ChatGPT has ads now. If you missed this because you've been heads down writing code or whatever, let me catch you up — OpenAI started testing advertisements in the free tier for logged-in US users back in January, and as of last week they announced the pilot has already crossed $100 million in annualized revenue. In six weeks. They hired Dave Dugan, a former Meta advertising executive, to lead ad sales, and they just expanded testing to Canada, Australia, and New Zealand. The money printer is warming up and I dont think there's any going back.
On one hand I get it. OpenAI burns through an absolutely insane amount of cash running inference at scale. They raised $110 billion in February at a $730 billion valuation, but even with $25 billion in annualized revenue from subscriptions, the economics of running GPT-5 level models for hundreds of millions of free users dont add up without another revenue stream. Ads are the obvious answer and always have been. The surprise isnt that they did it — its that it took this long.
But heres where my brain starts going in circles. ChatGPT isnt Google search. When you search Google, you type a query, get results, and move on. The relationship between you and the search engine is shallow by design. With ChatGPT, people have actual conversations. They share their anxieties, their business plans, their relationship problems, their medical symptoms. The AI knows your budget when youre planning a vacation. It knows what kind of code youre writing for what kind of project. If you've got memory turned on, it accumulates all of this over time. And now someone at OpenAI is going to figure out how to sell that intent data to advertisers, and honestly the thought of that makes me want to self-host Ollama for everything.
A former Meta employee named Miranda Bogen wrote a great piece in the Globe and Mail about this and she basically said we've seen this movie before. The ads start small and benign — clearly marked sponsored suggestions set apart from organic responses, with promises that they'll be helpful. But then more advertisers show up. Then they need to figure out which ads to show to which users. Then they build prediction models based on your demographics, your interests, your behavior patterns. Then someone gets creative and the line between organic response and paid promotion starts blurring. She should know because she watched it happen at Meta and now she runs the AI Governance Lab at the Center for Democracy & Technology.
The thing that keeps me up about this is that chatbot ads are fundamentally different from any advertising weve seen before. When Instagram shows you an ad, you know its an ad because its in a seperate little box with "Sponsored" written on it. When a chatbot recommends a product to you mid-conversation, the line between genuine recommendation and paid placement is almost impossible to draw. Is ChatGPT suggesting this tool because its actually the best option for my question, or because someone paid for that placement? Once ads are in the response stream, every answer becomes suspect. Trust erodes fast and its really hard to rebuild.
OpenAI's new CEO of applications is Fidji Simo, who came from Instacart where she basically invented sponsored products and last-minute checkout suggestions. Thats the person now running the consumer side of ChatGPT. If you think the ad integration is going to stay tasteful and minimal, I have a bridge to sell you. The playbook is already written — it worked at Meta, it worked at Google, it worked at Instacart, and its going to work at OpenAI because the financial pressure is enormous and the user data is the richest any ad platform has ever had access to.
For developers specifically this matters because a lot of us use the free tier for quick questions, code snippets, debugging help. If the free tier starts getting polluted with sponsored suggestions — like ChatGPT recommending a specific cloud provider or dev tool because someone paid for that placement instead of because its actually the best answer — thats a real problem. I already trust AI answers with some skepticism but adding financial incentives to the response generation makes that skepticism non-optional.
The $100 million in six weeks number is what really tells the story though. Thats not a test. Thats a business. OpenAI found product-market fit for advertising faster than most ad startups find their first customer. They're building a self-hosted ad system thats expected to launch broadly in April (thats like... now). And the fact that Meta already announced plans to use AI chatbot conversations to power ad targeting across Facebook and Instagram means this isnt just an OpenAI thing. Its an industry thing. Google shows ads in AI-generated search summaries. Meta will use your chatbot conversations to sell you stuff on Instagram. OpenAI puts ads directly in ChatGPT responses. The entire AI industry is converging on the same business model that turned social media into an attention-harvesting machine, except this time the machine knows you better than social media ever did because you literally told it everything.
I'm not saying everyone should delete ChatGPT or anything dramatic like that. I still use it every day. But I think we're watching the exact moment where AI assistants stop being tools that work for you and start being tools that work for advertisers while pretending to work for you. We watched this happen with search. We watched it happen with social media. And now we're watching it happen with AI and the speed is honestly terrifying — $100 million in six weeks means the incentive structure is already locked in. The only real defense is paying for the sub or running local models, and tbh thats exactly what I've been doing more and more lately.
Top comments (1)
Did you know that integrating ads into AI platforms like ChatGPT can actually improve model performance? In our experience with enterprise teams, using real-world data from ads helps refine AI models by exposing them to diverse language patterns and user interactions. This can lead to more robust language understanding and ultimately, better agent performance in real-world applications. - Ali Muwwakkil (ali-muwwakkil on LinkedIn)