Who really pays for free ChatGPT, and does it change the answers?
Rewind to February. OpenAI starts showing ads to some people in the US, a small pilot. By August the same thing is switched on across 31 European countries. And the reported figure that makes everyone sit up: an annualised run rate of around a billion dollars in under 200 days. Ads now reach well over forty countries, on the free tier and the cheap Go plan. Paid plans stay clean.
That is not a side experiment any more. That is a business.
Why it was always going to happen
Running these models is expensive, and most users pay nothing. Somebody has to cover the bill. Search engines solved this decades ago with ads, and a chatbot that people ask for advice is, if anything, a better place to show them than a list of links. The question was never whether. It was how.
But hang on. When you ask a chatbot for a recommendation and an advertiser is paying nearby, how do you know the answer is not quietly tilted?
That is the trust problem, and it is the one thing to watch. The stated approach is to keep ads clearly separate from the answer itself. That separation is the whole deal. A search result page trained us to spot the sponsored line. A conversation that sounds like a friend giving advice makes the line much blurrier, so the discipline has to be much stronger.
There is a second effect nobody talks about much. If free users become the product and paying users get the clean version, the quality of the information people receive starts to depend on their wallet. That is worth keeping an eye on, calmly, not with panic.