Anthropic published findings on 18 March from nearly 81,000 Claude users who had been invited to describe how they used AI, what they hoped it might make possible, and what they feared. The scale and multilingual approach made the study unusual among qualitative AI research.
A large invitation to speak in people's own words
The participants were Claude users who chose to respond. Their answers can reveal themes and language in depth, but they do not form a random survey of the whole public. A large number of responses does not remove that selection effect.
Detailed stories about desired and unwanted uses.
Respondents were not a random public sample.
Repeat the questions inside the affected organisation.
What the sample could not represent
The study is most useful as a prompt for local listening. Before writing an organisation's AI rules, ask staff which tasks are frustrating, what help they would welcome, where they fear loss of control, and what decisions must remain clearly human.
Key takeaways
- Participation is voluntary.
- Comments are not used to judge individual staff.
- Minority concerns are not erased by the average.
- Policy decisions explain the evidence behind them.
People's hopes and concerns often exist together. A worker may want help with repetitive administration while still wanting control over evaluation, pay, or customer decisions. A sensible policy has room for both sides.
