Population census · Generative AI · Early 2026

8.2 billion people. How many have actually touched AI?

A corrected, sourced remake of a waffle chart that went viral on X in early 2026, claiming that 84% of humanity has never used AI. The answer depends entirely on how you define "used AI" — so this version lets you switch definitions, and it magnifies the populations too small for the original to even render.

100× magnifier

The part the original chart couldn't show you

At ~3.3 million people per dot, paying subscribers and agentic-coding users round to a few barely-visible squares. Below, each dot is 1 million people — the same populations, magnified until they become legible.

Where every number comes from

Full disclosure of the seams. Three kinds of numbers appear on this page — and only one kind is a direct measurement.

Vendor-reported Author estimate Original chart's claim

Raw inputs — vendor-reported, not independently audited

50M paying subs
OpenAI, announced Feb 27, 2026 alongside a $110B funding round. Includes Plus/Pro/Team tiers; announced with fundraising, so read with that incentive in mind. TechCrunch
~20M paid seats
Microsoft 365 Copilot paid seats, Microsoft Q3 FY2026 earnings. Seats ≠ active individuals.
8M+ paid seats
Gemini Enterprise across 2,800+ companies, Alphabet earnings / Google disclosures.
~1.2B / 1.1B / 900M / 420M / 245M MAU
Meta AI (Zuckerberg statement) / ChatGPT app (Sensor Tower via Reuters) / Gemini app (Google I/O 2026) / Copilot (Microsoft) / Claude (Sensor Tower). Different companies, different counting methods, same label "MAU."
~2–2.5B monthly
Reach of Gemini-powered AI Overviews in Google Search, per Google. "Saw an AI summary" — the loosest usage definition on this page.
1.6M → 5M weekly
OpenAI Codex users, early 2026 → June 2026, per OpenAI.
1 in 6 people
Microsoft estimate of worldwide genAI use in H2 2025 ≈ 1.35B people. Closest thing to an independent global figure, but still one company's model.
8.23B people
UN world population estimate for 2026. The only number here from a non-tech-company source.

Author estimates — my arithmetic on top of those inputs

~1.8B chatbot users
No source publishes a deduplicated total; I constructed it. 1.2B Meta AI + 1.1B ChatGPT + 0.9B Gemini + 0.42B Copilot + 0.25B Claude ≈ 3.9B raw ≈ 1.8B after ~55% overlap discount The overlap discount is a judgment call, informed by cross-usage data (e.g. ~66% of Claude users also use ChatGPT) and Meta AI counting anyone who touched the feature in-app. Honest range: 1.5–2.2B. Could be off by ±25%.
~90M paying
50M OpenAI + 20M Copilot seats + 8M Gemini Ent. + ~15M others (Claude, Gemini AI Pro, Perplexity, Cursor…) − ~5M overlap ≈ 88M The "others" term is the softest part — vendors don't break these out. Range: 80–110M. Even the floor of this range is 3× the original chart's ceiling.
~8M agentic coding
5M Codex (June) + Claude Code + Cursor agents + Copilot agent mode − heavy overlap (most use several) ≈ 8M Only Codex has a public figure; the rest is inference. Range: 5–15M. Doubling every few months, so stale on arrival.
~3.5B "touched genAI"
2.5B AI Overviews reach + 1.8B chatbot users − large overlap + non-Google surfaces ≈ 3.5B Lowest-confidence number on the page. The overlap between "searches Google" and "uses a chatbot" is huge but unmeasured. Range: 3–4B. Treat as an illustration of the definition problem, not a statistic.
Grey remainders
6.33B and 4.63B "haven't used" figures are just population − everything else. They inherit every error above.

What would falsify these estimates

Dedup total
A survey-based global study (Pew-style, but worldwide) finding monthly chatbot use materially outside 18–27% of adults would break scenario ②.
Paid total
Any vendor restating subscriber counts downward, or evidence that OpenAI's 50M includes large numbers of trial/comped accounts, would pull ~90M toward ~60M.

Fact-checking the original, claim by claim

Plausible

"Free chatbot user: ~1.3B (16%)"

Matches Microsoft's estimate that roughly 1 in 6 people worldwide used a generative AI tool in H2 2025. But it's the floor, not the ceiling: monthly actives reported for early–mid 2026 include Meta AI ~1.2B, ChatGPT ~1.0–1.1B, Gemini app 750–900M, Copilot ~420M, and Claude ~245M. Even with heavy overlap between them, deduplicated monthly users plausibly reach 1.6–2.2B.

Wrong ~4×

"Pays $20/mo for AI: ~15–25M (0.3%)"

Contradicted by primary sources for the very month depicted. OpenAI alone announced 50 million paying subscribers on Feb 27, 2026. Add ~20M paid Microsoft 365 Copilot seats, 8M+ Gemini Enterprise seats, Claude subscribers, and 9M+ paying ChatGPT business users. A defensible total is ~80–110M people paying for AI — roughly 1% of humanity, not 0.3%.

Stale

"Uses coding scaffold: ~2–5M (0.04%)"

Roughly right for Feb 2026 if narrowly defined as agentic coding tools (Codex was at ~1.6M weekly users early in the year). Already outdated: Codex alone crossed 5M weekly users by June 2026, and that excludes Claude Code, Cursor, and GitHub Copilot's tens of millions of users under a broader definition. This number roughly doubles every few months.

Definitional

"Never used AI: ~6.8B (84%)"

Only true under the narrowest definition (never deliberately opened a chatbot). Google's AI Overviews alone reach ~2.5B people monthly, and Meta AI sits inside WhatsApp, Instagram, and Facebook. Count anyone who has received generative-AI output and "never touched AI" shrinks toward ~55–60% — still a majority of humanity, but a very different chart.

Minor

"8.1 billion humans"

Slightly low; UN estimates put world population at ~8.2B in 2026. Changes nothing material.

Sources & caveats

  1. The original viral chart this page fact-checks: X post, Feb 2026 — "each dot is 3.2 million people."
  2. OpenAI via TechCrunch, Feb 27 2026 — 900M weekly active users; 50M paying subscribers. techcrunch.com
  3. Google I/O 2026 — Gemini app 900M+ MAU; AI Overviews ~2–2.5B monthly reach.
  4. Microsoft Q3 FY26 — Copilot ~420M MAU; 20M+ paid M365 Copilot seats; "1 in 6 people worldwide used genAI, H2 2025."
  5. Meta (Zuckerberg) — Meta AI ~1–1.2B MAU across apps.
  6. Sensor Tower / Similarweb, May–June 2026 — Claude ~245M MAU; ChatGPT app 1B+ MAU.
  7. OpenAI, June 2026 — Codex 5M weekly users, up from 2M in April.
  8. Caveat 1: nearly all user counts are vendor-stated, not independently audited, and often announced alongside fundraising.
  9. Caveat 2: WAU, MAU, "seats," and "ever used" are different units; cross-platform overlap means totals can't simply be summed. Ranges here are deliberately wide.