispatch | Asked for the best software for a three-van operator, ChatGPT put a vendor first and cited five of that vendor's own pages, including a 'best software' roundup the vendor had published two weeks earlier. Days earlier the same question had a different first pick, set by a different page. |
| Leather goods | Asked for the best full-grain brands, ChatGPT ranked seven and wrote the answer entirely from the brands' own pages, including one brand's own 'best handmade wallets 2026' post. The client, with 32,000 on-site reviews, was absent. |
| Hospitality | Asked whether a resort group was a good company to book with, ChatGPT read seven of the group's own pages and the OTAs, gave a property-by-property verdict, and concluded "I'd book the property, not the company". On the category questions it never opened the group's site. |
| Architecture | Asked for a design-driven custom-home architect in a city, ChatGPT named the client first with a map card and read its FAQ page; asked about two neighbouring regions, it built the shortlists from Houzz directories and rivals' project pages and never named the client. |
| Supplements | Asked whether a brand was legit, ChatGPT confirmed the company, the BBB rating and the certification page, then wrote that it "could not establish that every report came from a named, accredited outside laboratory": a caveat built from the brand's own lab-results page not naming the lab. |
Every week the 50 questions run on ChatGPT with search on, the way buyers use it, and every answer is archived with the pages it read and the product cards it showed. A person grades each flagged answer. The work follows the sources: the page that answers the question ChatGPT is building from a rival's roundup, the correction to the record it quoted, the inclusion request to the list it read, and, for stores, the catalog feed check. The next check shows whether ChatGPT read the page and whether the answer changed.
Not according to OpenAI: the search crawler is OAI-SearchBot, and the vendor says a site can allow it "in order to appear in search results while disallowing GPTBot". Blocking OAI-SearchBot is what removes a site from ChatGPT search answers.
Because it reads the live web at answer time and the pages it reaches change: a rival publishes a roundup, a directory updates, a search returns a different page first. That is why Rightcited checks weekly and only changes a verdict after two consecutive checks agree.
OpenAI says product results "are selected independently by ChatGPT and are not ads". What a merchant can do is make sure its catalog reaches the feeds the cards are built from (Shopify Catalog, or a direct feed) and that price, availability and maker are right on the product page.
Or start smaller: the AI Page Check runs one page through all four assistants for $2.99.