Obility Editorial · · 4 min read
Shopping answers are assembled from product records
Ask ChatGPT for waterproof hiking boots under $150 and it can answer with a carousel of products, each with an image, a price, a short description and links to merchants. The brand that wrote the best buying guide on the subject is not necessarily in it. The carousel is built from product data, and the page copy most marketing teams polish is only one input to that data.
OpenAI’s help center article on shopping with ChatGPT search lists what ChatGPT considers when deciding which products to surface: structured metadata from first-party and third-party providers, such as price and product description, along with other third-party content; the responses the model generated before it looked at new search results; and OpenAI’s safety standards and product policies. The same article says a product appears when ChatGPT perceives it as relevant to the user’s intent, taking into account the query and context such as Memory and custom instructions. Two shoppers asking the same question can see different products for reasons that have nothing to do with your catalog.
ChatGPT gets product data through three routes
The first route is third-party providers. OpenAI says prices shown in ChatGPT come from third-party providers, and that the list of merchants for a product is built from merchant and product metadata supplied by those providers or directly by merchants. If you have never sent OpenAI anything, this is where your products come from, and you do not choose what those providers hold about you.
The second is Shopify. The help center states that product data for Shopify merchants is already integrated into ChatGPT through Shopify Catalog and that no additional work is required from individual merchants. For a Shopify store, the job is making sure the catalog itself is accurate.
The third is a direct product feed. OpenAI’s commerce documentation says onboarding product feeds is currently available to approved partners, who apply for access. The feed specification requires nine fields per item or variant: an item ID, title, description, product page URL, brand, seller name, image URL, availability and price. A row with a missing or unrecognised availability value is rejected, and the standard upload currently targets the United States only, with other markets enabled only after OpenAI confirms them. Several guides on this topic describe the feed as something any merchant can switch on today. The documentation does not say that.
Google’s AI Mode draws on the same product data as Shopping
Google described its AI Mode shopping experience, announced in May 2025, as Gemini combined with its Shopping Graph, which it said held more than 50 billion product listings, with more than 2 billion refreshed every hour. Your products reach AI Mode through that record, so its accuracy matters as much as the page itself.
Google’s Search Central guide to sharing ecommerce product data explains how that record is built. Product structured data is not required to appear in Search but improves Google’s understanding of price, discounts and shipping costs. A Merchant Center feed is optional for Search results, required for the Shopping tab, and gives you control over update timing, since Google does not guarantee how quickly crawling picks up changes. The same page warns that combining site data with feed data causes inconsistencies when one lags the other, such as a product that sold out on your site but still shows as available in the feed. Google’s guide to AI features lists checking that Merchant Center information is up to date among the fundamentals that apply to AI Overviews and AI Mode.
Reviews and labels are written about you, not by you
The words shoppers see next to your product are often not yours. OpenAI says ChatGPT may generate simplified titles and descriptions from information it receives, because merchants describe the same product in different ways. Labels such as “Budget-friendly” or “Most popular” are generated by ChatGPT and, in OpenAI’s own example, may mean reviewers frequently mention good value rather than that the price is lowest. Review summaries are generated from reviews on public websites, and OpenAI says it does not verify reviews or ratings.
A feed can carry a product review count and star rating, but the specification says these must describe product reviews only, not store reviews, and raw review text is not part of the discovery feed. So the framing of your product in a shopping answer depends heavily on what customers have written elsewhere. When a product has several sellers, OpenAI says merchants are ranked on factors such as availability, price, quality and whether the seller is the maker or primary seller of the item. A brand selling direct and through retailers is competing with its own resellers for the first slot.
Eligibility is the part you control
The feed specification is blunt about its limits: its search eligibility flag defaults to on, and the documentation notes that eligibility does not guarantee display. A clean feed makes your products available to be chosen. It does not make ChatGPT or AI Mode choose them over a better reviewed product at a similar price.
The work that pays off is consistency. Take your ten most important products and check that the name, variants, price, availability and description agree across your product pages, your structured data, your Merchant Center feed and, where you have one, your Shopify catalog or OpenAI feed. Update prices when a sale starts and ends; OpenAI’s specification says sale dates in the feed do not schedule price changes. Then read the review sites and marketplaces that describe those products, because that is where generated labels and summaries come from. Obility’s Shopping workflow starts from the same check: names, variants, availability and descriptions that agree across the pages you control, with no claims you cannot substantiate.
To see whether it worked, run the same shopping prompts on a schedule and record which products appear, which merchant is listed first, what price is shown and what label is attached. A product that starts appearing with the right price is evidence. A single carousel on a single day, shaped by one account’s memory and instructions, is not.