How ChatGPT Ads Manager works

Ads now sit between the lines of your ChatGPT answers. Here is what the ad manager costs, how targeting works, and what marketers say after six months

People have spent the last year trying to sneak into what ChatGPT says - by poisoning its sources¹, gaming the data it reads², or hiding instructions inside a prompt³. That is the gray door. Now there is an official one, and it is paid: through targeting, you can appear inside ChatGPT's answers on a regular basis.

That raised one question for me right away: what stops someone from buying their way in with a political message? So I read OpenAI's ad policy. Political ads are banned outright - anything that argues for or against politicians, elections, public policy, or contested social issues⁴. Every ad also passes a three-step review: the advertiser is verified, the creative and landing page are checked by classifiers and language models, and the placement is checked so the ad shows only in a fitting conversation. Money-sensitive areas like finance, health, and legal get manual review on top.

The pilot, and how you get in

The ChatGPT Ads pilot launched six months ago, on 9 February 2026. At first the minimum ad budget was $250,000. Then the threshold was removed. Now you can test ads with as little as $5, and freelancers can advertise too⁵. You apply for the self-service Ads Manager at ads.openai.com, and approval takes about two weeks.

The placement itself sits in a far more intimate space than we are used to - right between the lines of the dialogue. The first user reaction is rarely a pleasant one.

Only a few countries - and no European Union

There are still only a few markets. The United States, Canada, Australia, and New Zealand were recently joined by the United Kingdom, Brazil, Japan, South Korea, and Mexico - nine markets in total⁶. The European Union is not on the list. OpenAI says it is still building the infrastructure to comply with European law: the tracking pixel that measures conversions needs a European version that can exclude certain actions from tracking.

One European partner has already appeared: Zalando. Its ads can be shown when people ask ChatGPT about shopping. David Dugan, OpenAI's head of global advertising, announced it on LinkedIn⁷.

How much a click costs

A lot. People are reluctant to click - why click if ChatGPT has already told you everything? In real campaigns I have seen, a click costs around $10-20, and the click-through rate is around 1-2%.

The first reviews from marketers are cautious. The Canadian agency Longhouse does not sound thrilled with its campaigns:

We're seeing the potential for strong visibility, but fewer immediate conversions than we would expect.

Austin LeClear points at a deeper problem:

The audience problem is bigger than the numbers. Ads only show to free ChatGPT users - and business owners and marketers are almost all on paid plans.

So for B2B, where deal values are much higher, there is not much to do here yet.

Targeting is a prompt

The most interesting part: targeting in ChatGPT Ads Manager is written as a prompt. There are no keywords and no demographic audiences.

What the ad looks like

For now there is only one very basic format: one line of text, up to 48 characters, plus a small square image, 512 × 512 px. Most advertisers simply drop in their logo.

In shape it is Google Pay Per Click advertising when it first appeared. History has come full circle: the most futuristic product of the decade is monetised like a search engine from 2002.