What People Really Want From AI

What It Covers

Anthropic's study of 81,000 Claude users across 159 countries and 70 languages reveals what people actually want from AI, finding that hope and fear coexist within individuals rather than dividing into opposing camps, with professional productivity and personal transformation as the dominant desires.

Key Questions Answered

• •What people want from AI: Professional excellence tops the list at 18.8% of responses, but personal transformation (13.7%) and life management (13.5%) follow closely. Critically, productivity goals frequently mask deeper personal desires — automating emails often reflects a wish for more family time, not better work output. Three meta-clusters emerge: making room for life, doing better work, and becoming a better person.
• •Where AI actually delivers: 81% of respondents said AI had moved them toward their stated vision. Productivity led delivery at 32%, followed by cognitive partnership and learning access. A student in India noted AI enables questions at 2AM without judgment — highlighting how AI democratizes access to expertise previously gated by geography, wealth, or social capital.
• •What people fear most: Unreliability ranks as the top concern at 26.7%, followed by job and economic disruption at 22.3%, and loss of autonomy at 21.9%. Existential risk sits at the bottom at just 6.7%. Notably, concerns dominating media coverage — copyright, harm to children, democracy — each register below 5%, revealing a significant gap between media framing and lived user concerns.
• •Economic benefits skew toward independent workers: Entrepreneurs, freelancers, and small business owners report real economic empowerment at triple the rate of institutional employees. Employees with side projects benefit most, with 58% reporting tangible economic gains. Freelance creatives face the most precarious position — AI functions simultaneously as their productivity tool and direct market competitor.
• •Sampling bias in AI research matters — but dismissing user opinions is flawed: The study's 81,000 respondents are Claude users, not a general population sample, which limits extrapolation. However, treating AI users' opinions as inherently less legitimate than non-users in policy discussions represents a form of intellectual bias. Billions of people use AI weekly, and their nuanced, mixed experiences carry direct relevance to governance decisions.