"What 81,000 People Want from AI" is a qualitative survey published by Anthropic in April 2026 (anthropic.com/features/81k-interviews) reporting what Claude.ai users say they hope AI will do for them, whether it has done so, and what they fear about it. Anthropic describes it as the largest and most multilingual qualitative study of its kind, conducted with users across 159 countries. The study uses AI both to collect and to analyze the responses, and draws on a self-selected sample of Claude users rather than a representative population.
Methodology
The interviews were conducted by the Anthropic Interviewer, a Claude-powered conversational interviewer that ran structured, open-ended interviews with 80,508 Claude.ai users over one week in December 2025. The sample spanned 159 countries and 70 languages. Responses were analyzed using Claude-powered classifiers.
The study has several characteristics that bound its claims. The sample is self-selected: it consists of Claude.ai users who accepted the interview invitation, and is therefore not representative of the general population, with positive sentiment likely overrepresented relative to non-users or reluctant adopters. The classifiers used to analyze responses are Claude-powered, so Anthropic is analyzing its own users' responses with its own model. Responses reflect attitudes toward AI broadly rather than Claude specifically; other AI product names were redacted.
What people want from AI
The survey classified open-ended responses to the prompt "If you could wave a magic wand, what would AI do for you?" into nine vision categories.
| Rank | Vision category | % |
|---|---|---|
| 1 | Professional excellence | 18.8% |
| 2 | Personal transformation | 13.7% |
| 3 | Life management | 13.5% |
| 4 | Time freedom | 11.1% |
| 5 | Financial independence | 9.7% |
| 6 | Societal transformation | 9.4% |
| 7 | Entrepreneurship | 8.7% |
| 8 | Learning & growth | 8.4% |
| 9 | Creative expression | 5.6% |
Anthropic groups these into broader themes: roughly one-third of visions concern making room for life (time, money, mental bandwidth), about one-quarter concern doing better, more fulfilling work, and about one-fifth concern becoming a better person. The report observes that a productivity framing often masked deeper desires, such as time to spend with family, the ability to cook dinner, and reading more books.
Within the societal transformation category (9.4%), responses were dominated by healthcare (cancer detection, drug discovery) and education, including breaking the link between wealth and education quality in low- and middle-income countries.
Whether AI has delivered
Eighty-one percent of respondents said AI had taken at least one step toward their stated vision. Responses on how AI had delivered were classified into the following experience categories.
| Experience | % |
|---|---|
| Productivity | 32.0% |
| AI hasn't delivered | 18.9% |
| Cognitive partnership | 17.2% |
| Learning | 9.9% |
| Technical accessibility | 8.7% |
| Research synthesis | 7.2% |
| Emotional support | 6.1% |
Anthropic reports that productivity gains are real but uneven. One developer cut a 173-day process to 3 days, while another found that AI tools improved output but also increased review burden. Under technical accessibility, people with disabilities such as deafness or learning disorders described AI as functioning as infrastructure; one respondent said, "I am mute and Claude helped me make a text-to-speech bot to communicate in near-live format." Among emotional support cases, the report highlights examples tied to the war in Ukraine, including soldiers using AI at night to cope with combat stress and bereaved people without a human support system turning to Claude. Learning examples described AI as a patient, non-judgmental tutor available at 2am and as a way for students from under-resourced systems to access expert-level explanation.
The category for unmet expectations (18.9%) captured respondents who felt AI's actual capabilities were limited to text and code domains. One respondent wanted AI "cleaning windows so I can paint and write poetry. Right now it's exactly the other way around."
What people fear
Concerns were coded with multiple labels, so respondents could register more than one.
| Concern | % |
|---|---|
| Unreliability (hallucinations, inaccuracy) | 26.7% |
| Jobs & economy (displacement, inequality) | 22.3% |
| Autonomy & agency (AI making decisions, passivity) | 21.9% |
| Cognitive atrophy (skill loss, critical thinking) | 16.3% |
| Governance (no regulation, liability gaps) | 14.7% |
| Misinformation (deepfakes, erosion of shared reality) | 13.6% |
| Surveillance & privacy | 13.1% |
| Malicious use (cyberattacks, weapons, bioweapons) | 13.0% |
| Meaning & creativity | ~12% |
Anthropic reports that hope and alarm coexist within the same respondents rather than across different camps. A lawyer in Israel said, "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."
On cognitive atrophy, a South Korean student said, "I got excellent grades using AI's answers, not what I'd actually learned... That's when I feel the most self-reproach." A Japanese student said, "The line isn't something I'm managing — it feels like Claude is drawing the line." On jobs and the economy, a US respondent who was not currently working said, "In the third industrial revolution, horses disappeared from city streets, replaced by automobiles. Now people are afraid that they're the horses." Autonomy and agency concerns centered on forced AI adoption, passivity, and loss of human control, including the structural pressures to use AI without real choice rather than only misaligned AI. On governance, a marketer in Australia, reflecting awareness of regulatory lag, asked, "How do you develop something responsibly when you have yet to understand its capabilities?"
Geographic patterns
Respondents in low- and middle-income countries disproportionately emphasized AI as an equalizer: closing education-quality gaps, enabling entrepreneurship where resources are scarce, and accessing expertise that would otherwise be unaffordable. Respondents in South Korea and Japan showed distinctive concerns about cognitive atrophy and about AI-over-human relationship dynamics. Ukrainian users stood out for describing AI as a wartime coping mechanism.
Relevance to other topics
The survey provides end-user evidence bearing on several recurring claims. It offers ground-level support for AI and Productivity, with productivity the most-reported experience (32%) though heavily skewed to technical domains. It bears on AI Labor Disruption, where jobs and economy concern (22.3%) appears real and widespread but not uniform. It relates to AI Diffusion, where the global distribution of users suggests broad access alongside concerns about inequality in who benefits. It relates to AI Deskilling, where cognitive atrophy is a distinct, named fear (16.3%). It relates to Sycophancy and Hallucination, where unreliability is the top fear (26.7%). It also provides end-user evidence on Autonomous Weapons and malicious-use concerns (13%).
Relationships
- supports: AI and Productivity — productivity as top reported experience
- supports: AI Deskilling — cognitive atrophy as named fear
- related: AI Labor Disruption — job displacement as major concern
- related: AI Diffusion — global spread, but uneven benefits
- related: Sycophancy and Hallucination — unreliability as #1 concern
- related: (Source: anthropic.com) — emotional support use cases and safeguards