This analysis examines 80,000 People Actually Want from AI and separates supported implications from claims that still depend on context or further evidence.
Read the evidence below as a decision trail: what changed, why it matters, which trade-offs shaped the result, and where the conclusion still depends on context.
Anthropic conducted the largest-ever qualitative study of AI among 81,000 Claude users, spanning 159 countries and 70 languages. The conclusion is not that "people like or hate AI", but that the same person holds expectations and fears at the same time - and the ability to create benefits is often also the ability to create risks.
- 67% of users globally have a positive attitude towards AI, but developing countries (76% in Africa, 74% in Latin America) are more optimistic than developed countries (64% in Western Europe, 66% in North America)
- 81% of respondents said that AI has brought substantial help to them, but "unreliability" (26.7%) is still the biggest concern
- People’s expectations and concerns about AI are not either/or, but coexisting — people who benefit from AI emotional support are 3 times more likely to worry about AI dependence
The first thing 80,000 people told Anthropic: AI isn’t going to replace jobs, it’s going to replace chores
The most noteworthy finding of this study is not which feature is the most popular, but the structure of people’s expectations for AI. The top-ranked demand is "professional excellence" (18.8%), but looking carefully at the content, what people want is not to let AI do more work for them, but to let AI take over trivial tasks that consume mental effort but do not produce value.
A medical administrator’s statement is accurate: after importing AI, the pressure of documentation disappeared and she became more patient with nurses. This is not a story of “increasing productivity”, this is a story of “giving attention back to people”.
The top four needs — professional excellence (18.8%), personal growth (13.7%), life management (13.5%), and time freedom (11.1%) — totaled more than 57%, all pointing to the same thing:People don’t want to use AI to do more things, but they want to use AI to do less unimportant things.
A complete map of nine expectations: from personal efficiency to social vision
The first half is "liberation" - pulling people out of trivial matters
Professional excellence, personal growth, life management, and time freedom. The common logic of these four needs is "cognitive offloading." One software engineer said that with AI support, he can now pick up his children from get off work on time. One manager said that if AI can really handle mental load, what she will get in exchange is "undivided attention."
Notably, "personal growth" ranked second (13.7%). Some people say that AI demonstrates the communication method of emotional intelligence, and he uses these behavioral patterns in interacting with people. This has gone beyond the scope of tools and entered the field of AI as "behavioral coach".
The second half is "empowerment" - making things that were previously impossible possible.
Financial independence (9.7%), social change (9.4%), entrepreneurship (8.7%), learning and growth (8.4%), creative expression (5.6%) - the logic of these five needs is different. They are not about "what less to do", but about "what can be done".
The need for entrepreneurship is particularly strong in developing countries. A Cameroonian entrepreneur said that AI is an "equalizer" that allows people who lack resources to start a business. One game developer said that without AI, his game took three years and had to scale back its ambitions. These are real-world examples of AI as a “substitute for capital.”
81% say AI is helpful, but “helpful” and “reliable” are two different things
Productivity acceleration is the most direct gain
Positive experiences focused on productivity for 32% of respondents. A Japanese engineer said that this was the first time he felt that the quality of AI surpassed humans in business tasks. But here’s a key distinction: What people mean by “productivity” is not doing more things, but doing the same things faster.
Cognitive partnerships are the second biggest gain
17.2% use AI as a thinking partner. A homeless shelter resident used AI to brainstorm a personal branding strategy for digital marketing. A doctor found two research papers on rare neurological diseases through AI and slept peacefully ever since. These cases show that the value of AI is not only in execution, but also in "helping you think of things you haven't thought of."
But 18.9% said their expectations were met with failure
Nearly one in five people feel AI is unreliable or inadequate for their needs. This is no small number. When your product has 81% positive feedback, that 19% gap often contains the most critical clues for improvement.
Geographical disparity reveals: Anxiety is a luxury in rich countries
Globally, 67% have a positive attitude toward AI, but the distribution is highly uneven. Sub-Saharan Africa 76%, Latin America 74%, Southeast Asia 72%, much higher than Western Europe 64% and North America 66%.
The logic of this gap is very straightforward:When you already have complete systems and resources, AI is a threat that may break the status quo; when you lack these, AI is an opportunity to bypass the threshold.
The specific differences are also interesting. The demand for entrepreneurship is strongest in Africa, South Asia, the Middle East, and Latin America, and AI is regarded as a "capital bypass mechanism." The need for learning is highest in Central Asia (14%) and South Asia (13%), where it is seen as a tool to break the cycle of poverty. What North America and Oceania care most about is life management - what they face is not time poverty, but cognitive scarcity.
The pattern in East Asia is completely different: personal growth needs are as high as 19%, often linked to family obligations and filial piety. Concerns about cognitive degradation (18%) and loss of meaning (13%) both outrank governance issues.
What this means
The most valuable thing about this study is not any single piece of data, but the structure it reveals: attitudes toward AI are not a position on a spectrum but a set of simultaneous tensions. This is completely different from the “optimists vs. pessimists” narrative we see in the technology media.
I think the most noteworthy thing for AI companies to pay attention to is the “3x” number—people who receive emotional support from AI are three times more likely to worry about dependence than other people. This is not the fear of the ignorant, this is the sobriety of the experienced user. If your best users are also the most anxious users, then your product design cannot only pursue "better usability", but also must design an "exit mechanism".
The geographically diverse data also breaks a common assumption. We are used to thinking that AI ethics discussions are universal, but this study clearly shows that while developed countries are discussing governance frameworks and privacy rights, people in developing countries are using AI to start businesses, learn, and change their lives. These two groups of people need completely different AI policies, and managing them with the same framework will not work.
One final observation: a qualitative study of 81,000 people is itself a methodological breakthrough. Traditional qualitative research involves in-depth interviews with 30-50 people. This study used AI to process open-ended responses in 70 languages. Anthropic uses its own products to study its own product users—which is both an advantage (the scale is unprecedented) and a limitation (the sample is all Claude users, not representative of the general public). But as information for understanding “what people who are already using AI think”, this research is currently unmatched.
The study, which spanned 159 countries and 70 languages, found that 67% of users have a positive attitude towards AI, and 81% said AI has brought substantial help. The key insight is that people have both expectations and concerns - the users who will benefit most are often the most aware of the risks, and developing countries are more optimistic about AI than developed countries.
The top four needs are professional excellence (18.8%), personal growth (13.7%), life management (13.5%), and time freedom (11.1%), totaling more than 57%. The common logic is not "use AI to do more things", but "use AI to do less unimportant things" and return attention to truly valuable work and relationships.
Sub-Saharan Africa (76%) and Latin America (74%) are much more positive than Western Europe (64%) and North America (66%). The core reason is that AI plays the role of an "equalizer" in resource-scarce environments, allowing people to bypass traditional barriers to entrepreneurship and learning, while developed countries are more worried about AI breaking existing institutional advantages.
Original source:What 81,000 people want from AI — Anthropic, March 18, 2026
Five groups of "light and shadow" tensions: benefits and risks come from the same ability
The most insightful part of this study is that it identifies five sets of simultaneous tensions. This is not "some people think it's good and some people think it's bad"; it'sThe same group of people benefit and worry about the same thing at the same time。
Learning vs. Cognitive Degeneration
33% cited learning benefits (of which 91% had actually experienced them), but 17% were concerned about cognitive degradation. A Korean student admitted that he just memorized the answers given by the AI. That was the moment when he blamed himself the most. Educators observe cognitive decline at 2.5 to 3 times the rate of the general population.
Emotional support vs. dependence
16% describe emotional benefits from AI, and 12% worry about dependence. There is a surprising correlation in the data:People who receive emotional support from AI are three times more likely to worry about dependence.This means that those who benefit most are also most aware of the risks. One Ukrainian user mentioned that AI provided critical emotional support during wartime.
Time Savings vs. Unreal Productivity
50% said it saved time, and 18% warned it was "running faster but staying put". Freelance workers feel efficiency gains and competitive pressure at the same time - when everyone uses AI to speed up, the time you save is only filled with more work.
Economic Empowerment vs. Job Displacement
28% see economic opportunities and 18% are worried about losing their jobs. Independent workers experience both feelings simultaneously, while agency workers experience relatively low replacement anxiety. A good analogy is that in the third industrial revolution, horses disappeared. Now people are afraid that they are that horse.
What to take away
The article's value is in the evidence and trade-offs behind what 80,000 People Actually Want from AI, not in treating the conclusion as universal.