Ads arrived in ChatGPT. Your buyers first search is now a paid channel
For twenty years, the first thing a B2B buyer did when a problem appeared was type it into Google. That habit is moving, and this year it acquired a price list.
OpenAI began testing advertising inside ChatGPT in February. Search Engine Land reported the rollout on 9 February 2026:
OpenAI will show ads to logged-in users on the free tier and its lower-cost Go subscription. The ads will appear in a clearly labeled section beneath the chat interface, not inside responses.
Anu Adegbola, OpenAI starts testing ChatGPT ads, Search Engine Land
Paid tiers stay clean. Advertisers do not see conversations and, OpenAI says, cannot influence what the model answers.
That last detail is the one worth sitting with. The answer stays editorial; the space beneath it is for sale. Which means there are now two distinct ways to be present when someone asks a question about your category — and only one of them can be bought.
What this changes for a B2B company
Being cited is now the scarce position. On a search results page, ten links compete. In an assistant’s answer, a handful of sources are synthesised into a paragraph and the rest do not exist. The question stops being “do we rank” and becomes “are we one of the sources this thing reads”. Those are not the same problem, and the second one is decided by whether your content is unambiguous, structured and attributable — not by how many keywords sit on the page.
Zero-click stops being a complaint and becomes the design constraint. If the answer is delivered in the chat, the visit you used to get does not happen. What you are optimising for is no longer a session; it is a mention, and after the mention, a reason to come anyway. Concretely: your pricing has to be legible, your process describable in a sentence, and your claims checkable. Assistants repeat what is easy to state and safe to state.
Attribution gets worse before it gets better. A buyer who first heard your name inside a chat and typed your domain a week later arrives as direct traffic. Your analytics will file the most valuable touch in the funnel under “unknown”. Any dashboard that judges channels on last click will now systematically underprice the one that is growing.
And a paid entry point exists where none did. Whether it is worth buying is an open question — the format is young, the inventory limited, and the targeting contextual rather than intent-based in the way search advertising is. But it exists, competitors will test it, and the cost of finding out is falling as self-serve access opens up.
What we would do about it, in order
- Make the site readable by a machine that has to summarise it. Structured data that states what the company is, what it sells and who it serves. Unambiguous service descriptions. Facts stated once, in one place, in a form that can be quoted without a caveat.
- Write the answers to the questions buyers actually ask — the comparison, the decision framework, the “when not to”. Assistants surface material that resolves a question, not material that markets around it.
- Fix measurement before spending anything. Establish what direct and branded search look like now, so that a change six months from now can be read at all.
- Test paid placement last, and only against a baseline you can already measure.
The honest caveat
Nobody has a reliable model for how much of this traffic converts, and anyone quoting one is guessing. What is certain is narrower: the channel exists, it is commercial, and the qualities that make a site quotable by an assistant — clarity, structure, verifiability — are the same qualities that make it convincing to a human. That is a rare case where preparing for an uncertain future costs nothing, because the work is worth doing anyway.
Source: Anu Adegbola, “OpenAI starts testing ChatGPT ads”, Search Engine Land, 9 February 2026.
