You’ve probably seen the graphic. A grid of 2,500 dots - each representing 3.2 million people - showing how much of humanity uses AI. A massive grey sea of “never used AI” (84%), a modest green band of free users (16%), a tiny yellow sliver of paid users (0.3%), and a single red dot of developers (0.04%).
It went massively viral. Created by AI consultant Damian Player on X, reshared across LinkedIn, Threads, and every other platform. The message was clear: almost nobody is using AI. You’re early. The opportunity is enormous.
And as someone who does an AI news roundup every week and works with data daily, my first reaction was: these numbers feel off.
So I did something that apparently very few of the people resharing it bothered to do. I checked the sources.
What the graphic claims
The four-tier breakdown goes like this:
🔴 0.04% (~2-5M) use coding scaffolds / build with AI
🟡 0.3% (~15-25M) pay $20/month for AI
🟢 16% (~1.3B) are free chatbot users
⬜ 84% (~6.8B) have “never used AI”
The 16% comes from the Microsoft AI Economy Institute’s “Global AI Adoption in 2025” report, published January 8, 2026. The other three tiers are Player’s own estimates assembled from public subscriber counts. No formal methodology exists for the lower segments.
The Microsoft source is legitimate. The rest? Let’s look.
Problem 1: The paid user number is 3-5x too low
This is the easiest claim to debunk because we have official company disclosures.
OpenAI, on February 27, 2026, announced: 50 million consumer subscribers + 9 million paying business users. That’s 59 million paying users from one company. The graphic’s entire yellow category is 15-25 million.
Microsoft, during their FY26 Q2 earnings call on January 28, 2026, disclosed 15 million paid Microsoft 365 Copilot seats.
Google’s Gemini AI Premium has attracted “millions” of subscribers (their word - no exact number published). Anthropic’s Claude doesn’t disclose subscriber counts either, but revenue estimates suggest millions more.
Now, you can’t simply add these numbers together. Power users hold multiple subscriptions - I personally pay for three. So you need a discount for overlap, maybe 25-30%. But even with that adjustment, you land around 60-70 million unique paying humans. Not 15-25 million.
The graphic’s paid estimate may have been roughly accurate for mid-2024. By early 2026, it’s embarrassingly outdated.
Problem 2: The developer number is about half the reality
The graphic claims 2-5 million people use AI coding tools.
GitHub Copilot alone has 4.7 million paid subscribers (Microsoft earnings, January 2026). Add Cursor (1M+ daily active users), OpenAI Codex (1.6M weekly users), Claude Code, Windsurf, and others. There’s heavy overlap - many developers use 2-3 tools simultaneously. But even accounting for that, 6-10 million unique paid developer users is more realistic.
Problem 3: The Microsoft 16.3% - what it actually measures
The Microsoft report is the strongest data source in the graphic. But it’s worth understanding what it actually measures:
It tracks visits to generative AI tools (ChatGPT, Gemini, Claude, DeepSeek, Midjourney, and others - not just Microsoft products) using Windows telemetry
It then scales for other operating systems and device types
The 16.3% is expressed as a share of working-age population (ages 15-64), not total population
Here’s where the graphic introduces a subtle error. It applies the 16.3% to total world population (8.1 billion) to get 1.3 billion users. But the Microsoft methodology uses working-age population (~5.3 billion) as the denominator. If you apply 16.3% correctly, you get ~864 million, not 1.3 billion.
AI Diffusion by Economy H2 2025 (Microsoft Report)
Problem 4: “84% never used AI” hides two completely different realities
This is the most important issue.
2.2 billion people don’t have internet access at all (ITU Facts & Figures 2025). They haven’t used AI, true. But they also haven’t used email, online banking, or YouTube. This is an infrastructure and electrification crisis, not an AI adoption story.
That leaves roughly 3.9 billion people who ARE online but haven’t opened a chatbot. The graphic labels them identically to the offline population: grey dots, “never used AI.”
But when survey companies actually ask connected populations, the picture looks different:
Ipsos/Google, 21,000 people across 21 countries, January 2026: 62% have used an AI chatbot (up from 48% in 2024)
KPMG/University of Melbourne, 48,000 people across 47 countries, 2025: 66% intentionally use AI
Eurostat, official EU statistics, December 2025: 32.7% used gen AI in the last 3 months alone
Estonia: 46.6% 🇪🇪
Use of GenAI tools (Eurostat)
Survey rates are high everywhere. And although these surveys only reach online adults, they show the % who is actively using AI, not the ones who have tried, but stopped (for whatever reasons). Microsoft's low country-level figures (Ethiopia 6.8%, Nigeria 9.3%) reflect that most people in those countries simply don't have internet access, not that connected Africans aren't using AI.
So you also can’t just label all of them “never used AI” and color them grey.
The corrected picture
Based on cross-referencing ITU connectivity data, Microsoft telemetry, platform-reported user numbers, and multiple independent surveys, here’s a more honest breakdown:
I built an interactive version of this visualization where you can toggle between the original graphic and the corrected one, and tap each segment to see the data source and confidence level.
Let me be direct: my corrected version isn’t 100% right either.
The “online, never tried gen AI” (~3.9B) is the least certain number in the whole analysis. It’s a residual - what’s left over after subtracting estimated users from the online population. Nobody has measured this directly at a global scale. The surveys cover 21-47 countries, mostly richer ones, and only ask online adults.
The overlap coefficient for paid users is an estimate. The trial-vs-regular split is derived, not measured. I’m not a researcher with a team and a budget. I’m an AI trainer who reads too many reports.
But that’s the point
A single graphic with four clean color categories got millions of impressions across LinkedIn, X, and Threads. Smart people - executives, investors, consultants - reshared it without checking a single source. Nobody asked “wait, is 15-25M paid users still accurate when OpenAI alone has 59M?” or “how exactly was the 84% calculated?”
This happens in finance too. A clean chart with a clear narrative gets accepted faster than a messy spreadsheet with caveats. A single metric from one report becomes “the truth” the moment someone designs a pretty visualization around it.
Data is powerful. Data with good design is even more powerful. But data without context, without methodology notes, without confidence intervals - that’s just a story with numbers attached.
The world is messier than four colors. And understanding the mess - the sources, the definitions, the limitations - is where the actual value lives.
Gerlyn Tiigemäe is an AI implementation consultant and trainer for finance teams, and host of the AI Powerment Podcast. She does a weekly AI news roundup in LinkedIn and shares her thoughts in Substack. Visit gerlyntiigemae.com for contact.
Sources referenced in this analysis: