Seven questions a day is the unit of OpenAI's advertising business

OpenAI's finance chief has just told investors, without quite saying so, that the free tier is the product.

Sarah Friar put numbers on ChatGPT usage in an interview released on 9 September 2026. Free users run about seven turns a day. The first paid tier does roughly double that. Plus, at $20 a month, does about three times a free user, and Pro about eleven times.

Those are subscription numbers on their face. Read them again as inventory.

Seven turns a day across a month is a bit over two hundred questions from somebody paying nothing. A Plus subscriber at three times that rate produces about 630 questions for $20, which works out at 3.2 cents a question. That is the price OpenAI has established for one interaction with a paying customer, and it is the only per-unit figure here that anyone has verified.

Now price the other side

SemiAnalysis was writing a year ago about a free base above 700 million and growing quickly, and argued then that the model released that summer was built for those users rather than for power users, with indirect monetisation as the prize. Take that figure, stale and conservative as it now is, and multiply by seven. Call it five billion free interactions a day, and the real number is higher.

Tibor Blaho's weekly recap, citing OpenAI, put ChatGPT Ads at a billion dollars of annualised run rate in under two hundred days. Divide that across five billion daily turns and you get roughly 0.06 cents per free interaction. That is our arithmetic, three sources deep, so treat it as an order of magnitude rather than a figure.

A Plus turn is therefore worth something like fifty times an advertised one, today.

That ratio is the whole business question

It reads two ways. Pessimistically, ads on assistant traffic monetise terribly and always will, because a person asking how to reformat a spreadsheet has no commercial intent worth bidding on. Optimistically, a business that reached a billion dollars in under two hundred days while capturing a fiftieth of the obvious value has an enormous amount of room, and the room is the point.

We lean optimistic, with a caveat that matters: the free user count we used is a year old and certainly higher now, which pushes the per-turn advertising figure down rather than up. What would move us is disclosure of turns with an ad shown, which no lab publishes and all of them measure.

Friar paired her numbers with a line about why the free tier exists at all. "Our mission at OpenAI is AGI for the benefit of humanity, not for the benefit of humanity who can pay, or for the benefit of humanity who live in an enterprise."

Read literally, that is a mission statement. Read as a chief financial officer speaking to investors, it is the case for advertising, phrased as a case against paywalls. The two readings do not conflict, which is what makes it a good line. You keep the free tier open because it is right, and the free tier turns out to be the asset.

Which raises the obvious question. If a free user is worth a fraction of a cent per turn and a Plus user a few cents, why would OpenAI ever push a free user to subscribe rather than to ask more questions?

Past a point, it would not. That is the shift to watch. Our expectation is that the next usage metric an OpenAI executive volunteers in public will be turns per day rather than subscriber counts, because turns per day is the number that gets better when the free tier grows.