Back in December, Anthropic did something unusual. Instead of running another survey with multiple-choice boxes, they let Claude itself interview tens of thousands of users. The result? 80,508 people across 159 countries and 70 languages sat down with an AI interviewer and talked about what they actually want from this technology.
That’s not just a big number. It’s the largest and most multilingual qualitative study ever conducted on AI, period. And the findings are more nuanced than any hot take I’ve seen on Twitter.
Hope and alarm live in the same head
The biggest takeaway for me? People aren’t split into “optimists” and “doomers.” Almost everyone holds both hope and fear simultaneously. A lawyer in Israel put it bluntly: “I use AI to review contracts, save time… and at the same time I fear: am I losing my ability to read by myself? Thinking was the last frontier.”
That tension runs through the entire dataset. A freelancer in the US credits Claude with leading to a proper diagnosis after nine years of misdiagnosis. An entrepreneur in Nigeria sees AI as a potential escape from the hand-to-mouth cycle. Meanwhile, a technical support specialist in the US got laid off because their company replaced them with an AI system. A software engineer in South Korea warns we’ve never dealt with something smarter than ourselves.
Both sets of experiences are real. Both are happening right now.
What people actually want
Anthropic classified each person’s primary desire into categories. Here’s what stood out:
Professional excellence (18.8%) was the single biggest category. People want AI to handle the grunt work so they can focus on the stuff that actually matters. A healthcare worker described receiving 100-150 text messages per day from doctors and nurses. After implementing AI, the documentation pressure lifted. They had more patience with nurses, more time for families. That’s not abstract — that’s a direct quality-of-life improvement.
Personal transformation (13.7%) surprised me. People are using AI for emotional growth, therapy-adjacent work, even learning how to be better humans. One person in Hungary said, “AI modeled emotional intelligence for me… I could use those behaviors with humans and become a better person.” I’ve seen this in my own circles too. People are having conversations with AI that they’d never feel comfortable having with another human.
Life management (13.5%) and time freedom (11.1%) round out the top desires. A manager in Denmark said, “If AI truly handled the mental load… it would give me back something priceless: undivided attention.” That resonates. The dream isn’t that AI does everything. It’s that AI handles the noise so humans can focus on what matters.
The fears are concrete
The concerns aren’t abstract Elon Musk-style existential risks (though those came up too). They’re immediate, personal, and grounded in real experience. Job displacement. Loss of cognitive skills. Over-reliance. The creeping sense that if you let AI do your thinking, you might forget how to think.
I appreciate that Anthropic didn’t try to sugarcoat this. The study captures both sides honestly. People are using AI to diagnose diseases and write contracts, and they’re also terrified it’s making them dumber or taking their livelihoods.
How they did it
The methodology is worth noting. Instead of a static survey, Anthropic built “Anthropic Interviewer” — a version of Claude prompted to conduct conversational interviews with follow-up questions based on responses. Then they used Claude-powered classifiers to categorize the conversations across dimensions: what people want, whether they’re getting it, what they fear, their occupation, and overall sentiment.
This bridges the classic tradeoff in qualitative research between depth and volume. You usually get either rich interviews with 50 people or shallow surveys with 50,000. Here they got rich interviews at massive scale. That’s genuinely novel.
What’s missing
I wish the study went deeper on what “AI going well” actually looks like in practice. The categories are useful, but they’re still labels. The real texture is in the quotes, and Anthropic did publish a Quote Wall where you can filter by region, concern, and vision. That’s worth browsing.
I also wonder about selection bias. These are Claude users — people who already opted into using AI. The study doesn’t capture people who’ve tried AI and walked away, or people who refuse to touch it. That’s a legitimate limitation they acknowledge in the appendix.
Still, this is the most grounded picture I’ve seen of what regular people — not VC partners, not think tank fellows, not Twitter thought leaders — actually want from AI. They want it to make their work more meaningful, their lives less chaotic, and their time their own. They’re scared it might take the last of those away.
Both things can be true. The study shows they usually are.
Comments (0)
Login Log in to comment.
Be the first to comment!