On February 9, 2026, OpenAI started testing ads in ChatGPT. Sponsored results at the bottom of responses, matched to the topic of your conversation, your past chats and your history with previous ads. Free and Go tier users in the U.S. see them now. Plus, Pro and Enterprise users don't.
The CPM (cost per thousand impressions) is $60. For context, Meta charges about $20. OpenAI is charging three times more because the data is worth three times more.
Think about the difference. Facebook knows you visited a running shoe website and liked a post about marathons. ChatGPT knows you're training for your first marathon, your left knee has been bothering you since November, you're worried it might be a meniscus tear, and you haven't told your doctor yet because you don't want to be told to stop running.
That's not browsing data. That's a window into your head. And it's now being monetized.
OpenAI says conversations stay private from advertisers. Advertisers get aggregate stats like views and clicks, not individual chats. But the targeting runs on your past chats and saved memories. OpenAI's own help center confirms it: with personalized ads enabled, targeting uses "your current chat thread, including personalized model responses." The ad system reads your conversations to decide what to sell you. The advertisers just don't see the source material.
I used to do this for a living
Before BeatMask, I spent years in advertising technology. Most of our time was spent trying to connect anonymous signals into a picture of a real person.
Someone shopped at Macy's on Sunday. Someone else posted a photo of themselves tagged at the same Macy's that afternoon, wearing new hiking shoes. A third person lives at this address and is a mom in her 30s. Connect the dots: the Macy's shopper lives at that address too (loyalty card data). The person posting photos is into outdoor gear based on everything else on their social feed. Buy a data set from a broker, and suddenly you can layer in travel history and spending patterns too.
That work was hard. It took multiple data sources, matching algorithms, broker relationships and constant gaps where the signals didn't connect. You were always guessing.
Then Facebook came along and crushed a generation of ad tech companies overnight. Facebook didn't need to stitch anything together. People told Facebook who they were, who their friends were, what they liked, where they went and what they cared about. Voluntarily. In one place. The data was better, the targeting was sharper, and Facebook could charge more because the results were better. A lot of companies I knew didn't survive that shift.
ChatGPT just did the same thing to Facebook.
What we pieced together in ad tech was worth pennies. A fraction of a cent per data point. Facebook's version was worth dollars. What OpenAI has is worth orders of magnitude more because people don't curate what they tell a chatbot. They type what they actually think. A single ChatGPT user is worth more to an advertiser than a thousand of the cobbled-together portraits I used to help build.
OpenAI's internal projections put "free user monetization" at $1 billion in 2026, growing to $25 billion by 2029. For context, that's about what YouTube's entire ad business generates today.
And that's just ads. The real money hasn't started yet.
ChatGPT already has in-chat checkout. It already connects to shopping, travel and financial apps. The next step is obvious: AI agents that browse, compare and buy things on your behalf. When that happens, the model doesn't just know what ad to show you. It knows your budget, your taste, your health restrictions, your schedule, your kid's shoe size and the anniversary you keep forgetting. It makes the purchase for you. The advertiser doesn't pay for an impression. They pay for a conversion that the AI delivered because it knows you better than you know yourself.
The stitched-together data I used to help build in ad tech was worth a fraction of a cent. A ChatGPT user who shops, books travel, manages health and plans their life through an AI agent is worth thousands of dollars a year in commerce revenue alone. That's before you count the ads, the subscriptions, the API licensing and the enterprise deals.
A fraction of a cent to thousands of dollars. That's not 1,000x. That's closer to a million.
Why this one is different
I did this work for years. I know what the old data was worth and I know what made it better.
The stuff we stitched together in ad tech could be embarrassing. If it leaked, someone might learn you visited a gambling site or browsed engagement rings before a breakup. Awkward. Survivable.
What ChatGPT has on its users isn't embarrassing. It's dangerous. These aren't records of what you clicked on. They're records of what you thought. The legal strategy you're weighing. The medical symptom you haven't told anyone about. The marriage you're not sure about. The financial hole you're trying to dig out of. The things you typed at midnight when you thought nobody was watching.
That data, in the wrong hands, isn't an advertising tool. It's leverage. And "wrong hands" doesn't require a hack. A lawyer subpoenas your chat history in a divorce. A government agency requests it. An employer's legal team pulls it in discovery. An insider at the company looks where they shouldn't. None of those are breaches. They're things that happen every day, through normal legal channels, to data that's already tied to your name because you were logged in when you typed it.
The gap between what we had to work with in ad tech and what ChatGPT has now isn't incremental. It's the difference between knowing someone shops at Macy's and knowing what keeps them up at night.