Are You Behind on AI? What 1.2 Billion ChatGPT Users Actually Do
Is it true that everyone around you has already mastered artificial intelligence and is building agents — and how can you tell you're not falling behind if ChatGPT's 1.2 billion users mostly ask for advice, texts and information?
According to OpenAI's open Signals statistics, the three largest categories of personal ChatGPT messages are practical advice (about 32%), working with text (22%) and information seeking (19%), 73% together, while technical help, including programming, is only 4%: mass reach is not mass automation. Over the year the share of advice grew from 28% to 32% and working with text fell from 26% to 22%, and the chat is increasingly becoming a way to work through a situation rather than a generator of ready-made text. In the ranking of messages per inhabitant Singapore comes first and the US is in 51st place; Anthropic and Microsoft data show the same lead for Singapore and the Emirates, where banks and government services support adoption. China's seven hundred million generative AI users cannot simply be added to the 1.2 billion, the share of messages from people 35 and older has grown to 34%, and instead of comparing yourself with other people's demos Anna Volchek offers three questions — about the tasks you now solve with AI, about what has really got better and about checking that improvement.
What to watch for
Key takeaways
The company named this weekly ChatGPT audience at the end of September at its developer conference. The figure raises the questions "am I behind?" and "who are these people?": it seems everyone around already knows how to do what you haven't even started.
An acquaintance from her professional circle didn't know what Anthropic is, and a young man didn't understand the question of which model he uses. Meanwhile, videos show people building agents and AI-native companies; neither, Anna says, describes the world.
With 8.3 billion people on Earth, 1.2 billion is roughly one in seven. But a birthday greeting once a week and daily document automation count the same in the weekly audience.
In the Signals sample of three hundred thousand messages from personal accounts — without Codex and business products — practical advice takes up about 32%, working with text 22% and information seeking 19%.
The remaining 25% are conversations about oneself, communication and relationships, images and media; technical help, including programming and data analysis, is only 4%. Mass reach, Anna stresses, does not equal mass automation.
Perhaps the main mass change is not that 1.2 billion people learned to build agents but that millions gained an everyday way of dealing with understandable tasks. Whether decisions got better, the data do not show.
Over the year practical advice rose from 28% to 32%, working with text fell from 26% to 22%, and search barely changed. Anna's hypothesis: the chat is becoming a habitual way to work through a situation rather than a generator of ready-made text.
Singapore, Malta, the Netherlands and Montenegro lead the Signals ranking of ChatGPT messages per inhabitant in the second quarter of 2026. The ranking measures personal ChatGPT relative to population, not Americans' knowledge, corporate AI or other tools.
According to Anthropic, Singapore has less than 1% of global Claude usage, but relative to its working-age population its share is almost six times larger. According to Microsoft, about 19% of people aged 15 to 64 worldwide use AI, 73% in the Emirates, 64% in Singapore and around 33% and 32% in the US and Poland.
In Singapore the bank UOB, together with a government agency, helps small companies choose tasks for AI and build prototypes; in the Emirates AI is being built into government digital services. A high place in the ranking doesn't mean people are smarter, and a low one doesn't mean they aren't interested.
According to the China Internet Network Information Center, by mid-2026 China had more than seven hundred million active users of generative AI. They can't be added to the 1.2 billion: definitions and overlaps differ. ChatGPT is a huge part of the market, but not the whole market.
Two thirds of messages come from people aged 18 to 34, but the share of those 35+ grew from 28% to 34%, and of those 65+ from 1.9% to 2.5%. The notion that AI is a tool only for the very young is blurring.
Assess yourself not by the number of prompts, your age or your country but by three questions: which tasks do I now solve differently with AI, what has actually got better — speed, quality, depth, access — and can I check that the improvement happened.
What this episode is about
A solo ToTheMoon episode with Anna Volchek about the figure OpenAI announced at the end of September at its developer conference: 1.2 billion ChatGPT users every week. The host admits the figure impressed her and raises understandable questions — am I behind, who are these people, do I use the chat the way everyone else does. Two recent conversations — an acquaintance asked what Anthropic is, and a young man didn't understand a question about models — and tech videos about agents and AI-native companies equally fail to describe the world.
About 8.3 billion people on Earth give a ratio of roughly one in seven, but the weekly audience counts equally someone who once asked for a birthday greeting and someone who automates documents and writes code every day. So Anna looks at what happens after a person opens the chat.
OpenAI's open Signals statistics — the August update, data as of June 2026, a sample of three hundred thousand messages from personal accounts without Codex and business products — show that practical advice takes up about 32% of messages, working with text 22% and information seeking 19%, 73% together. The remaining 25% are conversations about oneself and relationships, images and media; technical help, including programming, is 4%. Mass reach is not mass automation: the main change may be a new everyday way of dealing with tasks, and over the year the share of advice grew from 28% to 32% while working with text fell from 26% to 22%.
Geography is uneven: by ChatGPT messages per inhabitant in the second quarter of 2026 Singapore comes first, followed by Malta, the Netherlands and Montenegro, while the US is in 51st place. Anthropic's data on Claude and Cowork and a Microsoft study across AI services show the same lead for Singapore and the Emirates; Anna separates three measurements — the intensity of messages, a country's relative activity and how widespread use is among the population — and gives examples of support for adoption: the bank UOB in Singapore and government digital services in the Emirates.
1.2 billion is not the whole world of AI: there are Gemini, DeepSeek, Qwen, corporate solutions and AI features inside work programs, and in China, according to the China Internet Network Information Center, by mid-2026 there were more than seven hundred million active users of generative AI who cannot simply be added to ChatGPT's audience. The share of messages from people 35 and older grew from 28% to 34%, and from people 65+ from 1.9% to 2.5%.
The finale offers neither a reassuring answer nor a frightening conclusion: the analytics don't show whether people have started working and living better, so you should assess yourself not by the number of prompts, your age or your country but by three questions — which tasks do I now solve differently with AI, what has really got better and can I check it. Anna asks viewers to share concrete examples.
The episode strips the figure of 1.2 billion users of the feeling of a universal race: behind it are mostly advice, texts and search, uneven geography and data that don't show whether anyone has started working and living better. The value of the analysis lies in telling the measurements apart: the weekly reach of one product, the intensity of messages, the share of the population and AI use beyond ChatGPT answer different questions. Anna Volchek's three questions move the assessment of progress from other people's demos to your own tasks and a result you can check.
Episode transcript
The episode is in Russian; below is an English reading guide to the transcript (the full EN transcript is a machine translation). Voice matching applied to 34 segments: 34 identified, 0 mixed, 0 probable, and 0 unresolved.
Read transcript on a separate page
Loading…