Adoption, measured honestly
Not provider hype. Probability-sampled surveys, each strand labeled with exactly what it asks.
- 800M weekly ChatGPT users, the largest disclosed AI user count OpenAI · Oct 2025
- 44% of US adults have used ChatGPT Pew · Feb 2026
- 61% tried · 34% weekly generative AI across six rich countries Reuters Institute · 2025
- ≈18% of US businesses use AI, up from under 4% in 2023 Census BTOS · 2026
- 20% of EU enterprises use AI; Denmark leads at 42% Eurostat · 2025
- ≈1 in 10 humans has a weekly AI habit floor estimate · method below
There is no single "AI adoption number." There are strands, and each answers a different question. Tried it once? The rich world crossed half in 2025. Weekly habit? A third. In the US, ChatGPT alone has gone from 18% of adults to 44% in three years. That is the fastest consumer technology diffusion Pew has ever polled.
Sources: Pew Research Center (Feb 2026) · Reuters Institute Digital News / Gen AI Report 2025 · Google · Ipsos "Our Life With AI" 2025 · provider milestones per company disclosures. Pulled 2026-07-15.
Every state. Every country. One curve.
The explorer
Pick up to four geographies and scrub six years. US states come from the Census Bureau's biweekly business survey; European countries from Eurostat's annual one. The frames differ. Census asks all firms about the last two weeks. Eurostat asks enterprises with 10+ staff. So compare trajectories, not decimal points.
Sources: Census Bureau BTOS (US + states, share of all businesses using AI in the last two weeks, biweekly since Sep 2023; lines show a 3-release rolling average, faint dots are the raw readings, suppressed cells omitted) · Eurostat isoc_eb_ai (countries, share of enterprises 10+ using any AI technology; solid dots are the annual waves 2021 · 2023 · 2024 · 2025, dashes interpolate between them, dotted tails carry the last wave forward). The Businesses scale multiplies each share by the geography’s business base. US states use Census CBP establishments (2023, all sizes); European countries use Eurostat SBS enterprises 10+ (2024). That base is held fixed across the timeline, so counts are estimates. Download: btos-state-series.csv · eurostat-country-series.csv · geo-business-base.csv.
Every engine's own scoreboard
The same engines you can seat in your harness, on their own disclosed milestones. Read the metric tags carefully. These numbers cannot be added together. The same person shows up in three of them. Weekly users, monthly users, downloads, revenue are different animals.
● company disclosure ○ press / third-party estimate · WAU weekly · MAU monthly · DL cumulative downloads · ARR annualized revenue
Adoption has a zip code
Businesses, by state and by country. From the two official statistical agencies that actually ask.
The US Census Bureau asks businesses every two weeks whether they use AI to produce goods or services. In late 2023 the answer was under 4%. By early 2026 it is roughly one in five. And the spread between states is nearly 2×. Europe's official count tells the same story at the same altitude: 20% of enterprises, with an 8× spread between Denmark and Romania.
US businesses using AI, by state
Census BTOS, avg of Jan–Mar 2026 releases
EU enterprises using AI, by country
Eurostat isoc_eb_ai, 2025 (10+ employees)
Intensity is a different map again: on Anthropic's Economic Index, Washington DC uses Claude at 3.8× its population share, the highest of any US state-level geography. Israel and Singapore lead countries. Enterprise surveys run far hotter than official counts (Stanford HAI: 88% of organizations use AI in at least one function; McKinsey: 71% for generative AI) because they ask bigger firms a broader question.
The doors still haven't opened
The gap
≈ 9 in 10 humans
have no weekly AI habit.
ChatGPT's 800 million weekly users is the biggest number any provider has ever disclosed. And it is under 10% of the planet. Trial is mainstream in rich countries. Habit is not. Paying is 50-odd million seats. Building on it is a village.
Right now, getting in is cheap, and it is cheap for a reason. Every frontier plan is subsidized today, roughly $100 of compute sold to you for one. Same move Uber made: half price across the bridge until the habit sets, then the premium lands. When the subsidy lifts, one question decides who comes out ahead. How is your org deploying this, and who is the benefactor?
Adoption curves reward whoever is ready before the crowd arrives. The model layer is rented and it changes monthly. The context layer is the part you own. Your memory, your files, your harness. It compounds while everyone else is still trying the free tier.
The numbers, sourced
- Never sum provider users. ChatGPT's 800M + Gemini's 750M + Copilot's 20M overlap heavily. One person is in several counts. Only surveys estimate unduplicated people.
- Metrics differ: weekly users ≠ monthly users ≠ cumulative downloads ≠ paid seats ≠ revenue. Each engine row above is tagged with its own metric.
- Survey wording drives the number: "ever used" (61%) vs "weekly" (34%) vs "used ChatGPT" (44%) are different questions. Strands are never spliced.
- Company figures are unaudited and usually disclosed at launches and earnings. All are dated to the disclosure.
- The dot grid is a model: weekly tier is a floor (ChatGPT alone); "tried it" extrapolates surveyed ratios to online adults; paying and building tiers are order-of-magnitude seat counts.
Questions people actually ask
How many people use ChatGPT?
OpenAI reported 800 million weekly active users in October 2025. That is the largest disclosed figure for any AI product. In the US, 44% of adults told Pew in February 2026 they have used it, up from 18% in 2023.
Which US state uses AI most?
Colorado, at 23.2% of businesses, with Arizona (22.9%) and Washington DC (22.5%) close behind; West Virginia is lowest at 10.8%. On per-capita usage intensity of Claude, DC is in a league of its own at 3.8× its population share.
Can you add ChatGPT and Gemini users for a total?
No. The counts overlap (same people, several tools), they are self-reported, and they mix incompatible metrics: weekly vs monthly users vs downloads vs seats. That is why this page anchors its index on population surveys and treats provider numbers as dated milestones.
Early is a fact. Ready is a choice.
Nine in ten humans haven't built the habit yet. The ones who compound will be the ones whose context was ready before the crowd showed up. The engine is rented. The harness is yours.