← All Reports
AI Field Notes · AI Use Demographics · September 2026

Who uses AI in 2026?

In the United States, just over half of everyone, children included, has tried AI, and one in three uses it every week. Worldwide, about a third have tried it and one in seven uses it weekly. The two pull apart hardest at the top: 8% of Americans pay for AI against 1% of the world, and 3% of Americans build with an AI coding agent against 0.4% of the world.

These charts count everyone: 349 million Americans and 8.3 billion people on Earth, babies and grandparents included, on the United Nations' count. That is why the "never used" share looks so large. Adults-only numbers run higher. Pew found 49% of American adults using an AI chatbot in February 2026, the St. Louis Fed found 62% of adults aged 18 to 64 using generative AI by May, and Common Sense Media found 86% of kids aged 9 to 17 using AI. The people who have never knowingly used it are mostly under 9, over 65, rural, or, worldwide, offline altogether. About 2.2 billion people still have no internet.

Everyone lands in exactly one of five groups, in rising order of how deep they are in. Two of the groups overlap on purpose in the captions: "tried" means anyone who has ever prompted a model, and "weekly" means anyone who uses one at least once a week, whether they pay or not.

Never knowingly used AIHas not prompted a chatbot, image tool or agent. AI hidden inside search or a social feed does not count.
Tried it, casual or lapsedHas used one, but less than weekly, or stopped.
Uses a free model weeklyA regular user who does not pay.
Pays for AIA subscription to ChatGPT, Gemini, Claude, Copilot or similar.
Builds with an AI coding agentCursor, Claude Code, Codex, Copilot's agent mode, Lovable, Replit. Half are not professional programmers.

Every square is 3 million people. Each grid is filled from the top left, never-used first, then each group in order. Press play, or drag the year, and the grid and the chart beside it move together. Tap the timeline to jump to a year.

Who to show. Beside each title there is a switch for men or women and a slider for the youngest age to include, from all ages up to 70 and older. Move either one and the grid, the chart, the table and the population count all redraw for that slice of people. Everyone, all ages, is where it starts.

The World

All ages
2026

One square = 3 million people

Share of everyone, 2022 to 2026

Each column is 100% of the world's population that year. Gray is the people who have never used AI. Hover or tap for the numbers.

The numbers behind the charts · World

The United States

All ages
2026

One square = 3 million people

Share of everyone, 2022 to 2026

Each column is 100% of the US population that year. Gray is the people who have never used AI. Hover or tap for the numbers.

The numbers behind the charts · United States

What the numbers say

Four things worth knowing

1. Trying spread like a consumer app. Paying spread like business software. Building spread like a new craft.

Play either chart and the shape is the same: the gray shrinks fast, the blue and green swell, and the amber and pink stay thin. In the United States the share of people who have tried AI went from 2.5% at the end of 2022 to 53% today. The share who pay went from almost nobody to 11%, builders included. The share who build with a coding agent, a category that did not exist in 2022, is 3%, about 10 million people, and roughly half of them are not professional programmers.

Worldwide the same three curves run two to three years behind. About a third of humanity has tried AI, one in seven uses it weekly, one in a hundred pays, and four in a thousand build. The world is catching up on free, phone-based use. It is not catching up on $20 subscriptions or coding agents.

2. The United States leads on money and building, and trails on trying.

Two big 2026 rankings measure who uses AI, and they place the United States in very different spots. Microsoft's AI Diffusion Report (May 2026) uses device telemetry across 147 economies to estimate the share of people aged 15 to 64 who used a generative AI product in the first quarter of 2026. On that measure the United States is 21st, at 31.3%, behind the United Arab Emirates (70.1%), Singapore (63.4%), and most of Western Europe.

Google and Ipsos asked online adults in 21 countries whether they had used generative AI in the past year. There the United States came last of the 21, at 40%. Americans are also unusually wary: Pew finds 52% of adults more concerned than excited about AI, up from 37% in 2021.

Yet the United States is where the money is. Roughly 20 million Americans pay OpenAI alone, and the country's paying share (8% of everyone, not counting builders) is eight times the world's 1%. Its builder share is seven times the world's. Americans try AI at European rates and pay for it at rates nobody else matches.

3. The countries using AI the most

Two rankings, two answers, because they count different people. Microsoft counts everyone of working age, online or not. Ipsos asked only people who are already online.

Top 10 · Microsoft, 2026

People aged 15 to 64 who used generative AI, by device data

#CountryShare
1United Arab Emirates70.1%
2Singapore63.4%
3Norway48.6%
4Ireland48.4%
5France47.8%
6Spain44.2%
7New Zealand43.0%
8United Kingdom42.2%
9Netherlands42.1%
10Qatar41.8%
21United States31.3%

Top 10 · Google/Ipsos, 2026

Online adults who said they used generative AI last year

#CountryShare
1United Arab Emirates85%
2Nigeria84%
3India83%
4Brazil74%
5Singapore71%
6South Africa70%
7Mexico66%
8Argentina65%
9Spain64%
9South Korea64%
21United States40%

Nigeria, India and Brazil rank high in the survey and low in the telemetry for the same reason: their internet users are young and enthusiastic, but a large share of their populations is not online at all. Both lists are true. The Gulf states and Singapore top both, small rich countries that put AI into government services early. China is a special case: its own internet authority counts 602 million generative AI users, 42.8% of the country, almost all on Chinese models such as DeepSeek, Doubao and Qwen, since Western tools are blocked there.

4. The countries using AI the least

At the bottom of Microsoft's table are Cambodia (5.7%), Afghanistan, Tajikistan and Turkmenistan (6.1% each), Cuba (6.7%), then a band of East African and South Asian countries at 7 to 8%: Ethiopia, Somalia, Sudan, South Sudan, Uganda, Tanzania, Bangladesh and Laos. Russia and Ukraine sit near 9.5%, Nigeria at 10%, Pakistan at 11%, Indonesia at 14%, and India at 17.6%.

The reason is mostly connection, not interest. The ITU counts 2.2 billion people still offline. Only 36% of Africa is online, against 88 to 93% of Europe and the Americas, and in low-income countries 23% of people use the internet at all. Where people are online, the appetite is often stronger than in the rich world: 84% of online Nigerians had used generative AI, and the survey found people in emerging markets far more excited than worried about it. Microsoft's own split makes the gap plain: 27.5% of working-age people in the Global North used generative AI in early 2026 against 15.4% in the Global South, and the gap grew during the quarter.

Who, inside a country

How these numbers were built

No single survey asks these exact questions of an entire population, so the five groups are estimates assembled from named sources and cross-checked against each other: Pew Research Center (February 2026, 5,119 adults), the St. Louis Fed's Real-Time Population Survey (May 2026), SSRS, Gallup, Common Sense Media (June 2026), Microsoft's AI Diffusion Report (Q1 2026), DataReportal, OpenAI's own disclosures (900 million weekly users and 50 million paying subscribers, February 2026), Menlo Ventures, Parks Associates, PNC and Bank of America card data, China's CNNIC, the ITU and the UN's population estimates. Adult and household survey results were converted to whole-population shares, double-counting across subscriptions and coding tools was removed, and 2022 figures were reconstructed from launch-era data. Confidence is highest for the American "tried" and "weekly" figures and lowest for the number of builders worldwide, where vendors report seats, not people. Year-end figures; 2026 is year to date as of September.

How the men, women and age slices were built

Every population number on this page, including every slice, is the United Nations' own count of how many people of each single year of age and each sex are alive that year. One source, the same method for the United States as for the world, so the slices always add back up to the whole. It is a little higher than the US Census Bureau's own estimate for the United States, and it is used here for consistency rather than because it is the only count.

The gaps between groups are measured, not assumed. Pew asked 5,119 American adults in February 2026 and found chatbot use falling with age: 66% of 18 to 29 year olds, 61% of 30 to 49, 42% of 50 to 64 and 23% of 65 and over, with daily use peaking in the thirties and forties. Pew also found the overall gap between American men and women has closed to 50% against 47%, while daily use has not: 27% of men against 20% of women. Common Sense Media supplies the under-18s, at 86% of 9 to 17 year olds. For the world, the ITU's Facts and Figures 2025 gives the access gap the rest of it rests on: 82% of people aged 15 to 24 use the internet against 72% of everybody else, and women's internet use worldwide runs at 0.92 of men's. The builder group is anchored to the Bureau of Labor Statistics count of who writes software for a living, where women are 20.3% of software developers, and to Stack Overflow's 2025 age profile of developers.

Those measured rates are applied age band by age band and then scaled by one constant for each group, so that everyone, all ages, reproduces exactly the totals in the tables above. Two consequences are worth stating plainly. The scaling lifts each band a little, because the surveys ask about chatbots while this page counts anyone who has ever prompted a model. And the shape of the age and sex gaps is the 2026 shape held steady across the earlier years, since nobody measured those gaps in 2022; only the totals move as you drag the year back.

Provenance

This page began as three separate deep-research projects. Grok, Gemini and Claude were each handed the same question: who uses AI in the United States and in the world, how many of them, and how that has changed since 2022. The three came back with three different answers. They counted different populations, read the same surveys in different ways, and each carried a few claims that traced back to a single statistics website. Claude then audited the three reports against each other, went back to the primary sources to settle every disagreement, dropped what could not be verified, and assembled the one set of figures used here. The charts, the tables and this essay were built from that synthesized report on September 10, 2026.

← All Reports