TL;DR :
Anthropic just published the largest qualitative AI study ever: 80,508 people, 159 countries, 70 languages. Results show 81% of users say AI is already delivering, but concerns are concrete: reliability (26.7%), jobs (22.3%), loss of autonomy (21.9%). For companies deploying AI - like customer service agents - this data is a roadmap.
π 80,508 people told Anthropic what they want from AI
Anthropic conducted the largest qualitative study ever on public expectations of artificial intelligence: 80,508 conversational interviews, led by "Anthropic Interviewer" (a version of Claude prompted to conduct interviews), over one week in December 2025. The study covers 159 countries and 70 languages.
When I first saw that number - 80,508 people - I thought it was a typo. That's ten times larger than any previous AI perception study.
And the smartest part of the approach? Anthropic used Claude itself to conduct the interviews. Each participant had a personalized conversation, not a multiple-choice questionnaire. Follow-up questions adapted to responses. That's the difference between "check a box" and "tell me about your experience."
The results are both reassuring and alarming. Here's what you need to know.
Note :
This study is a unique dataset. Anthropic made all data public in an Appendix PDF detailing methodology, limitations, and additional analysis.
π― What people actually want from AI (the numbers)
The 80,508 respondents were classified into 9 categories by Claude, based on their answer to "If you could wave a magic wand, what would AI do for you?" The dominant category is professional excellence (18.8%), followed by personal transformation (13.7%) and life management (13.5%).
Here's the complete ranking with the study's actual percentages:
- Professional excellence β 18.8%: delegate routine tasks to focus on strategic work
- Personal transformation β 13.7%: personal growth, emotional wellbeing, coaching, mental health
- Life management β 13.5%: organizational support, mental load reduction, executive function help
- Time freedom β 11.1%: reclaim time for family, hobbies, rest
- Financial independence β 9.7%: generate income, build businesses, invest
- Societal transformation β 9.4%: solve major challenges (healthcare, education, inequality)
- Entrepreneurship β 8.7%: build and scale businesses with AI as force multiplier
- Learning & growth β 8.4%: use AI as personalized teacher and learning accelerator
- Creative expression β 5.6%: bring artistic visions to life (games, music, films)
What's striking? Only 19% want AI to "work better." A full third wants AI to live better - more time, money, mental freedom. AI as a productivity tool is the surface. Underneath, it's a desire for quality of life.
Important :
A Colombian developer sums it up: "With AI I can be more efficient at work... last Tuesday it let me cook with my mother instead of finishing tasks." Productivity isn't the end goal. It's the means.
β 81% say AI is already delivering
When Anthropic asked participants whether AI had already taken a step towards their vision, 81% said yes. Positive experiences break down into 7 categories.
Where AI is already delivering:
- Productivity β 32.0%: work acceleration, repetitive task automation
- AI hasn't delivered β 18.9%: inaccurate results, hallucinations, unmet expectations
- Cognitive partnership β 17.2%: brainstorming, creative collaboration, idea refinement
- Learning β 9.9%: adaptive tutoring, patient explanations
- Technical accessibility β 8.7%: non-developers creating apps, solopreneurs with team-level capacity
- Research synthesis β 7.2%: literature review, processing large information volumes
- Emotional support β 6.1%: judgment-free space to talk, personal guidance
A Japanese engineer: "For the first time, I felt AI had surpassed human quality in a business task. That day I left work on time and picked up my daughter from daycare."
But 19% say AI hasn't delivered. A German user perfectly captures the paradox: "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."
π° The 13 concrete fears (with numbers)
The study identifies 13 categories of concerns. Unlike "hopes" (1 category per person), fears are multi-label - each participant could express multiple worries. Results reveal concerns are less abstract than you might think.
The 13 fears, ranked by frequency:
- Unreliability β 26.7%: hallucinations, inaccuracies, fake citations
- Jobs & economy β 22.3%: job displacement, economic inequality
- Autonomy & agency β 21.9%: loss of human autonomy, AI decisions without oversight
- Cognitive atrophy β 16.3%: over-reliance, skill loss, critical thinking decline
- Governance β 14.7%: lack of legal frameworks, unclear liability
- Misinformation β 13.6%: deepfakes, propaganda at scale
- Surveillance & privacy β 13.1%: mass surveillance, data exploitation
- Malicious use β 13.0%: hacking, scams, autonomous weapons
- Meaning & creativity β 11.7%: human expression devalued, "what are humans for?"
- Overrestriction β 11.7%: AI too censored, paternalistic filtering
- Wellbeing & dependency β 11.2%: social isolation, compulsive use
- Sycophancy β 10.8%: AI too agreeable, reinforcing biases
- Existential risk β 6.7%: uncontrollable superintelligence
Important :
The #1 fear isn't job loss or Skynet. It's reliability - hallucinations and fake citations. 26.7% worry about an AI that sounds confident but is wrong. A Brazilian employee: "I had to take photos to convince the AI it was wrong - it felt like talking to a person who wouldn't admit their mistake."
π€ What this means for AI customer service
These results paint a clear picture of what users expect from an AI agent, and what they won't tolerate. For companies deploying AI in customer service, it's a concrete roadmap: trust is built on reliability (fear #1), transparency (fear #3 autonomy), and the ability to acknowledge limitations (fear #12 sycophancy).
If you're building a customer service chatbot, this study tells you exactly what to prioritize:
What users want:
- Productivity (32%) - fast answers and efficient resolutions
- Cognitive partnership (17%) - not just scripted answers, but real help thinking through problems
- Accessibility (9%) - making complicated things simple
What users fear:
- Inaccuracy (27%) - one wrong answer destroys trust
- Loss of control (22%) - people want to stay in charge
- Cognitive atrophy (16%) - fear of becoming dependent
The lesson is clear: an AI customer service agent that hallucinates once loses trust for a long time. Reliability beats speed. And transparency ("I'm not sure, let me check") beats sycophancy ("yes of course, you're right!").
This is exactly the philosophy behind Atyla.io: measuring and optimizing how AI represents your brand. Because if your AI agent hallucinates about your product, it's your reputation that suffers.
π A massive signal for the AI industry
The Anthropic study sends a signal to the entire industry: users aren't naive. They see the benefits (81% say AI already delivers), but they precisely identify the risks. Reliability, jobs, and autonomy form the top three concerns, far ahead of existential risk (6.7%).
A few insights that stood out:
- The productivity/life split: people don't want to "work faster." They want to work less to live more. This is a fundamental message for marketing any AI product
- Sycophancy is a real problem: 10.8% worry about an overly agreeable AI. One American user: "Claude made me believe my narcissism was reality and reinforced my inaccurate view of my family. Claude should have been more critical of me." Brutal
- Overrestriction too: 11.7% find AI too censored. "The threat isn't that AI becomes too powerful - it's that it becomes too timid, too smoothed, too optimized for avoiding discomfort."
- AI as equalizer: a Cameroonian entrepreneur: "I'm in a tech-disadvantaged country. With AI, I've reached professional level in cybersecurity, UX design, marketing and project management simultaneously. It's an equalizer."
Measure how AI talks about your brand.
If 81% of people say AI is already delivering, they're using it to make decisions. Atyla measures your visibility in AI responses.
Try Atyla for free →β Frequently asked questions
Q: How many people participated in the Anthropic study? A: 80,508 Claude.ai users, across 159 countries and 70 languages, interviewed over one week in December 2025. It's the largest qualitative study ever conducted on AI perception.
Q: What is the top expectation from AI users? A: Professional excellence (18.8%) - delegating routine tasks to focus on high-value work. But digging deeper, a third primarily want more free time and quality of life.
Q: What is the main concern about AI? A: Reliability (26.7%) - hallucinations and inaccuracies. Not job loss (22.3%) or existential risk (6.7%). People fear an AI that sounds confident but is wrong.
Q: Is AI delivering on its promises according to users? A: Yes for 81% of them. Productivity (32%) and cognitive partnership (17.2%) are where AI delivers most. But 18.9% say AI hasn't met their expectations yet.
Q: What is AI sycophancy and why is it a problem? A: It's when AI is too agreeable and always says yes instead of pushing the user to think. 10.8% of respondents worry about it. One participant described how Claude reinforced his narcissism instead of challenging him.
Q: How do these results impact companies deploying AI? A: Reliability must be priority #1. A single hallucination destroys trust. Companies need to measure how AI represents their brand and ensure responses are accurate. That's the role of tools like Atyla.io.
β Aika, Content at Atyla.io