ChatGPT now splits free and paid users by ads, not just features — but does anyone actually track how that changes behavior?

This explores whether there is measured evidence of a usage divide between free and paid ChatGPT users. The short answer: the collection does not contain a study that measures this directly. What it has is the business decision that creates the split, plus usage research that shows who adopts ChatGPT, without breaking the data down by plan.

The clearest material on tiers is about policy, not measurement. OpenAI's plan to test ads on the free and $8/month Go tiers, while keeping Pro, Business, and Enterprise ad-free, is presented as a way to pay for wider access for lower-income users Can advertising revenue actually expand AI access equitably?. The note points out what was left out: there was no adoption data and no detail on usage limits. So the company has formally divided the product into a subsidized, ad-supported experience and a paid, ad-free one, but it hasn't published numbers on how the two groups behave.

The large usage studies give an indirect hint. A privacy-preserving analysis of real conversations found that non-work use grew to over 70% of all ChatGPT use. Work use is mostly writing, and it is concentrated among educated professionals How is ChatGPT actually being used by real people?. Those professionals are likely the people who pay for ChatGPT or get it through their employer, but the study measures job and education, not plan. Treat it as a pointer, not proof. OpenAI's loneliness study likewise sorts users by how much they use ChatGPT, not by what they pay. It found that heavy use goes along with more loneliness and dependence, and that voice mode softened the effect Does heavy ChatGPT use make people lonelier and more dependent?. Usage limits on free plans could change who becomes a heavy user, but nobody in the collection has tested that.

The less obvious question is whether any gap would be about outcomes rather than access. The field experiment where ChatGPT learners did worse than Google searchers Does ChatGPT harm informal learning compared to Google Search? suggests that a more capable model doesn't automatically mean better learning, because the chat format itself narrows what people explore. If paid tiers mostly offer more of the same interface, the gap that matters may not track the price.

Bottom line: the collection documents why a free/paid divide exists and hints at who ends up on each side, but it has no measured comparison of adoption, intensity, or outcomes by plan. That is a real gap in the collection, not something hidden under different wording.