Does ChatGPT Recommend Products? How It Actually Picks

August 23, 2026

ChatGPT recommends products, and OpenAI documents part of how it picks them. For shopping questions it shows products it judges relevant to the shopper’s intent and context. It also weighs structured product data from first party and third party providers. The part a store can influence this month is the live web search. It needs pages that state plainly what a product is and who it is for.

By Margareta Petrovic, founder of Visibility Mesh. I am CISA and CISSP certified, and we measure how readable websites are to AI assistants and publish what we find. I last updated this article on September 27, 2026.

Key takeaways

OpenAI says an answer comes from what the model learned in training, from a live web search, or from both.
For shopping, ChatGPT weighs structured product data, the shopper’s context and its own earlier response, and OpenAI publishes no weights.
Nobody buys a place in the answer. OpenAI says ads sit below it with a sponsored label.
The gaps our scans find most often sit on a store’s own pages, and those are the cheapest to fix.

Where do the names in a ChatGPT recommendation come from?

OpenAI names two sources. Its page on accuracy says that without search, a response is based on what the model learned in training. With search, ChatGPT can cite real time web sources. In our observation, product answers usually blend the two. Well known brands can appear with no search at all, and the rest of the list comes from pages ChatGPT found and read.

For shopping questions, OpenAI’s shopping help page is more specific. It says ChatGPT considers structured metadata from first party and third party providers, such as price and description. It also considers its own response from before any new search, and OpenAI’s safety policies. It weighs what the buyer asked for, so a stated budget moves the focus to price.

The live search is the part you can influence this month. It works on pages that state plainly what a product is, who it is for, what it costs and whether it is in stock.

Why does ChatGPT not recommend my product?

OpenAI does not publish reasons for leaving a product out, so I start from what it does document. ChatGPT weighs structured product data such as price and description, what the shopper asked for and its own earlier answer. A product drops out when ChatGPT cannot read those facts on your pages, or when a rival page states them more plainly.

In our scans the gaps usually sit on the store’s own pages. Crawlers are blocked, or the content appears only after scripts run. The product page has a name, a price and two adjectives. The brand shares its name with something bigger, and nothing on the site confirms which one it is. Each of these can be fixed this month: open access and serve the content in the HTML, state the materials, the sizing, the use case and who the product suits, and keep one consistent identity across your site and your profiles. Reviews and roundups on other sites matter too, and they take longer to earn.

What decides between two similar products?

OpenAI’s own example shows the kind of tie breaker it uses. The question is about goofy dog costumes. OpenAI says ChatGPT may weigh the available options, price, reviews and ease of use, plus anything else the shopper said. When several stores sell the same product, OpenAI ranks the stores by availability, price and quality. It also looks at whether a store is the maker or primary seller.

So a close call comes down to facts ChatGPT can read and compare. A missing field, a stale price or a vague description gives it less to work with than a rival page that states all three.

What kind of product page gets picked?

One that answers a buying question without making ChatGPT guess. Say a buyer asks for a waterproof jacket for cycling in winter. The page with something to match says it is waterproof and made for cycling. It gives the temperature range, and it shows a price and stock status in a form a machine can read.

A page that says “built to last” answers none of that. Specific claims make a page quotable, and we explain the method in putting claims in fields rather than prose.

How do I check whether it recommends me today?

Ask it the question a customer would type, then see which names come back. Asking whether it knows your brand tells you very little, since the brand name is already in the question. Use a Temporary Chat so your own memory does not shape the answer, and ask more than once.

Repeat the check every month, since the answers move. AI visibility drift explains why a store named in one month can be missing in a later one even when nothing on its site changed.

What our study found on store pages

In our study of 216 Shopify storefronts (June 2026), every store was scored on five metrics. A store failed a metric when it earned less than half of its points.

Of the 216 storefronts, 41 percent failed Front Door, our metric for whether an AI crawler can read the page at all.

Of the same 216, 71 percent failed AI Readability, 72 percent failed Entity Integrity and 68 percent failed Authority Signal. Listed versus recommended product schema walks through the product fields one by one.

Our scan reads your own site, so it does not measure whether other sites mention you. You can check that by asking ChatGPT the question a buyer would ask. The median store in the study scored 46 out of 100, and everything the study measured sits on a store’s own pages, so start there.

Common questions

How does ChatGPT recommend products?

ChatGPT answers from what the model learned in training, from a live web search, or from both, according to OpenAI. For shopping questions, OpenAI says it shows products it judges relevant to the shopper’s intent and context. It weighs structured metadata from first party and third party providers, such as price and description. Its own earlier response and OpenAI’s safety policies count too. A store can influence the live search, and that needs pages that state plainly what a product is and who it is for.

How does ChatGPT decide which companies to recommend?

OpenAI has not published a list of factors for recommending companies. It documents that ChatGPT may answer from training or search the web when current information helps. It ranks search results on several factors aimed at relevance and reliability, and memory, custom instructions and approximate location can shape the answer. In our scans, a company is easier to describe accurately when its own pages say plainly what it does and for whom. Treat any published list of ranking factors as a guess.

Why does ChatGPT give different recommendations each time I ask the same question?

Each answer is generated fresh, and OpenAI documents several inputs that change between conversations. One is whether ChatGPT searched at all, and another is the queries it wrote for that search. The user’s memory, custom instructions and approximate location also vary, and so do the results themselves from one moment to the next. So one screenshot proves little. Ask the same questions several times, count how often each brand appears, and note the date, the account settings and whether search was used.

Does ChatGPT personalize answers, so different people see different brands?

It can. OpenAI says that when memory is on, ChatGPT may use saved memories when it rewrites a search query. For shopping it considers context such as memory and custom instructions, and for local results it uses an approximate location from the IP address. Two people asking the same question can therefore see different businesses. When you test your own brand, use a Temporary Chat. OpenAI says it does not use or update memory. Then repeat the question more than once.

Can I pay ChatGPT to recommend my products?

You can buy an ad, but not a place in the answer. OpenAI says ads appear below the response with a sponsored label, run on separate systems from the chat model and do not influence the answer. It also says product results in shopping are not ads. Does ChatGPT get paid to recommend products? covers the details.

Does ChatGPT only recommend big brands?

OpenAI does not say, and we have no measurement that settles it. In our observation, well known brands appear often. That fits OpenAI’s note that its models were trained on vast amounts of text from the internet. For search results, OpenAI says it ranks for relevance and reliability and does not mention size or fame. Small stores can appear when ChatGPT searches, but we have not measured how often. Test your own category before you assume the pattern.

I read the OpenAI help pages behind this article on September 26, 2026. They are Shopping with ChatGPT Search, Searching the web with ChatGPT, Does ChatGPT tell the truth?, What is ChatGPT? and Ads in ChatGPT. The study figures come from The State of AI Visibility 2026 (216 Shopify storefronts, scored June 28, 2026), and our AI legibility dataset explains how the score is produced.