That Viral AI Dot Chart Got Its Numbers Wrong. Here's What the Data Actually Shows.

A fact-check of the "Each dot is ~3.2 million people" visualization, with corrected figures and sources.

---

You've probably seen it by now — a grid of 2,500 dots where each represents 3.2 million humans. A sea of grey ("never used AI"), a sliver of green ("free chatbot users"), and a barely visible speck of gold and red for paying subscribers and coding tool users. The message is clear: almost nobody actually uses AI.

The visualization went viral because it tells a compelling story. And the core of that story is true — roughly 84% of the world's population has never intentionally used a generative AI tool. But several of its specific claims are significantly wrong, and it's missing the most important dimension entirely.

I decided to fact-check every claim against the best available data. Here's what I found.

What it gets right

The headline number holds up well. Microsoft's AI Economy Institute published its AI Diffusion Report in January 2026, based on aggregated telemetry data adjusted for device market share, internet penetration, and population differences across countries. Their finding: 16.3% of the global population used a generative AI product in the second half of 2025. That puts the "never used AI" share at 83.7% — essentially what the visualization claims.

The dot-grid format itself is genuinely effective. Seeing 2,500 dots where only the bottom few rows change color drives the point home viscerally in a way that percentages alone cannot.

What it gets wrong

Paid subscribers: undercounted by 3–4×. The visualization claims 15–25 million people pay $20/month for AI. In reality, ChatGPT alone had approximately 35 million paying subscribers as of July 2025, according to Reuters. Add Claude Pro, Google AI Pro, Perplexity Pro, Grok, and others, and the real number is likely 50–80 million — and prices range from $8 to $300 per month, not a flat $20.

Coding tool users: undercounted by 5–8×. The claim of 2–5 million "coding scaffold" users doesn't match publicly reported data at all. GitHub Copilot alone crossed 20 million all-time users by July 2025. Cursor surpassed $500 million in annual recurring revenue with over a million daily users. Stack Overflow's 2025 developer survey found 84% of developers now use or plan to use AI tools. The corrected figure is closer to 15–25 million active users.

The "free chatbot user" label is misleading. The 16% figure from Microsoft includes all generative AI users — free and paid, chatbot and non-chatbot alike. Calling this entire group "free chatbot users" obscures the fact that it covers everyone from casual ChatGPT visitors to paying Claude Pro subscribers to developers running code through AI assistants.

No sources are cited. For a data visualization making specific quantitative claims, this is a significant omission — especially when several of those claims are wrong.

What it completely misses: the hidden billions

Here's the biggest gap. The visualization treats "using AI" as a conscious choice — you go to ChatGPT or Claude and type a prompt. But billions of people now interact with AI without making that choice at all.

Google's AI Overviews — the AI-generated summaries that appear above search results — now reach over 2 billion monthly users across more than 200 countries and 40 languages, as reported during Alphabet's Q2 2025 earnings call. Meta AI, embedded into WhatsApp, Instagram, and Facebook, hit 1 billion monthly active users by early 2025.

Many of these users don't know they're "using AI." They're just Googling something or messaging on WhatsApp. But AI is shaping their information, their search results, and their conversations.

If we count embedded AI, the "never touched by AI" share drops from 84% to potentially 57–67% of the world's population. That's a fundamentally different story.

The story that matters most: the digital divide

The visualization presents a global average, which hides what may be the most important pattern in AI adoption: the gap between rich and poor countries is not closing — it's widening.

Microsoft's data shows that in the Global North, 24.7% of the working-age population used generative AI by the end of 2025. In the Global South, just 14.1%. That gap grew from 9.8 to 10.6 percentage points in six months. All ten countries with the largest adoption gains were high-income economies.

The country-level variation is staggering. The UAE leads at 64% adoption. Singapore follows at 61%. Australia sits at roughly 37%, ranking 11th globally — more than double the world average and well above the United States at 28.3%. At the bottom, countries like Cambodia sit at just 5.1%.

What drives these differences? Not AI model development — the US leads there but ranks only 24th in actual usage. The key factors are digital infrastructure, AI skills education, government adoption policies, and crucially, language. Countries where low-resource languages dominate show 20% lower AI adoption even after adjusting for GDP and internet access.

The notable counter-trend is DeepSeek, the open-source AI platform released under a free MIT license with no payment barriers. Its adoption has surged across Africa, China, Russia, and other markets historically underserved by Western AI platforms — a reminder that accessibility and cost are the real bottlenecks for the next billion users.

What a better visualization would show

The original is effective at one thing: communicating that most people haven't used AI. But an honest picture needs several additions:

Overlapping categories, not a pipeline. The visualization implies a hierarchy — never used → free → paid → coding — as if they're stages in a funnel. They're not. Many paid subscribers also use free tools. Many coding tool users don't pay for chatbots. A Venn or layered approach would be more accurate.

An embedded AI layer. The 2+ billion people exposed to AI through Google Search and Meta's platforms represent a distinct and important category that the original completely ignores.

Regional breakdowns. A global average obscures the 35+ percentage point gap between the UAE (64%) and the least-connected countries (5%). The digital divide in AI access is one of the defining issues in technology today.

Corrected numbers with sources. At minimum: ~50–80M paid subscribers (not 15–25M), ~15–25M coding tool users (not 2–5M), and citations to the Microsoft AI Diffusion Report, OpenAI/Reuters subscriber data, and TechCrunch/GitHub reporting.

Why this matters

Getting these numbers right isn't academic pedantry. How we understand AI adoption shapes policy decisions, educational investments, and public understanding of who benefits from this technology.

If you think only 15 million people pay for AI, the market looks tiny. If the real number is 50–80 million and growing fast, the economic and cultural implications are quite different. If you think "using AI" is limited to people who type prompts into ChatGPT, you miss the billions already being shaped by AI-generated search results and social media features.

And if you think AI adoption is a global phenomenon spreading evenly, you miss the widening divide between nations with infrastructure, skills, and language access — and those without.

The original visualization tells a true story: most people haven't used AI. But the full story is more nuanced, more urgent, and more consequential than a simple dot grid can capture.

---

Data compiled February 2026. Primary sources: Microsoft AI Economy Institute AI Diffusion Report 2025 (H2), OpenAI via Reuters, Alphabet Q2 2025 earnings call, Meta Q1 2025 earnings, TechCrunch, DataReportal Digital 2026, Stack Overflow Developer Survey 2025. Full source list and interactive visualization available in the companion piece.