How ChatGPT's ad system reads a shopper, and what it means for your store

TL;DR: On October 21, 2026, OpenAI described how ChatGPT chooses ads. A feed or search ad competes for a slot each time someone searches or scrolls. A ChatGPT ad waits for the point in a conversation where a need is clear, and showing nothing is a normal result. The auction underneath works like a conventional one. What changed is how the system reads the conversation: it follows the whole conversation, can split it into several needs, and runs apart from ChatGPT's answer, so an ad and a place in the answer are two different jobs. Below: the stage-by-stage comparison, and what to do about it this week.

---

What did OpenAI publish?

In How we built ChatGPT's novel ad stack for user trust and relevance, OpenAI's engineering team explained how ads in ChatGPT are selected. The system is built around "a custom, post-trained frontier LLM that makes tool calls for retrieval (identifying the top options), ranking (predicting user engagement), relevance (evaluating the usefulness of the ad), and running auctions for final ad selection and pricing." OpenAI calls this LLM the orchestrator.

OpenAI also gave the scale: ChatGPT Ads "reached $1 billion in annualized revenue run rate less than 200 days after launch," and "more than 100,000 advertisers have run ads on ChatGPT since launch." Those are OpenAI's figures.

How is a ChatGPT ad different from a feed or search ad?

OpenAI describes a conventional system this way: a search query or page view arrives as a request, and the system "retrieves potential ad candidates, ranks them, runs an auction to appropriately price the opportunity, and shows the resulting ad to the user." Stage by stage, against what OpenAI says ChatGPT does, with what a merchant can do at each stage:

Stage	Feed and search ads	ChatGPT ads, per OpenAI	What a merchant can do (our view)
What starts it	"a well-defined event": a query or a page view	A conversation where intent "shifts over multiple turns". The orchestrator first writes "one or more queries representing the relevant commercial intent"	Answer the questions a shopper asks on the way to buying, not only your product names. Citable lists questions each of your buyer personas is likely to ask
Memory	"request-response": one ad request per query or page view	"Stateful ads sessions": remembered "across turns of a conversation and across multiple conversations"	Cover the whole path on your site: the goal, the comparison, then the product
Finding candidates	Broad retrieval	Still broad: "a deliberately broad pool of ad options"	If you run product-feed ads, keep the feed complete and your reviews visible
Filtering	Targeting, budget, policy	Still there: campaign targeting, policy, "budget availability", pacing	Same discipline as any channel
Ranking	Predicted clicks and conversions	Still there: relevance, "predicted engagement", "historical performance"	Write ad copy for the shopper's goal, not one keyword: OpenAI scores relevance to "the user's goal"
Auction	Bids and price	Still there. The orchestrator "doesn't directly decide which ad will be served, nor does it set a price"	Bid as you would in any auction
Next to the content	Ads sit inside the feed or the results	A separate service, "to keep ads from influencing answers"	Work on the answer separately, through your pages and the sources AI reads. Citable drafts pages and posts for your team to approve
Where it shows	A slot in the feed or the results	"below ChatGPT's response, clearly labeled as sponsored"	Know which brands ChatGPT names when asked the questions your buyers are likely to ask. On paid plans that include ChatGPT, Citable checks this weekly for the questions it tracks
Showing nothing	The flow ends with "the resulting ad"	"Returning no ad is an ordinary, successful outcome"	Judge a test on results, not impression volume

So the auction works like a conventional one. The new parts are the reading at the start and the decision at the end about whether to show anything at all.

Why doesn't ChatGPT treat each message as an ad request?

OpenAI names four risks of doing that: "misreading the user goal by oversimplifying it," "beginning the advertising flow too soon, before the user's real need has become clear," "showing ads that get in the way of the user journey instead of supporting it," and "inefficiently rerunning the ads flow every time the request changes."

OpenAI's example: a conversation starts with how to train for a race, and "several turns later, they ask about shoes to help train." In OpenAI's words, "a system that acted on the first message could retrieve the wrong intent entirely."

The orchestrator also has to "recognize when a user is speaking hypothetically, and avoid treating every mention of a product or service as a request to buy something."

What it means for your store: the product question can come several turns into a conversation about a goal, sometimes after a few what-ifs. Your pages should answer the questions that lead up to the purchase and the trade-offs a shopper tries out, not only the product name.

How does the ad system turn a conversation into what it looks for?

It can split it. OpenAI says "multiple queries can preserve different aspects of intent": "a conversation about planning a trip, for example, might contain distinct needs related to transportation, lodging, or activities." A conversation about a Yosemite trip and a shoe brand becomes two queries: "[shoes appropriate for hiking trip]" and "[new [brand] product lines good for travel]".

What it means for your store: answer each need on its own page: who the product is for, how it compares, what it costs, how it fits a specific use.

Do reviews matter for ads in ChatGPT?

For product-feed ads, yes. OpenAI writes that "the index for product feed ads is designed to be aware of positive or negative product reviews as it fetches the best possible ads."

Can paying for ads get my brand into ChatGPT's answers?

OpenAI says no. The services use "separate models and separate code," and "ChatGPT has no dependency on the ads system." Being named in the answer and appearing in a sponsored slot below it are two separate jobs. An ad budget pays for the second. Being named in the answer is work on your own pages and on the sources AI reads.

---

What to do this week

1. Put product facts in text. Sizes, materials, ingredients, specs, price, shipping and returns, written on the page, not only in images.
2. Write one page per need. Take the questions a shopper is likely to ask on the way to your product (who it's for, how it compares, what if the budget is lower), and answer each on its own page.
3. Make reviews visible. OpenAI names reviews as an input to the product-feed ad index.
4. If you run ChatGPT ads, check your feed. It's the input the product-feed index reads.
5. Measure the organic side separately. ChatGPT adds utm_source=chatgpt.com to the links it sends to websites. With Google Analytics connected, Citable shows visits from AI assistants, as Google Analytics attributes them.

---

See how often AI names you, next to your competitors. Get your free AI report →