OpenAI
ChatGPT users active in a week
- 8 Dec 2025
- Over 800 million
- 27 Feb 2026
- Over 900 million
These are thresholds OpenAI reported on each date. They give a sense of scale and leave the exact counts open.
AI in everyday life
A business can adopt AI to read invoices. A person can try a chatbot once. Both count as adoption in different studies. Start with businesses, then see what changes when you count people.
By 2025, one in five businesses covered by Eurostat’s survey used at least one AI technology. These include tools for recognising speech, analysing text and automating work.
| Year | Share |
|---|---|
| 2021 | 7.7% |
| 2022 | No figure |
| 2023 | 8.1% |
| 2024 | 13.5% |
| 2025 | 20.0% |
Two changes to read alongside the bars. Eurostat did not ask its AI questions in 2022. In 2025 it added AI-generated images, video and audio, and explicitly mentioned generating code. The final rise can reflect both a change in use and a change in the question.
Link to this chartEurostat figures and survey scope, 11 Dec 2025 · Eurostat questionnaire history, §2.3.1 (PDF)
An editor wants to describe the gap in the EU chart. Choose the sentence you would publish.
Look at the label beside the dashed line. A zero would be a measured result. Eurostat did not ask the question that year.
Supported. There are figures on either side of the gap, but the survey leaves that year open. Even a plausible estimate would need to be labelled as an estimate.
Microsoft estimates how many people use a generative AI product. Its worldwide estimate rose from 15.1% in the first half of 2025 to 18.8% in the quarter ending June 2026.
| Period | Share |
|---|---|
| Jan–Jun 2025 | 15.1% |
| Jul–Dec 2025 | 16.3% |
| Jan–Mar 2026 | 17.8% |
| Apr–Jun 2026 | 18.8% |
The first two estimates come from half-year reports and the last two from quarterly reports. Each bar represents its labelled period. Their spacing does not measure the speed of growth.
Link to this chart| Place | Share |
|---|---|
| Worldwide | 18.8% |
| Singapore | 64.3% |
The worldwide estimate includes Singapore. Using the same age range, reporting period and measurement method lets us compare the shares side by side.
Link to this chartMicrosoft combines anonymous usage signals and adjusts for differences in devices, operating systems, internet access and population. Think of it as estimating the crowd from the entrances you can observe. The tools Microsoft can see affect the estimate. The figures tell us about use during a period. They cannot tell us how often or how skilfully someone used a tool. Read Microsoft’s explanation of its measure.
Microsoft AI Economy Institute · 2025 estimates · First quarter 2026 · Second quarter 2026 and method
You are writing about the quarter ending June 2026. Choose the headline you would publish.
Supported. The chart names the population, the type of tool and the period. The word “estimated” keeps the method visible.
This widens the claim twice. The estimate covers ages 15–64 and tracks generative AI products. It cannot establish the share for every age or every kind of AI.
The age group fits, but “every day” adds a frequency the chart never measures. Someone who used a tool during the reporting period can count.
Each company reported the figures below on different dates. Read each pair on its own. Where it reports a threshold such as “over 800 million”, the figures cannot tell you the exact increase. A person can use several services, an organisation can have many employees, and a month gives a user longer to appear in a count than a week.
ChatGPT users active in a week
These are thresholds OpenAI reported on each date. They give a sense of scale and leave the exact counts open.
Business customers
Anthropic described the earlier figure as two years before its announcement. One business customer may represent many users.
Gemini app users active in a month
Alphabet gave these figures for the Gemini app. Someone who saw an AI-generated answer in Google Search belongs to a different measurement.
Registered users who used a Grok AI feature in the 30 days ending on each date
SpaceX counted feature use across Grok and X. It used sign-in activity to identify people using both platforms and acknowledged that some duplicates may remain.
Try reading the evidence
No. Imagine one person using ChatGPT at home and Gemini at work. Adding their activity counts them twice. If their employer also buys Claude, Anthropic counts the organisation as a business customer. The disclosures cover different dates and windows. To estimate people worldwide, start with the population estimate and check who it covers.
Practise checking an AI answer · Explore more lessons · Return to release trends