ChatGPT ads hit $1bn a year. They buy nothing in the answer

What a billion a year actually measures

The figure came from OpenAI, not from an analyst. Its own post states that ChatGPT Ads reached $1 billion in annualised revenue run rate in less than 200 days after launch, on a platform now used by tens of thousands of advertisers. That attribution is worth getting right, because several write-ups have credited the number to Sensor Tower. Sensor Tower published a separate analysis of ad density and category mix the following day, and the version of it written up by PPC Land carries no monthly revenue figure at all.

A run rate is one month multiplied by twelve. Digiday did that arithmetic in reverse and put the underlying figure at roughly $83 million a month, with around $330 million actually booked across the first eight months of 2026. Both are Digiday derivations rather than OpenAI disclosures, so treat them as a well-argued estimate. The growth curve underneath is real enough on its own terms: the pilot crossed $100 million annualised within six weeks of its 9 February launch, according to a company spokesperson quoted by Reuters at the time. Dave Dugan, OpenAI's VP of global ad solutions, called the milestone the beginning of a new chapter for advertising.

The ads are thickening, and so far that is an American story

Sensor Tower's read of the platform, as reported by PPC Land, is that the number of ads shown per user per hour in ChatGPT's United States mobile app rose 163% between April and August 2026, an average of 26% every month. It counted roughly 1,200 unique US advertisers in August, a population growing 43% month over month since May. The mix moved as well: shopping and retail fell from 37% of US spend to 21%, while financial services climbed from 2% to 13%, travel from 5% to 11% and health from 1% to 5%. Four of August's ten biggest US spenders were financial services firms, against a single one anywhere in April's top 100.

Two things to hold on to about those figures. Sensor Tower publishes shares of spend rather than dollars, so a category tripling means its slice of the pie tripled, not that anyone's budget did. And Sensor Tower's own disclaimer, quoted in the same report, is that the figures may not accurately represent metrics for any individual app or advertiser. The direction is solid. The decimal places are not.

How often you actually meet an ad is genuinely contested, and the spread is enormous. SE Ranking ran 50,006 commercial prompts on a single day in July and found ads on 25.94% of them, close to the 29.45% the same firm measured in Google's AI Mode. Seer Interactive, tracking a repeating prompt set daily, put it at 77%, up 34 percentage points in two weeks. Otterly measured 76.4% across US commercial answers, from 85.9% in finance and insurance down to 42.3% in healthcare. LLM Pulse found 71.9% of 36,153 US answers carried at least one ad. The one-day wide sample and the two-week narrow samples are measuring different things, and nobody should quote a single number as the ad rate.

If you are reading this in Europe, none of those percentages describe your screen yet. LLM Pulse checked the United Kingdom, live since June, and found ads in 7 of 8,436 answers. OpenAI said on 18 August that ads would reach 31 European countries the following week, and confirmed on 31 August that self-service access through Ads Manager was live across all of them. Adweek's list names Ireland among them, though OpenAI's own post does not. European ads launched contextual, and OpenAI was explicit that they are not personalised at launch. Expect to see very few of them for now, and expect that to change.

The same day the run rate was announced, Brussels designated ChatGPT a Very Large Online Search Engine under the Digital Services Act, giving it four months, until January 2027, to meet the extra obligations that come with the label. One of them is a public repository of every ad it shows. We covered that designation in the EU just called ChatGPT a search engine. The two events landed together by coincidence, but the consequence is convenient: from January 2027, anyone should be able to look up who is buying what.

The advertiser in the slot is almost never the brand the answer cites

Four teams looked at the same question from four directions and landed in the same place. Seer Interactive checked 3,697 unique advertisers and found the advertiser appeared in ChatGPT's actual response 5.4% of the time and was cited as a source 3.3%. SE Ranking put the mirror image of that at 96.37% of placements where the advertiser was not among the sources used in the answer above it. LLM Pulse, across 137,038 answers, found 2.8% of ads pointed at a domain the same answer had used as a source, and 73% of advertisers never appeared as a source anywhere in the sample. Otterly came at it from the other end and found that in 12 of 14 industries the market leader ran zero ChatGPT ads, leaving the paid slot to challengers.

Read together, those four numbers describe two doors into one answer that barely open onto each other. Money buys you a labelled card beside the answer. It does not buy you a sentence inside it, and no study has found that it does. SE Ranking says so in plain language: do not expect ad spend to improve your visibility inside ChatGPT's organic answer, because being cited is a content authority problem rather than a media buying one.

One caution, because the opposite claim is already circulating. Nobody has shown that ads displace or reduce organic citations. No study measured a before and after on the same prompts, and the evidence available points the other way, towards two layers running independently. The honest version of the worry is that the answer is being colonised by a paid layer alongside the organic one, which is a different problem with a different response. SE Ranking also found that about 1 in 7 ads has no topical connection to the prompt it appears on, which tells you how loosely coupled the two layers currently are.

Your best prospects may be on a plan with no ads in it

This part has been true since the beginning and gets forgotten every time the revenue headline runs. When ads launched on 9 February 2026 they were for logged-in adults on the Free and Go tiers only, with Plus, Pro, Business, Enterprise and Education excluded. The European expansion kept the same rule. If you sell to people who pay for ChatGPT, which for most business-to-business companies is a large slice of the buying committee, the ad simply cannot reach them. They see the organic answer and nothing else. Organic citation is the only route into that conversation, and it is the segment with the most money in it.

Otterly's finding that market leaders in 12 of 14 industries bought nothing supports the same conclusion from the other side. Either the leaders have not got around to it, or they already occupy the answer and do not need the slot. Whichever it is, the paid layer today is a challenger's channel sitting underneath an organic layer that the incumbents own. That is worth knowing before you decide which one to fund. It is also the reason the traffic argument matters: AI referrals arrive in small numbers and convert unusually well, which we went through in AI sends less traffic, but it is worth far more.

On cost, be sceptical of anything you read. No OpenAI rate card is public. The CPM and minimum-spend figures circulating on agency blogs trace back to no primary source and contradict each other. The two defensible data points are both narrow: an independent account-level benchmark on real spend reported a blended 1.49x return on ad spend and roughly $1.72 per click over 15 days on about $60,000, and OpenAI itself cites an ecommerce advertiser hitting 3x return over 28 days, which is a chosen example rather than a benchmark. The whole industry is working off one account and one anecdote.

What you actually do

Four steps, and the first one is not optional, because every decision after it depends on knowing where you stand today.

1. Measure your organic presence first. Run a mention scan on the questions your customers actually ask, and find out whether ChatGPT names you at all, how often, and which brands it names instead of you. That is a measurement you can take this afternoon, it costs nothing like a media budget, and it is the only thing that tells you whether the paid slot is a shortcut or a distraction. If the answer never mentions you, buying the box below it puts your logo next to a recommendation for someone else.

2. Treat the two budgets as separate questions. "Are we cited" and "should we buy the slot" have an overlap of roughly 3%. Answer the first with data before you commit to the second, and stop expecting one to move the other. Most brands find they are further behind on citations than they assumed, which is a familiar pattern: most brands are losing AI citations and do not know it.

3. If you do buy, watch the answer above your ad. The one report nobody is producing is the pairing: your sponsored card, and the brands the answer recommended in the same response. Track the questions you bid on and read what ChatGPT says on them. An ad under an answer that praises your competitor is a measurable problem, and it is a content problem rather than a bidding one.

4. Re-measure on a schedule. US ad density climbed 26% a month across four months, Europe is only weeks old, formats are still being added, and the Commission's public ad repository arrives by January 2027. A single reading taken today will be wrong by Christmas. Put the scan on a recurring run and watch the line rather than the point.

Run a mention scan on your brand

Questions people ask about this