Advertising in ChatGPT: what has already been confirmed by OpenAI, and what still remains rumors about CPC
It's easy to get confused with advertising on ChatGPT these days. Some things have already been confirmed by OpenAI itself, while others are still being discussed by the market based on indirect evidence and publications by specialized media. You cannot mix one with the other.
What has been officially confirmed? In January 2026, OpenAI released material "Our approach to advertising and expanding access to ChatGPT". The company explicitly wrote that it was going to start testing ads in the US for Free and ChatGPT Go, promised explicit labeling and specifically emphasized that responses should not be tailored to the advertiser. It also says that there will be no advertising in Plus, Pro, Business and Enterprise.
But the CPC ranges, screenshots and conversations about the first rates are no longer the official OpenAI price list, but early market signals. For example, Search Engine Land wrote that OpenAI is quietly preparing an advertising infrastructure, and Search Engine Journal published a message about early CPC rates. Is it useful to read this? Yes. Take this as a stable market benchmark? No.
What Really Happened
Essentially, OpenAI took the first official step: it admitted that advertising would appear and described the restrictions in advance. This is more important than any rumors about rates, because it was from this moment that the market ceased to be a fantasy and became a real direction for testing.
This is an important point. Businesses can now discuss advertising on ChatGPT not as a fantasy, but as a future working channel. But this is still an early stage. This means that you need to test it more carefully than the usual search advertising.
Why it's dangerous to focus on one CPC
In regular search, we are used to looking at the cost per click as a clear guide. In AI assistants, CPC alone is no longer enough. The user scenario itself changes there.
A person in ChatGPT often asks not just a request, but an entire task. This may indicate a more mature interest. But it can also mean something else: the user receives part of the benefit directly in the response, but not everyone comes to the site anymore.
Because of this, the same CPC can hide very different quality of traffic. An expensive click sometimes brings a strong lead. And sometimes it’s just an interesting transition from someone who hasn’t intended to buy yet.
What should you consider instead of just one click cost?
If you are testing advertising in ChatGPT or another AI assistant, it is more useful to look at this combination:
- lead quality, not just the number of leads;
- viewing depth after transition;
- time until the first meaningful action;
- share of branded and non-branded scripts;
- assisted conversions;
- returns to the site;
- the difference between exploratory and transactional intent.
Otherwise, you can see “expensive but beautiful” traffic and decide that the channel is already working, although it does not reach sales.
How to test such a channel without self-deception
1. Separate official facts and market rumors
What OpenAI has already confirmed ad tests, can be taken as a fact. Anything related to specific market rates and early interfaces is best considered a hypothesis until you have your own test.
2. Don’t transfer expectations from Google Ads one to the other
Advertising in ChatGPT does not have to behave like mature search. There is a different context of the question, a different format of interaction and a different depth of contact before the click.
3. Keep the pilot short and tightly limited
Such a test must have a clear budget, a time frame, and a predetermined limit after which the test stops. Not “let’s look at it for a month,” but “we check three scenarios in two weeks, after which we decide what to do next.”
4. Divide campaigns by intent type
The most common mistake is to lump all transitions from an AI channel into one basket. In fact, you need to look separately:
- research questions;
- commercial comparisons;
- brand scripts;
- top of the funnel versus almost ready demand.
5. Compare more than just search
It is useful to compare the new channel not only with Google Ads, but also with other formats that work on attention and early interest: discovery placements, content campaigns, retargeting, branded defense.
Who might benefit from this test?
Most likely, the pilot will make more sense to those who are selling not an impulse purchase, but a solution to a more complex problem. This is B2B software, finance, education, professional services, medicine, expensive services, enterprise choice with a long cycle.
There, the user more often actually comes with a question, and not just with the intention of “buy now.” This means that the AI assistant can become part of the selection scenario.
If a business lives on strict performance here and now, you can still test such a channel, but with even greater caution. Otherwise, it’s very easy to overpay just for interest.
Why the article is also useful for SEO
There is a direct demand here: реклама в ChatGPT, ChatGPT ads, CPC в ChatGPT, как тестировать AI-рекламу, реклама в AI-ассистентах. But a good article on this topic should not play along with the noise, but help businesses distinguish proven facts from early market speculation.
For NBM, this is a good entry into a broader story: if a company wants not just to “touch a new channel”, but to build a normal system for checking AI sites and AI search, this is already a task for integration of AI into business processes, analytics and experimental design.
Useful on the topic
- Google AI Mode and query fan-out: why one strong page is no longer enough for SEO
- How to get into ChatGPT answers: GEO and AI SEO
- Case on the topic
- Profile service
FAQ
Is it possible to take CPC ranges from industry articles as a ready-made guide?
No. This is a maximum signal for a hypothesis, but not a normal guideline for a media plan.
Since the ad is verified by OpenAI, is the market already mature?
No. The course of advertising testing itself was confirmed, and not a ready-made and stable market model.
Is it worth testing such a channel now?
Yes, if you do it as a controlled pilot, and not as a bet on the new “magic” growth button.