ChatGPT Statistics: Users, Usage & Which Numbers To Trust (2026)

ChatGPT had 900 million weekly active users in February 2026, and that is the newest user count we could trace to an OpenAI disclosure. It counts people, not visits or accounts. Later headline figures circulate, but traffic panels, app estimates and roundup tallies measure different things and should never sit beside it unlabelled.

The figure applies when you need a count of people using ChatGPT in a given week. For paid users, traffic or survey shares, use the separate measure and source listed in the table below.

OpenAI publishes these numbers as company announcements rather than audited filings. Quote them as "OpenAI said" with the month, and treat anything newer than February 2026 as unconfirmed until it links to an OpenAI statement.

ChatGPT Usage Statistics At A Glance

Most ChatGPT roundups hand you a pile of big numbers and leave you to guess what each one counts. This table does the opposite. Every row says what the number measures, when it was true, who said it and whether you should quote it.

The sources behind each row are linked here. The 700 million milestone was reported from OpenAI's Nick Turley in the week of 4 August 2025. The 800 million figure came from Sam Altman at the October 2025 developer day keynote. The 900 million weekly users and 50 million paying subscribers both sit in OpenAI's February 2026 announcement. The 10% and 70% shares come from the NBER working paper on how people use ChatGPT.

How Many People Use ChatGPT Now?

The short answer: 900 million weekly active users, as OpenAI announced in February 2026. That is the newest weekly count we could tie to an OpenAI statement.

The series behind it is short and dated, so here it is in order.

• August 2025: OpenAI app product manager Nick Turley said ChatGPT was set to reach 700 million weekly users, reported in the week of 4 August 2025. Read that as a milestone the product was approaching, not a confirmed count.
• October 2025: Sam Altman said at OpenAI's developer day keynote that ChatGPT had more than 800 million weekly active users.
• February 2026: OpenAI put the figure at 900 million weekly active users in its own announcement.

ITdaily later reported the 900 million figure as attributed to an OpenAI press release in March 2026. That is a useful cross-check, but when you cite the number, cite OpenAI and not the press report.

Notice how these disclosures arrive. OpenAI's user counts come out as blog posts, press releases and keynotes, and the press relays them. There is no filing or audit behind them, which is why the speaker and the venue belong in your citation.

Higher figures have circulated since February 2026. We left them out. None of them traced to an OpenAI statement in our research, and a number you cannot trace is a number you cannot defend when someone asks where it came from. If a newer figure matters to your piece, find the OpenAI post or keynote first. If you also want to see how ChatGPT stacks up against the rest of the field, we have put the main players next to each other in AI search engines compared.

Weekly, Monthly, Paid And Visits: What Each Measure Counts

Roundups blur these measures all the time. Put a weekly user count next to a visit count and the page looks precise while saying nothing. These are the measures you will meet, and what each one does and does not count.

• Weekly active users: People who used the product at least once in a seven-day period. It does not count how often they came back, how long they stayed or whether they pay.
• Monthly users: People who used the product at least once in a month. It runs higher than a weekly count for any product with occasional users, so a monthly figure and a weekly figure are never interchangeable.
• Paying subscribers: People or accounts that pay a recurring fee. It does not count free users, and a consumer subscriber figure does not cover business seats or API customers unless the source says it does.
• Website visits: Sessions on a website. One person opening the site ten times is ten visits and one user, and the same person on a phone app is not a website visit at all.
• Survey shares: The percentage of a surveyed group who say they use or have used the product. The percentage applies to that group and that date, not to everyone.
• Third-party panel estimates: Modelled figures from traffic panels or app-store data. They are estimates built from a sample, and they should be labelled that way every time.

The one that causes the most damage is the visit count. A visit is an event, and a user is a person. Divide visits by an assumed number of visits per person and you have invented a user count, not found one. If a roundup quotes billions of monthly visits next to millions of users, it is comparing two different things on one line.

Our own review of the current top ten results found that none of them labelled the measure type or the as-of date for each figure. That is the whole problem in one sentence. A reader cannot tell whether 900 million means people, visits or accounts, and neither can the next writer who copies it.

How Many People Pay For ChatGPT?

OpenAI said ChatGPT had more than 50 million paying consumer subscribers in February 2026. The word consumer matters. The figure covers individual subscribers, and it excludes business seats and API customers.

That exclusion is easy to lose when the number gets repeated. A reader who sees "50 million paying users" will assume every paid relationship is in it. It is not, and the source does not say so.

You can put the two February 2026 figures side by side with simple arithmetic. 50 million paying consumer subscribers against 900 million weekly active users is about 5.6%.

• Paying consumer subscribers: 50 million
• Weekly active users: 900 million
• Arithmetic: 50 divided by 900 is roughly 0.056, or about 5.6%

Label that result as arithmetic. OpenAI has not published it as a conversion rate. One count covers subscribers at a point in time and the other covers people active in a week, so the ratio is a rough scale check and not a measured share of users who pay.

What People Use ChatGPT For

The best public evidence on how people use ChatGPT is the NBER working paper "How People Use ChatGPT", built on OpenAI usage data. The paper reports that about 10% of the world's adults used ChatGPT weekly as of July 2025. It also reports that roughly 70% of ChatGPT usage was not work-related.

Read that second figure twice. Most use is personal. If you are writing for a B2B audience, a claim like "people use ChatGPT to do their jobs" needs its own source, because the best evidence we have says the majority of use sits outside work. That does not mean work use is small in absolute terms. With hundreds of millions of weekly users, the minority that is work-related is still a very large group of people. It means you cannot borrow a headline user count to prove a work-use claim.

The usage data is also from July 2025, which is before the February 2026 user count. Use the two together carefully. The 10% share applies to July 2025, and it should not be recalculated against 900 million users to produce a new percentage.

There is a wider shift behind these numbers. People ask ChatGPT questions that they used to type into a search box, and that changes how brands get discovered. We cover that in what happened to SEO after AI search. It also helps to know that one question typed into an AI tool is often split into many sub-questions behind the scenes, which is explained in how query fan-out works.

Which ChatGPT Statistics Are Safe To Quote

Revenue and valuation figures are all over the roundups. We left them out. Our research found no OpenAI disclosure to trace them to, so they would be numbers with no source behind them. If you need one, go to OpenAI's own announcements and any company statement first, and quote only what you can link.

For everything else, use these rules before a ChatGPT number goes into a deck, an article or a pitch.

1. Quote a user count only if it links to an OpenAI statement, keynote or press release. A roundup repeating a figure is not a source, and roundups mostly cite each other.
2. State the measure type and the month. "900 million weekly active users in February 2026" is a fact, while "900 million users" is a rumour with a good-looking number.
3. Label panel and app-store figures as estimates. They are modelled from samples, so say so in the same sentence as the figure.
4. Name the speaker and venue when the figure came from a person. "Sam Altman said at OpenAI's developer day keynote in October 2025" tells the reader exactly where to check.
5. Drop any figure with no primary link. If the best you can say is "credited to OpenAI" with nothing to click, cut it, because a number the reader cannot check is worth less than no number.
6. Keep milestones and confirmed counts apart. The August 2025 figure was a milestone ChatGPT was approaching, so it should not sit in a list as if it were a count.
7. Check the date before you trust the headline. A title stamped with the current year over figures from an earlier year is a stale page, and this happens often.

If you want broader numbers than ChatGPT alone, see our collection of AI marketing statistics. The same rules apply there. If your real aim is to be the source an AI engine quotes, that is the work of answer engine optimization, and it starts with being the page that gets the citation right.

Why ChatGPT Roundups Disagree And Go Stale

In our review of the current top ten results for "chatgpt statistics", nine were stale or undated. The disagreement between them is not a mystery. It comes from the same few habits repeated across pages.

• Stale numbers under fresh titles. A title carries the current year, and the figures underneath come from an earlier one. The reader sees a date that was never checked.
• Credit with no link. Several pages credit a figure to OpenAI and give the reader nothing to click. Once a number loses its link, the next writer copies it as if it were settled.
• Roundups citing each other. One page takes a number from another, and that page took it from a third. The chain has no primary source at the end of it, and an error at the start travels the whole way.
• Mixed measure types in one list. Weekly users, monthly visits, app estimates and survey shares sit in a single list, in the same font, as if they were the same kind of thing.

None of this needs a villain. It is what happens when the easiest thing to do is copy the last page you read. The fix is dull and it works. Go to the primary source, write down the measure type and the month, and link it.

A Citation Template For Your Deck Or Article

Use one line shape for every ChatGPT number. It takes ten seconds and it makes your figure checkable.

The template is this sentence, filled in each time.

[Source] said ChatGPT had [figure] [measure type] in [month and year], according to [linked announcement or paper].

Before you publish, run this checklist.

1. Write the figure with its unit. "900 million" alone is not enough. Say what is being counted.
2. Name the measure type. Weekly active users, paying consumer subscribers or share of usage, as the source states it.
3. Give the month and year. The as-of date is part of the number.
4. Name the source and link to it. Use the announcement, keynote coverage or paper, not a roundup.

Here are two worked examples using verified figures.

• Example one: OpenAI said ChatGPT had 900 million weekly active users in February 2026, according to its announcement on scaling AI for everyone.
• Example two: OpenAI said ChatGPT had more than 50 million paying consumer subscribers in February 2026, according to the same OpenAI announcement.

A primary source here means the party that produced the number. For a user count that is OpenAI, in a blog post, press release or keynote. For a usage share it is the paper that analysed the data, such as the NBER working paper. A news report that relays either one is a pointer to the primary source, and it should not replace it.

What These Numbers Mean For B2B Marketers

900 million weekly active users tells you how large the audience is. It does not tell you how many of those people ever see your brand in an answer, or whether any of them reach your site. Audience size and referral traffic are two separate measurements, and the first is no proof of the second.

So the number is a reason to measure, not a result. If a large share of your buyers ask AI tools questions before they ever open a search engine, you need to know whether your brand appears in those answers. Start with what AI search visibility is, then decide which AI search visibility metrics you will track.

Keep the NBER finding in mind as well. About 70% of ChatGPT usage was not work-related, so a big user count does not mean a big pool of people researching software at work. Your own data on who arrives and what they ask is worth more than any headline total.

Traffic from ChatGPT to a SaaS site is its own analysis, with its own method, and it is not covered by user counts at all. To work on the visibility side, read how to get cited in AI answers and look at the tools that monitor ChatGPT answers. If you want a team to run the tracking and the fixes, our AI search visibility service does exactly that.