What 81,000 people told us about the economics of AI.
Summary
This Anthropic report links a large-scale qualitative survey of roughly 81,000 Claude.ai users to the quantitative occupational exposure data tracked by Anthropic’s Economic Index. Using Claude-powered classifiers to infer occupations, career stages, and attitudes from free-text responses, the authors examine how worker-reported displacement concerns and productivity gains map onto measured AI usage across jobs. The headline argument is that subjective fears about AI track objective exposure: workers in occupations where Claude is observed to handle more tasks worry more about being displaced, even as most respondents — especially the highest- and lowest-paid — report substantial personal productivity benefits, often through doing new kinds of work rather than just working faster.
Key Contributions
- Connects qualitative, self-reported AI experiences to a task-based occupational exposure measure at unusual scale.
- Provides empirical evidence that perceived displacement risk tracks measured AI diffusion across occupations.
- Demonstrates a reusable methodology for extracting structured labor-economics variables (occupation, career stage, beneficiary of gains, type of productivity gain) from open-ended survey text via LLM classifiers.
- Extends the Anthropic Economic Index with a worker-sentiment layer alongside its task-usage analysis.
- Surfaces hypotheses for future structured work, notably a U-shaped relationship between AI-driven speedup and displacement concern.
Methods
- Open-ended online survey of 80,508 Claude.ai personal-account users.
- LLM classifiers used to infer respondent occupation and career stage and to code sentiments (job-threat language, productivity gain magnitude and type, beneficiary of gains).
- A 1–7 self-rated productivity scale, from “less productive” to “transformatively more productive.”
- Inferred occupations linked to the “Observed Exposure” measure of Massenkoff and McCrory (2026), capturing the share of an occupation’s tasks for which Claude is used.
- Linear fits and quartile comparisons relate exposure, reported speedup, wage, and career stage to displacement concern and productivity outcomes; a robustness check restricts to the 11% of respondents who explicitly named their occupation.
Findings
- About one in five respondents expressed concern about economic displacement from AI.
- A 10-percentage-point rise in observed occupational exposure corresponds to a 1.3-percentage-point rise in perceived job threat; top-quartile exposure occupations reported displacement concerns roughly three times as often as bottom-quartile ones.
- Mean self-rated productivity was 5.1 (“substantially more productive”); only 3% reported negative or neutral effects, while 42% gave no clear productivity signal.
- Management roles (largely solo entrepreneurs) and computer/math occupations reported the largest gains; legal and scientific professions reported the most modest gains.
- Among those describing how AI helped, 48% cited expanded scope (new tasks) and 40% cited speed, with quality and cost trailing.
- When a beneficiary was identified, most named themselves; about 10% named employers or clients; smaller shares named AI companies or described net negatives.
- 80% of senior professionals reported personal benefits versus 60% of early-career workers, who were also markedly more worried about displacement.
- The link between reported speedup and displacement concern is U-shaped: both those slowed down by AI and those most accelerated by it report higher anxiety.
- Occupation could be inferred for 39% of respondents (11% explicit, 28% from context).
Connections
No other papers have been provided under shared topics, so there are no sibling notes to link here. Intellectually, the work sits adjacent to the broader Economic Index literature on task-level AI exposure and to emerging studies of AI’s distributional effects on early-career labor markets.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify · Apple Podcasts