ChatGPT Paid Subscribers: Two Years of Disclosures, One Rate Stuck Near 5%

ChatGPT turns roughly 5.6% of its weekly users into paying subscribers. That is 50 million payers against 900 million weekly users, the pair OpenAI disclosed in February 2026. Put every earlier disclosure on one line and the ChatGPT paid conversion rate has barely left 5% since 2024, while the audience more than tripled. Holding the rate flat at that scale is the achievement. It is also the problem.

What the ChatGPT paid conversion rate actually measures

A conversion rate needs a numerator and a denominator. OpenAI usually publishes them at different times, through different channels, and only once has an executive stated the rate directly.

The numerator here is paying subscribers. The denominator is weekly active users, meaning anyone who used ChatGPT at least once in a seven-day window. Dividing one by the other gives a stock ratio: everyone paying today, divided by everyone active today.

That is not what most SaaS teams call a conversion rate. A SaaS funnel rate divides new payers by new signups over a fixed window, usually counted by monthly cohort. The two numbers answer different questions, and mixing them is the most common error in coverage of ChatGPT paid subscribers.

The stock ratio is still the right headline number for ChatGPT, in my view. OpenAI's own long-range target is expressed in exactly that form. Reuters, relaying a November 2025 report from The Information, said OpenAI was aiming for 8.5% of weekly users to pay by 2030.

What the stock ratio cannot show you is speed or loss. It does not reveal how quickly a new user decides to pay, and it hides every subscriber who cancelled. A flat 5% can mean a stable base, or heavy churn offset by heavy acquisition. The public figures cannot tell those apart.

Every ChatGPT paid subscribers disclosure on one timeline

Nobody appears to have put these numbers in one place, which is odd for what may be the most watched consumer subscription business of the decade. Here they are, with the source type stated for each row, because the quality varies a great deal.

Implied shares are this post's arithmetic, not OpenAI's. Where the user and subscriber figures were reported weeks apart, the share is approximate. Only two rows carry both figures from OpenAI itself: October 2024 and February 2026.

2024: the CFO's 5% to 6%

The cleanest data point is the oldest. In an October 2024 Bloomberg TV interview, OpenAI's CFO Sarah Friar said ChatGPT had 250 million weekly users and was converting free users to paid at 5% to 6%. She added that roughly 75% of OpenAI's business came from consumer subscriptions.

That stated rate matters because it is the only time OpenAI has given the ratio directly. Every other point on the timeline is derived by dividing two separately reported numbers.

Six weeks later, Sam Altman told the DealBook conference that ChatGPT had 300 million weekly users. The Information later put paid subscribers at 15.5 million at the end of 2024, which implies about 5.2%.

2025: growth outran payment, then payment caught up

The first quarter of 2025 was the audience shock. Weekly users went from 300 million in December to 500 million by the end of March, according to Nick Turley, who runs the ChatGPT app. Paid subscribers rose to just over 20 million in the same period.

The result was a paid share near 4%, the lowest point on the line. A wave of image generation use drove much of that spring growth, and casual users arriving for one feature do not convert like users who stayed for the product.

By July the gap had closed. Reuters, relaying The Information in November 2025, reported that about 35 million users paid for Plus or Pro as of July 2025, roughly 5% of weekly users. OpenAI said in early August that ChatGPT was on track to hit 700 million weekly users that week.

2026: 50 million payers and a quieter disclosure cadence

In February 2026, alongside its $110 billion funding announcement, OpenAI reported 900 million weekly users and 50 million paying subscribers. It also said January and February were on track to be its largest months for new subscribers ever.

Since then the official figures have stopped. A July 2026 report from The Information, carried by PYMNTS, said ChatGPT was approaching 1 billion weekly users with over 50 million paying subscribers. It said the user milestone arrived seven months later than OpenAI had planned.

I read the silence as information. A company that set a record for new subscribers in early 2026 had every reason to publish a bigger subscriber number by autumn, and as of this writing it has not.

The ChatGPT conversion rate on new users tells a better story

A stock ratio averages everyone together, from users who arrived in 2023 to users who arrived last week. To see what newer users do, you can divide the change in paid subscribers by the change in weekly users across each window. Analysts call this a marginal rate.

It is crude. It ignores churn, and the dates on the two series never line up exactly. It is still the closest public proxy for how recent arrivals behave, and the result surprised me.

From the end of 2024 to the end of March 2025, paid subscribers rose by about 4.5 million while weekly users rose by about 200 million. That is a marginal rate near 2.3%. The spring surge added a huge number of users who did not pay.

From April to July 2025, payers rose by about 15 million against roughly 200 million new weekly users, a marginal rate near 7.5%. From July 2025 to February 2026, payers rose by another 15 million against another 200 million weekly users. That is also near 7.5%.

There are two readings, and they are not exclusive. One is that the product, the plans and the paywall placement improved, so newer users found a reason to pay sooner. The other is that older free users finally converted, which would inflate the marginal number without new users behaving any differently.

Either way, the marginal figure is the one a product team should care about. It shows the paywall working harder over time, even while the headline ratio looks stuck. If you have read the history of ChatGPT's plan and pricing changes, the timing of new tiers and limits is worth setting against these windows.

ChatGPT against freemium conversion rate benchmarks

The obvious comparison is the SaaS freemium benchmark. It is also the most misleading one, because it measures a different thing.

The B2B SaaS freemium benchmark measures something narrower

The most cited benchmark comes from a 2023 study by Kyle Poyar and Lenny Rachitsky, run with OpenView, Pendo and Growth Unhinged. It drew on data from more than 1,000 products, mostly B2B SaaS. It rated 3% to 5% as a good self-serve freemium conversion rate and 6% to 8% as great.

Its definition is a six-month cohort rate. It counts new accounts that start paying within six months, divided by all new accounts created in the window. It does not divide current payers by current active users.

The same study found wide spread. About 20% of freemium products converted below 2.5%, and about 15% converted above 20%. A single average hides more than it shows, which is a fair warning for anyone quoting a benchmark at a board meeting.

Consumer subscription peers: Duolingo and Spotify

Two public consumer subscription companies disclose both sides of the ratio in their filings, which makes them the cleanest comparison available. Neither is an AI product, and both have had far longer to build paid habits.

Duolingo reported 12.7 million paid subscribers against 140.6 million monthly active users for the quarter to June 2026. That is about 9.0%. Spotify reported 300 million Premium subscribers against 777 million monthly active users, about 38.6%.

ChatGPT's denominator is weekly users. Every monthly active user count includes the weekly users plus occasional ones, so on a monthly basis ChatGPT's share would be lower than 5.6%, not higher.

My conclusion is blunter than most coverage. ChatGPT is the largest paid consumer software audience built in the shortest time, and it is also, by share, a weak converter against mature consumer subscriptions. Both statements are true at once.

Why 5% is the most important conversion number in AI

Every consumer AI product is priced, packaged and funded in the shadow of this ratio. If the category leader, with the strongest brand in the field, monetises about 1 in 18 weekly users directly, a smaller rival should not plan on doing better without a sharper reason to pay.

The ratio also anchors how investors value the company. Reuters reported that OpenAI's plan calls for 8.5% of an estimated 2.6 billion weekly users to pay by 2030, around 220 million subscribers. That is about 4.4 times the February 2026 count, and it requires the paid share to rise by roughly half.

Nothing on the public timeline shows the share moving that way yet. The marginal rate offers some hope, and advertising and commerce are meant to monetise the free majority instead. Whether that bet holds is the subject of the analysis of OpenAI's valuation against its run rate.

There is a cost side too. Every free weekly user consumes inference that someone pays for. A 5% paid share means each subscriber is carrying roughly 17 free users, which is why how free tiers meter compute matters as much as the price of the paid plan.

Where this analysis is weakest

This timeline is built from a mix of official statements and reported leaks, and three of its weaknesses are serious enough to change how you should use it.

The numerator is not defined the same way twice

The July 2025 figure covered Plus and Pro consumer plans. The February 2026 figure of 50 million carried no breakdown in the coverage reviewed for this post. If it includes business seats, the consumer-only ratio is lower than 5.6%. OpenAI separately reported millions of paying business users during 2025, so the difference is not trivial.

Cheaper plans matter here too. If a lower-priced tier added many subscribers, the count could rise faster than revenue. A ratio of subscribers says nothing about what each one pays, which is why the ranking of AI revenue disclosure metrics puts subscriber counts below revenue figures.

Half the timeline rests on leaked figures

Only the October 2024 and February 2026 points come entirely from OpenAI. The 15.5 million, 20 million, 35 million and July 2026 figures all trace to The Information, often relayed through other outlets. Reuters noted it could not verify the November 2025 report.

That reporting has a good record, and the numbers are consistent with each other. They are still one outlet's sourcing, and repeated coverage of the same report is not corroboration.

Weekly users flatter the ratio and hide churn

Weekly active users are a stricter count than monthly ones, so dividing by them makes ChatGPT look better against Duolingo and Spotify than a like-for-like comparison would. The direction of that bias is certain. Its size is not, because OpenAI does not publish monthly users alongside weekly.

Churn is invisible throughout. A rising subscriber count can coexist with heavy cancellation, and the marginal rates above assume none. If ChatGPT loses a meaningful share of payers each month, the real conversion of new users is higher than 7.5%, and the retention problem is larger than anyone has said.

Using the ChatGPT benchmark for your AI product conversion rate

If you run an AI product with a free tier, the ChatGPT figures are useful only once you match the definitions. Most teams compare their cohort rate against ChatGPT's stock ratio, and that comparison flatters or punishes them at random.

Track both numbers. The stock ratio divides current payers by current active users, using the same activity window every time. The cohort rate divides users from one signup month who pay within six months by everyone who signed up that month. Report them side by side, and you will see when a growth spike is diluting the headline.

The spring 2025 dip is the case study. ChatGPT's paid share fell by about a fifth in a single quarter, and nothing was wrong with conversion. A flood of casual users arrived through one viral feature. A team reading only the stock ratio would have panicked and fiddled with pricing, and a team reading cohorts would have stayed calm.

For conversion optimisation, the lesson I would draw is about where the limit bites. ChatGPT's paid tiers sell more capacity and better models, not access. The level at which free use runs out is the real price. That trade is explored in the post on AI usage limits, and for agent-driven products in product-led growth when the user is an agent.

Set your target against the right peer. A self-serve B2B tool should hold itself to the 3% to 8% cohort band. A consumer AI app with a broad free tier should expect to sit nearer ChatGPT's 5% of weekly users, and should be suspicious of any plan that assumes Spotify-like penetration.

Frequently asked questions

Where to start this week

Pull your last six signup cohorts and compute what share of each paid within 30, 90 and 180 days. Then compute today's stock ratio using a fixed seven-day activity window. If the two numbers tell different stories, the cohort number is the one to believe.

Write the definitions down before anyone presents them. Name the numerator, the denominator and the activity window in one sentence, and refuse to compare your figure with ChatGPT's 5.6% until those three match. Most bad pricing decisions in AI products start with two conversion rates that were never measuring the same thing.

Related reading