OpenAI is putting ads inside ChatGPT image generation. Free users go first.

The placement is the thing to notice

Ads inside ChatGPT are not new. A visual ad format that appears while the model is drawing you a picture is.

OpenAI said on 5 October 2026 that it is introducing "a new visual ad format in ChatGPT," built around images rather than text links. The company describes the creative as showing "product inspiration, product usage, or the experiences they make possible," so a person can "discover a new brand, understand what makes an offering distinctive, and decide what to buy."

The format is less interesting than the slot. "Initially, we'll test this new ad format during image generation in ChatGPT for users on the Free and Go plans," the post says. Testing "will begin later this month in the US with an initial group of advertisers."

Read that again. The first placement is the few seconds when you are staring at a progress indicator waiting for a picture, and the first audience is the tiers that pay nothing or nearly nothing. That is not an accident. It is the highest-attention moment in the product, attached to the users OpenAI cannot monetise any other way.

OpenAI commits to two things about it: ads will be "clearly labeled, and remain separate from the image being created," and "advertising does not influence the answers ChatGPT provides." Both are promises rather than checkable facts. Nobody outside OpenAI can currently audit the second one.

Three numbers, three vendors, nobody independent

The post's most quotable material is a set of early advertiser results. They are worth looking at closely, because the sourcing on each one is doing a lot of work.

According to DV Rockerbox, WeightWatchers' attributed cost per acquisition on ChatGPT Ads was 15.3% lower than its blended paid-search benchmark. WorkMagic reported statistically significant lift for the wellness brand Dose, with 67% of incremental purchases coming from net-new customers. And according to Triple Whale, 93% of Portland Leather's visitors from ChatGPT Ads were new.

Three different brands, three different measurement companies, three different metrics. Not one of them is an independent evaluation of the ad platform. DV Rockerbox, WorkMagic and Triple Whale are attribution vendors hired by advertisers, and OpenAI names all three in the same post as partners in its own measurement ecosystem. OpenAI is relaying its partners' reports about its own product.

That does not make the numbers false. It makes them unverifiable from outside, and it makes the comparison soft. A "blended paid-search benchmark" is whatever that advertiser's blend happened to be. A high share of new visitors is exactly what you would expect from any new channel, because nobody has been exposed to it before. "93% new" is close to arithmetically guaranteed on a platform that started selling ads this year.

OpenAI is also straight about the weakest link. On incrementality, the measurement that would actually settle whether the ads caused the sales, the post says the work "is still in its early stages," with geo-based experiments being explored alongside Haus, Measured and WorkMagic. That is the honest sentence in the release, and it is the one least likely to end up in a marketing deck.

Brand suitability is the harder problem, and OpenAI knows it

A chatbot is a worse place to sell ads than a search results page, for a reason that has nothing to do with targeting. People type things into ChatGPT that they would never type into Google with a brand watching.

OpenAI's answer is placement guardrails that assess "whether a conversation is appropriate for advertising, helping keep ads out of emotionally vulnerable, sensitive, or otherwise unsuitable contexts," plus automated review, human oversight, and a Negative Phrases control for qualifying advertisers. It is now running brand suitability evaluation pilots with DoubleVerify and Integral Ad Science, which assess how the safeguards hold up "in a controlled testing environment, without accessing private user conversations."

That last clause is the design constraint on this whole category. Independent verification and conversation privacy pull against each other, and OpenAI has chosen privacy, which is the right call and also means the audit will always be partial.

IAS chief executive Lidiane Jones framed the gap plainly in the announcement: "As advertising expands into AI-native environments, marketers need confidence that brand safety protections are durable, consistent, and built to operate at scale."

Note what she is describing: a confidence problem that does not yet have an answer. The pilots exist because the proof does not.

Why the free tier carries the ads

The strategic logic is in one sentence of the post. Advertising, OpenAI writes, "allows us to serve a much larger market than subscriptions alone."

With 1.2 billion weekly users, the overwhelming majority of whom pay nothing, subscriptions reach a ceiling fast. Ads are the only mechanism that turns the free base into revenue, and the free base is where the growth is. It is the same logic that put Finances in ChatGPT in front of free US users three days earlier, in that case to deepen engagement rather than to sell against it. Both moves treat the Free and Go tiers as the product's real centre of gravity, not its waiting room.

It also sits oddly next to everything OpenAI shipped at DevDay, where ChatGPT was reframed as a shared workspace for paying teams. Two products are being built in one app: a paid surface where you work, and a free surface where you are the inventory.

Our take

The announcement is more careful than most ad launches, and the carefulness is the tell. OpenAI published the weak spot in its own evidence, chose privacy over full auditability, and limited the first test to one placement, one country and a hand-picked advertiser list. Take the three performance figures as what they are, partner-reported results on a channel too young to have a benchmark, and ignore them until a geo-based incrementality test is published by someone who is not selling the ads.

The real thing to watch is not whether the ads perform. It is whether the labeling and the separation from the generated image still hold a year from now, when the test group is everyone on the free tier and the pressure to tighten the loop between a prompt and a purchase is a quarterly number. That is the commitment worth holding OpenAI to, and it is the one nobody outside the company can currently check.