53%
of people use generative AI, just three years after launch
Stanford HAI, AI Index 2026 · 13 Apr 2026Everyday, and uneasy
In three years, half the world started using AI. Half of Americans are more worried than excited about it.




53%
of people use generative AI, just three years after launch
Stanford HAI, AI Index 2026 · 13 Apr 2026In three years, half the world started using AI. Half of Americans are more worried than excited about it.
50%
of Americans are more worried than excited about AI
Pew Research Center · 17 Sep 2025Up from 37% in 2021. Only 1 in 10 is more excited than worried.
35 of 42
points, a gold medal at the world’s top maths competition for students
Google DeepMind · 21 Jul 2025 †At the International Mathematical Olympiad, under the same time limit and grading as human contestants.
AI solved five of six problems at the hardest student maths contest. Yet it still misreads a simple clock about half the time.
50.1%
of analogue clocks read correctly
Stanford HAI, AI Index 2026 · 13 Apr 2026That’s the best AI model available, and it’s no better than a coin toss.
83%
less time spent on paperwork
Stanford HAI, 12 Takeaways · 13 Apr 2026For doctors who use AI to write up patient notes.
AI already saves doctors time on paperwork. But very few studies have tested medical AI on real patient records.
5%
of medical AI studies used real patient data
Stanford HAI, 12 Takeaways · 13 Apr 2026Out of more than 500 studies. The rest relied on test or simulated data.
about 13 times
cheaper every year for the same quality
State of AI Report 2026, citing Epoch AI · 8 Oct 2026 †Getting the same result from AI costs about half as much every three months.
Using AI gets cheaper every year. But the costs show up elsewhere: electricity, water, and data centres in people’s neighbourhoods.
29.6 gigawatts
of electricity capacity used by AI data centres
Stanford HAI, 12 Takeaways · 13 Apr 2026Running one popular AI model for a year may use more water than 1.2 million people drink. Stanford HAI · 13 Apr 2026
71% of Americans would oppose an AI data centre near their home. That’s more than the 53% who would oppose a nuclear plant. State of AI Report 2026, citing Gallup · 8 Oct 2026 †
73%
of AI experts think AI will make jobs better
Stanford HAI, AI Index 2026 · 13 Apr 2026When asked if AI will improve how people do their work.
People who build and study AI are far more optimistic about its effect on jobs than the people who work alongside it. The gap is 50 percentage points.
23%
of the public think the same
Stanford HAI, AI Index 2026 · 13 Apr 2026Same question, 50 points lower.
AI isn’t finished, it isn’t something to fear, and it isn’t free. This year it won a maths gold medal but still misread clocks. It cut doctors’ paperwork, though the evidence behind medical AI is still thin. It got cheaper to use, while its data centres needed more power and water and faced local opposition. When left to use a computer on its own, it still fails about one task in three (Stanford HAI, AI Index 2026 · 13 Apr 2026). Half of Americans are more worried than excited, and experts are far more hopeful than the public. Both views have a point. Use these numbers, and their sources, the next time AI comes up.
Want to share a number? Press Copy next to it to copy the number with its source.
Where each number comes from and when it was published. All checked on 10 October 2026. Three marked † were confirmed through trusted news reports because the original page couldn’t be accessed.
† Confirmed through trusted news reports; the original page couldn’t be accessed.
Every image on this page was made with AI using one saved style, “Renaissance Mythic Classicism” on StyleRef. Using the same style for each image keeps the colours, texture and drawing consistent, so the pictures read as one series.