By Karsten Weide, Chief Analyst, W Media Research.
At Cannes Lions in June, OpenAI’s Chief Revenue Officer Denise Dresser told a room full of CMOs that the company is “clearly in the advertising business now,” and that we’re witnessing a shift from the attention economy to something she called the intelligence economy, where ads succeed by being useful instead of intrusive. It’s a nice sentiment. I’ve heard some version of it from every ad platform since the advent of the pop-up window, usually right before the platform in question got exactly as intrusive as everything that came before it. So let’s set the rhetoric aside for a moment and look at what advertisers can actually buy, whether it works, and whether anyone serious is showing up with real budget.
What OpenAI is actually selling
Strip away the intelligence economy branding and the current offer looks unexcitingly familiar. Ads currently appear only for users of the two lowest tiers, the Free and Go offers. A labeled sponsored card appears below a chat answer, matched to conversation context, bought through a self-serve Ads Manager. It opened in May on CPM, CPC, and, as of early June, CPA models, with recommended starting CPC bids around three to five dollars. There is no video and no carousel yet, and no product feed that would allow buyers to run retail media ads. Targeting runs on chat conversations and context hints, plus, since mid-June, on custom audience uploads and first-party data.
A lot of this is not being built in-house. Instead, OpenAI runs much of its stack on Criteo, Adobe, StackAdapt, Pacvue, Kargo, and LiveRamp. There’s also a Trade Desk courtship that so far consists only of OpenAI poaching TTD’s chief strategy officer Samantha Johnson. Using third-party tech is a sensible course to buy time to market. It is not a new kind of advertising.
Does it actually work?
Here’s where the intelligence economy pitch must earn its keep, and where the evidence gets thin. Early click-through-rates hover around 0.9 – 2% in various tests, lower than on Google Search. OpenAI claims that dismissal rates (meaning the share of ads that users hide because they find them irrelevant or annoying) are dropping, indicating improving targeting and effectiveness. Alas, keep in mind that every trust and dismissal metric cited so far is self-reported, with no outside auditor checking the homework.
OpenAI’s CPM opened at 60 dollars in February and roughly halved to around 25 dollars within ten weeks, which the company can spin as healthy price discovery and I would read as a market finding the true value of ChatGPT ads once the novelty had worn off.
The verticals actually spending, travel, retail, beauty, and financial services are exactly the categories that would show up on any platform with contextual targeting and a pulse. Their presence tells you the format is functional, not that conversational intent beats a search keyword. That’s the actual empirical question underneath all the Cannes poetry, and eleven months in, nobody outside OpenAI has answered it with real data.
Who’s actually buying
The topline numbers seem genuinely impressive at first glance. More than six hundred advertisers generated a hundred million dollars in annualized revenue within six weeks of launch. But watch the minimum spend, because that’s where the real story is. It started at $200,000 in February, a price expecting the curiosity factor and scarcity to have an impact. Ten weeks later it was $50,000 dollars. By the time Criteo got involved, partners were pitching entry at $10,000. That is a company lowering the toll because buyers weren’t showing up at the original price, indicating how they felt about the product. Agencies, for their part, describe themselves as learning alongside the platform rather than shifting real budget into it, which is agency-speak for a curiosity line item instead of one in an actual media plan. OpenAI is telling its investors to expect two and a half billion in ad revenue in 2026. Compare that with industry estimates that the entire AI-chatbot advertising category will do well under a billion dollars in 2026. Only one of the two numbers can be true, and I wouldn’t bet on the party with an IPO narrative to defend.
Everyone else says no – so far
What makes OpenAI’s confidence more interesting is how alone it is for now. Perplexity actually ran this experiment first, launched ads in 2024, and killed the whole program in February 2026, with executives telling the Financial Times that once a sponsored placement appeared, users started doubting everything else on the page. Anthropic never started and took out a Super Bowl ad specifically to say so: ads are coming to AI, but not to Claude – for the same reason Perplexity quit ads. Google’s posture is almost comically split. The standalone Gemini app stays ad-free by explicit, repeated statements from its own VP of Global Ads, Dan Taylor, while Gemini in Google Search is already selling sponsored placements inside AI Overviews and AI Mode. Elon Musk has used Grok to make advertising on X smarter, but ads do not run on Grok itself.
Where this actually goes
OpenAI’s own investor materials, per reporting this spring, project ad revenue climbing from two and a half billion this year to eleven billion in 2027, twenty-five billion in 2028, and a hundred billion by 2030, all resting on the assumption that OpenAI’s products reach 2.75 billion weekly users by decade’s end. This seems to be more of a narrative defending a trillion dollar valuation than a serious forecast.
The bull case requires the content of chats to prove measurably superior to keywords, at scale, with independent verification, none of which exists yet. The bear case is straightforward: inventory stays genuinely scarce because ChatGPT’s best use cases aren’t commercially adjacent, minimum spends keep sliding, and it settles into a real but modest line item somewhere far south of the search and social giants. Remember when Facebook first got into ads? When they pitched that because they knew who knows who, somehow, magically, ad effectiveness would be on a whole new level? That never panned out, either.
My own read, for what it’s worth: this becomes a legitimate, mid-sized advertising channel with genuinely useful high-intent inventory in a handful of categories, but it does not become the third pillar of digital advertising standing next to Google and Meta by 2030. And yet, because it will work in a fashion, I do see OpenAI’s competitors eventually joining the fray, given they are just as desperate for cash as is OpenAI.
The “intelligence economy” line was a great applause moment at Cannes. The eleven billion dollar milestone in 2027 is the one that will actually tell us whether it meant anything.
Leave a Reply