ChatGPT Ads: The End of the Subscription Fantasy

Sam Altman once called the combination of ads and AI “uniquely unsettling” and described advertising as a “last resort” for OpenAI, as Search Engine Land recounted. By February 2026, OpenAI was testing ads in ChatGPT. Weeks later, it reported more than 900 million weekly users and more than 50 million consumer subscribers.

Those two numbers explain the change of heart better than any statement could. Roughly one weekly user in 18 pays. The other 17 use ChatGPT for free, and every answer they get costs money to produce. For an advertiser, the more useful question is what that business model gives you, and what it withholds.

The arithmetic that made ads unavoidable

A subscription business earns from the people who pay. A product used free by hundreds of millions of people has to earn from the rest, and the established way to do that is to sell their attention. Search engines and social networks arrived at that answer long ago, and ChatGPT’s free users carry a real computing cost with every reply.

OpenAI’s plans show how much is riding on it. Search Engine Land reported that internal figures counted on about $1 billion in 2026 revenue from what OpenAI called “free user monetization.” In April 2026, OpenAI’s investor projections, reported by Axios and Reuters, put ad revenue at $2.5 billion in 2026, $25 billion in 2028 and $100 billion by 2030. By August, OpenAI said ChatGPT Ads had reached a $1 billion annual run rate.

Subscriptions still matter. They are simply not where the growth is meant to come from. The users who pay are a small minority, and in OpenAI’s own projections advertising becomes one of the largest lines in the business.

What that model gives you as an advertiser

A platform that needs to grow ad revenue tenfold in two years needs advertisers to succeed. That gives buyers more bargaining power than the early pilot suggested.

The pilot showed it. Early advertisers faced large minimum commitments, impression-only pricing and thin reporting, and complained publicly when campaigns underdelivered, Digiday reported. Within about three months, OpenAI had added cost-per-click bidding, a pixel and a Conversions API, and a self-serve Ads Manager with no minimum commitment. A seller under that much revenue pressure adjusts to its buyers.

Use that. Buy on cost per click rather than impressions where the goal is action. Insist on conversion data before scaling. Decline commitments the platform cannot back with delivery. And run your own holdout tests, because a platform that needs to prove its value has every reason to report its results generously. The practical details of buying are in the buyer’s guide to ChatGPT ads, and testing a channel against what already works is the job of performance marketing that measures what ads actually caused.

What it does not give you

The same model sets hard limits, because ChatGPT’s value to all those free users depends on people trusting its answers. OpenAI’s stated rules are that ads stay separate from answers and do not influence them, that advertisers see only aggregated data rather than conversations, and that ads are not shown near health, mental health or politics. You can buy a place below the answer. You cannot buy a place inside it, and you cannot target with exact keywords.

The subscribers are out of reach too. Plus, Pro and Business accounts carry no ads, so the people who value ChatGPT enough to pay for it are the ones advertising cannot touch.

And the revenue targets create their own pressure. Getting from $1 billion to $25 billion means more ads, more formats or higher prices. Each of those tests how much users will tolerate, and users already say that ads would lower their trust, the tension explored in what the research shows about trust in ChatGPT ads. If that trust erodes, the inventory you are buying gets worse.

Not every assistant has made the same choice. Perplexity is ending the ad program it started in 2024, saying users need to believe they are getting the best possible answer, Campaign reported. A smaller company can make that call. One with a billion free users and investor projections of $100 billion in ad revenue has already made the other.

For an advertiser, that settles the strategy. OpenAI needs ad revenue too much to turn your budget away, and needs user trust too much to let you get any closer to the answer than the space below it.

Frequently Asked Questions

Why did OpenAI start showing ads in ChatGPT?

Because most of its users do not pay. In February 2026, OpenAI reported more than 900 million weekly users and more than 50 million consumer subscribers, so roughly one weekly user in 18 pays. Ads let OpenAI earn from the free majority. Its investor projections, reported by Axios, put ad revenue at $2.5 billion in 2026, rising to $25 billion in 2028 and $100 billion by 2030.

Which ChatGPT plans show ads?

Ads appear on the Free and Go plans. Plus, Pro, Business, Enterprise and Education accounts are ad-free. Ads are not shown to accounts identified as belonging to users under 18 or near sensitive topics such as health, mental health and politics. Free users can also choose an ad-free version in exchange for fewer daily messages. In practice, advertisers reach the free majority of ChatGPT’s users, not its paying subscribers.

Do you need a minimum commitment to advertise on ChatGPT?

No, not since May 2026. During the early pilot, advertisers reported large minimum commitments, with Digiday citing $250,000, and pricing was based on impressions. Since May 2026, OpenAI’s self-serve Ads Manager has let businesses in eligible countries buy without a minimum, using either cost-per-impression or cost-per-click bidding, with a pixel and Conversions API for measuring results. Larger advertisers can still work through OpenAI’s ads team or its agency and technology partners.

Can advertisers influence ChatGPT’s answers?

Not according to OpenAI. Its stated principles are that ads are clearly labeled, shown separately from ChatGPT’s answers, and do not affect what ChatGPT says. Advertisers receive aggregated performance data rather than access to conversations. That is the main limit of the format: you can buy a placement below an answer, but appearing inside the answer depends on being a source the model draws on.