ChatGPT is selling ads. The money showed up before the measurement did.

8 min read

TL;DR: ChatGPT ads are live in the US (Free and Go tiers) and reached 100 million dollars annualized in six weeks. But until recently advertisers saw only impressions and clicks, with no conversion attribution. The money is moving ahead of the measurement. I would fund it as an experiment from upper-funnel budget, not reallocate search.

OpenAI spent years insisting it would not put ads in ChatGPT. That ended in early February 2026, when the ad pilot went live in the US for the Free tier and the lower-priced ChatGPT Go plan. Within about six weeks it was generating more than 100 million dollars on an annualized basis (Reuters).

That number is the headline everyone repeated. The number I could not stop thinking about came from CNBC: one enterprise advertiser had committed a 250,000 dollar budget and spent roughly three percent of it over several weeks, in part because a glitch in OpenAI’s ad manager stopped them from seeing their own campaign data (CNBC, reported via trade press).

Hold those two facts next to each other. Money is pouring in at a 100-million-dollar pace, and at least one buyer could not even see what their money was doing. That gap is the entire story of ChatGPT ads in 2026, and it should decide how you behave.

The measurement arrived after the money

Here is the part the excitement skips. For the first stretch of the pilot, advertiser reporting was, in the words of Adthena’s CMO Ashley Fletcher, “essentially impressions and clicks only, with no conversion attribution” (Digiday). Digiday’s own headline for the early advertiser experience was that they were “flying blind.”

OpenAI has since started fixing this. A measurement pixel and a conversions API arrived, and the company moved quickly from CPM-only buying to cost-per-click and then toward conversion-optimized campaigns (Digiday). So the tooling is coming. But read the timeline plainly: a credible, accepted way to know whether these ads convert did not exist for the first few months that hundreds of advertisers were spending real money.

The spend is a bet on presence and FOMO, not yet on proven performance.

I have watched this pattern before. A new surface appears, nobody wants to be late, budget arrives ahead of instrumentation, and for a while “being on it” quietly stands in for “it works.” The honest read of the early ChatGPT numbers is not that the channel is proven. It is that fear of missing out is a powerful media buyer. Even eMarketer’s coverage of the 100-million milestone carried the analyst caveat that agency leads had not yet seen measurable business outcomes (eMarketer).

Where the budget is actually coming from

When a new channel opens, the money does not appear from nowhere. It gets moved. The interesting question is moved from where, because that tells you what bet the market is really making.

The obvious guess is paid search, since a question typed into ChatGPT looks a lot like a search query. But that is not what is happening yet, and the most telling voice on this is OpenAI itself. Its ads lead, Asad Awan, has been steering buyers away from search budgets: “Maybe ChatGPT is a different kind of a place. This is not the same as a discovery platform feed or pure search” (Digiday). An agency executive in the same reporting relayed the pitch more bluntly: do not take it from search, take it from anything else you are doing up the funnel.

And that is where the money is coming from. Agencies in the same coverage describe pulling from programmatic display and from experimental and brand budgets, not from cash earmarked for search. The format is small and sits high in the funnel, so it is being funded like an awareness buy, not a direct-response one.

The analysts do think search budget gets hit eventually. As ROAST’s John Barham put it, the most effective route to taking Google search budget is to accept that most advertisers already have a Google Ads account (Digiday). But that is a prediction tied to measurement maturing, not something the data shows today. Treat the search-cannibalization story as a forecast, not a fact.

Why a low click rate is not the failure it looks like

One number that gets waved around as damning is the click-through rate. The research firm Adthena, which has been analysing the pilot’s placements, pegged early ChatGPT ad click-through at a low single-digit percentage, far below paid search (Adthena’s analysis is detailed in Digiday’s pilot coverage). On its face that looks like a channel that does not work.

I read it differently. A person mid-conversation with an assistant is in task-completion mode, not shopping mode. Low click-through on a conversational surface is a signal about the user’s intent state, not proof the placement is worthless. It means you should not buy this like last-click search and judge it on click-through ROAS. You should buy it like a brand or awareness surface and judge it on assisted conversions and lift.

This matters because of how conversion actually happens here. Someone absorbs a product inside a multi-turn dialogue, then converts later through branded search or by going direct. A click-based model gives ChatGPT zero credit for influence it genuinely had. Even OpenAI’s new pixel, which sets a first-party cookie with a 30-day window, cannot see the persuasion that happened inside the conversation. So if you measure this channel only by the click, you will systematically under-count what it did, conclude it failed, and turn it off for the wrong reason.

The risk that actually caps this

The thing that makes ChatGPT valuable to an advertiser, the trust of a private advisory conversation, is also the thing that limits it. Robert Webster, who runs the AI consultancy TAU, put the product risk precisely: if you ask for a great laptop and the answer is just whoever bid highest, the product breaks. The line between the answer and the ad is the whole thing (Digiday).

That is why OpenAI keeps stressing that ads do not influence the answer, that conversations are not shared with advertisers, and that fewer than one in five users are shown ads on a given day even though most are eligible (Reuters). Those are not just user-trust choices. They are the ceiling on ad inventory. OpenAI’s product risk is the advertiser’s scale problem: the surface can only carry so many ads before it stops being trusted, which means the bottleneck on this channel is not advertiser demand, it is how much OpenAI can show without breaking the thing people came for.

How I would actually play it

Not by sitting it out, and not by moving a real performance budget into it. Both are wrong.

I would carve a small, ring-fenced test budget and fund it as the cost of learning a surface early, not as a performance line that has to hit a ROAS target this quarter. I would instrument everything on my own side, because the platform’s attribution is new and partial, so first-party tracking, the “how did you hear about us” question, and branded-search and direct-traffic lift become the honest signals. And I would keep two reasons to be there separate in my own head: presence, being a brand the model surfaces and a buyer meets in a place they increasingly trust, which is real today, and performance, capturing trackable conversions, which the measurement has not yet earned.

The pattern here is older than ChatGPT. A surface gets hot, money races ahead of proof, and the operators who win are not the ones who moved the most budget fastest. They are the ones who spent a little, measured hard on their own terms, and kept the discipline to call it an experiment until the numbers earned the word channel.

That is where I am on ChatGPT ads. The rollout is real, the spend is real, the early interest is real. The proof is not here yet. Spend like it.

Are ChatGPT ads live, and who sees them? Yes. They launched in the US in early February 2026 on the Free and ChatGPT Go tiers, with expansion to Australia, New Zealand, and Canada announced shortly after. Paid Plus and Pro tiers are not in the ad pilot.

Can you track conversions from ChatGPT ads? Only recently. A measurement pixel and conversions API arrived in early May 2026 and conversion-optimized campaigns in early June. For the first months of the pilot, reporting was largely impressions and clicks, and attribution on a conversational surface is still unsettled.

Should I move my Google Ads budget into ChatGPT? Not today. The money flowing in is coming mostly from programmatic display and experimental budgets, and OpenAI is actively telling advertisers not to pull from search. Analysts expect search budgets to be the eventual target once measurement matures.

Do the ads change the answers ChatGPT gives? OpenAI says no. Ads are labeled as sponsored and appear separated from the response, and OpenAI’s stated position is that the ad does not influence the answer itself.