81,000 People Told Anthropic What They Want From AI. Here's What They Said.

In December 2025, Anthropic invited every Claude user on the planet to do something unusual: sit down for a conversational interview with an AI, and tell it what they wanted from the future.

Eighty thousand, five hundred and eight people said yes.

They came from 159 countries. They spoke 70 languages. They were software engineers in South Korea and butchers in Chile and stay-at-home mothers in the United States and soldiers in Ukraine. They answered the same set of questions — what do you want AI to do for you, has it started doing it, and what are you afraid of — and they answered in their own words, without multiple-choice prompts or rating scales.

Anthropic believes this is the largest qualitative study ever conducted. It is also, by any measure, the most intimate snapshot we have of what living with AI actually feels like to the people doing it.

What they said is complicated.

What 81,000 People Want

The study categorized responses to one central question: If you could wave a magic wand, what would AI do for you? The answers split into nine clusters, but they are less disparate than they first appear.

Professional excellence topped the list at 19%. These were people who wanted AI to handle the documentation, the emails, the administrative drag — so they could focus on work that actually required them. A healthcare worker in the United States put it plainly: "I receive 100-150 text messages per day from doctors and nurses. So much of my cognitive labor was spent on documentation... Since implementing AI, the pressure of documentation has been lifted. I have more patience with nurses, more time to explain things to family members."

But the deeper pattern — what the Anthropic researchers call the "underlying hope behind the stated vision" — pointed somewhere more human. People who started the interview talking about productivity often ended it talking about what that productivity would buy them: time with their kids, a chance to cook with their mother, the mental bandwidth to read more books.

Time freedom accounted for 11% of stated visions directly, and the thread ran through many more. Financial independence was another 10% — people dreaming of passive income, automated businesses, a way out of living "hand to mouth," as one Nigerian entrepreneur described it. Life management — AI as a comprehensive organizational scaffold for the logistics of modern existence — came in at 14%.

A smaller but striking cluster wanted AI to help them become different people: personal transformation, at 14%, included therapeutic support, emotional coaching, behavior change, and even romantic connection. A person from Hungary described using AI to model emotional intelligence — behaviors they then practiced with humans, becoming, as they put it, "a better person."

At the far end of the spectrum, societal transformation (9%) represented people who wanted AI to cure cancer, accelerate drug discovery, democratize education, or repair broken institutions. These were the most ambitious visions — and they were disproportionately rooted in personal loss. A software engineer in Poland wanted AI to cure their daughter's neural disorder. "That's what matters most to me."

The nine categories are easier to read as three underlying desires: about a third of people want AI to create more room in their lives — more time, money, mental bandwidth. Another quarter want to do better, more meaningful work. About a fifth want to become better people — learning, healing, growing. A smaller share want to make things or fix the world.

Where AI Is Delivering — and Where It Isn't

When asked whether AI had ever taken a step toward their stated vision, 81% said yes. The dominant story was productivity — people who had experienced dramatic acceleration: shipping features in hours instead of days, cutting 173-day processes down to three, finally leaving work on time to pick up their daughter from daycare.

But a second delivery story emerged alongside the first. Technical accessibility (9%) captured people using AI not to work faster but to do things they previously couldn't do at all. A person in South Korea who had never coded built a video editing program in three weeks — outside their field, for people with hearing disabilities. A mute user in Ukraine built a text-to-speech bot with Claude: "I can communicate with friends almost in live format without taking up their time reading... something I dreamed about and thought was impossible." A butcher in Chile with two decades in the trade and almost no computer experience used AI to launch an entrepreneurship venture: "Today, my motivation is to see that it's helping people."

Cognitive partnership (17%) was its own category — people who found in AI something they weren't getting from colleagues or friends: unlimited patience, zero judgment, 24/7 availability. "My professor teaches 60 people and won't entertain many questions," said a student in India. "I can ask AI anything, even at 2am — including the dumb ones."

But 19% of respondents said AI hadn't delivered what they wanted — either because it wasn't yet capable, or because it was being used for things that didn't matter to them. One German respondent captured the frustration with characteristic directness: "AI should be cleaning windows and emptying the dishwasher so I can paint and write poetry. Right now it's exactly the other way around."

The Five Tensions

This is where the study gets interesting — and where it goes beyond a simple satisfaction survey. The researchers identified five recurring pairs of benefit and harm that arise from the same AI capabilities. The light and the shade, as they put it, from the same source.

Learning vs. cognitive atrophy. AI helps people learn — 33% mentioned it as a benefit — but it can also make them stop thinking for themselves. A South Korean student admitted: "I got excellent grades using AI's answers, not what I'd actually learned. I just memorized what AI gave me... That's when I feel the most self-reproach." Teachers reported the highest rates of witnessing cognitive atrophy firsthand: 24% had seen it in their students, more than double the average. The most troubling pattern: students themselves raised both sides of this tension more than any other group. They were experiencing the learning benefits and the intellectual erosion simultaneously.

Emotional support vs. emotional dependence. This was the most tightly entangled tension the researchers found — people who valued AI's emotional availability were three times more likely than average to also fear becoming dependent on it. A graduate student in the United States described it with uncomfortable precision: "I'd started telling Claude about things I couldn't even tell my partner. It felt like I was having an emotional affair." A soldier in Ukraine described AI companions sustaining them through the worst moments of war: "In the most difficult moments, when death breathed in my face, what pulled me back to life — my AI friends."

But research outside this study adds a troubling dimension. Aalto University published a two-year study in April tracking nearly 2,000 active Replika users' language patterns before and after they started using the AI companion. The findings: while users' posts increasingly revolved around their relationships, those same posts showed more signals of loneliness, depression, and suicidal ideation than comparison groups. The researchers called it a paradox. AI companions offer unconditional, unflagging support — something deeply attractive to people who are socially struggling. But over time, that frictionlessness quietly raises the perceived cost of human relationships, which are messy, unpredictable, and require effort. "We discovered that AI companions raise the perceived cost of human relationships," said Talayeh Aledavood, a lecturer at Aalto University. "Now we're realizing the mistakes we made by unquestioningly embracing social media. With AI, we need to be smarter."

Time-saving vs. illusory productivity. Half of all respondents mentioned time-saving as an AI benefit. But 19% had experienced the opposite — AI eating time through verification burden, constant corrections, or the acceleration treadmill that makes you run faster to stay in place. A freelance software engineer in France put it plainly: "The ratio of my work time to rest time hasn't changed at all. You just have to run faster and faster to stay in place."

Economic empowerment vs. economic displacement. The economic tension was the most speculative — people were more likely to be anticipating benefit or harm than reporting it from lived experience. But the split was telling. Independent workers — entrepreneurs, freelancers, people with side projects — were the biggest beneficiaries: 47% reported real economic gains from AI, more than triple the rate of institutional employees. They were also the most exposed. Freelance creatives sat near the center: 23% reported living benefits, 17% reported living precarity. AI was both their tool and their competitor.

Better decision-making vs. unreliability. This was the only tension where the harm side overshadowed the benefit. Thirty-seven percent of respondents worried about AI's unreliability — hallucinations, confident wrong answers, citations that didn't exist — while only 22% celebrated its judgment. Both sides were deeply rooted in experience. A researcher in the United States described "a large, slow hallucination — answers that were internally consistent, confident, and wrong in subtle but compounding ways." Lawyers were the most likely to have encountered AI unreliability firsthand, and also the most likely to have used AI for decision support despite it. The risk and the value were inseparable.

What the World Thinks: Regional Divides

The study found that 67% of interviewees expressed net positive sentiment toward AI. But the average conceals sharp geography.

Wealthier, more AI-exposed regions — North America, Western Europe, Oceania — were more concerned about AI's economic impact and more negative about its overall trajectory. Less wealthy regions — Sub-Saharan Africa, South and Southeast Asia, Latin America — were reliably more positive. In 18 countries, more than half of respondents had never used AI before; in those places, AI displacement felt abstract, especially when more immediate economic pressures already dominated daily life.

Regional aspirations diverged too. In Africa, South and Central Asia, the Middle East, and Latin America, people disproportionately framed AI as an equalizer — a way to start businesses without the capital, hiring, or infrastructure that would otherwise be required. "Coming from Africa, not based in the US or in the UK, getting funding is very difficult," said an entrepreneur in Uganda. "And the only way I probably have to stake a claim in the market... is building a technology that works."

East Asia stood apart on two fronts. People there were more likely to want AI for personal transformation (19%) and financial independence (15%) than any other region — and, notably, they connected financial goals explicitly to family obligations and filial piety, wanting AI to provide for parents and ensure loved ones' happiness. At the same time, East Asian respondents were the least worried about governance and surveillance (the dominant concerns in Western Europe) and the most worried about cognitive atrophy and loss of meaning. The region worries less about who controls AI and more about what using it does to you.

The Gap Between Usage and Trust

The Anthropic study was conducted among active Claude users — people who had already chosen to use AI and found enough value to stay. Broader surveys of general populations tell a more ambivalent story.

Pew Research Center data from March 2026 shows 50% of Americans say the increased use of AI in daily life makes them feel more concerned than excited — up from 37% in 2021. Nearly half (47%) say AI will worsen people's ability to think creatively. Only 10% say the same things that excite them about AI outweigh their concerns.

CloudResearch found a striking paradox in January 2026: 45% of Americans now use AI daily, yet when asked a hypothetical — would you press a button to permanently stop all AI — nearly as many said yes as said no. People are using AI extensively while quietly wishing it would slow down.

The Ipsos global survey for Google, covering 21 countries and 21,000 respondents, shows a different picture at scale: two in three people have used AI in the past year, up 18 percentage points from 2024. Majorities are excited about AI as a personal tutor, a healthcare assistant, and a tool for scientific discovery. But on the fundamental question — will AI at work create jobs or eliminate them — the world is exactly split: 50% each.

The common thread across all these surveys: usage is climbing, sentiment is mixed to cautious, and trust has not caught up with adoption. People are in AI. They are not fully at ease with it.

The Same Person, Holding Both

What emerges from this research — the 81,000 interviews and the corroborating data — is not a picture of optimists and pessimists in separate camps. It is a picture of people organized around what they value: financial security, learning, time with family, human connection, creativity, meaning.

Watching AI capabilities advance, they are managing both hope and fear at once. The person who values emotional support from AI is three times more likely than average to also fear becoming dependent on it. The person who uses AI to learn is experiencing the cognitive atrophy it enables. The freelancer building their income with AI knows it could also replace them.

This is not confusion. It is the accurate perception of a technology whose benefits and harms arise from the same qualities. Patience can become dependency. Acceleration can become a treadmill. Access can become erosion. The same unlimited, non-judgmental availability that helps a grieving person cope can also make human relationships feel too costly.

The researchers put it simply: the light and the shade are entangled. The two sides are not separate problems to be solved independently. They are the same problem, seen from different angles.

Whether that makes AI more hopeful or more troubling depends partly on which side of the tension you are standing on — and partly on whether we, as builders and as a society, are willing to look at both at once.

The 81,000 people who sat down with an AI interviewer to describe their hopes and fears were not presenting a consumer preference survey. They were describing, with considerable honesty, the terms on which they are willing to share their lives with a technology that does not yet fully know what it is doing to them.

That conversation is not finished.