81,000 users reveal how AI is changing the economy | Keryc
Anthropic published a survey of 81,000 Claude users that links what people ask the model with how they feel about their work and the economy. What happens when real AI usage is measured alongside people’s voices? Here’s the essential takeaway, explaining the technical methodology without losing the practical side.
Key findings
Observed exposure to Claude (observed exposure) correlates with greater worry about job displacement: the more the model appears in tasks for a given role, the higher the job anxiety. For every 10 percentage points in exposure, perceived threat rises by 1.3 percentage points.
Workers in the highest and lowest income percentiles reported the biggest productivity gains, especially from expanded scope (being able to do new tasks).
There’s a U-shaped relationship between the speed AI brings and perceived threat: people who feel AI slows them down are worried, but for most, more acceleration also means more unease about the role’s future.
Early-career workers show higher levels of concern than senior professionals.
Methodology and metrics (technical detail)
The survey mixes open-ended responses with classifiers powered by Claude to infer attributes. Here’s what matters technically:
observed exposure: percentage of tasks in a job where Claude appears in observed traffic. It’s the main metric to estimate displacement risk.
Automated inferences: occupation and career stage were deduced from open responses (for example, “last thing I used with a chatbot”). That lets researchers group 81,000 accounts, but it introduces uncertainty in each person’s label.
Productivity measured on an adapted Likert scale (1 to 7): 1 = less productive, 4 = neutral, 7 = transformatively more productive. The mean was 5.1 ("substantially more productive").
Sentiment classification and quote extraction: Claude was used to find passages where respondents express worry about job loss and to tag the type of improvement (scope, speed, quality, cost).
For example: occupations with more use in coding tasks (software) show higher observed exposure and, in parallel, more mentions of replacement risk.
What people say (examples and patterns)
A software engineer mentioned constant worry about replacement at junior levels.
A customer service rep said AI saves them a lot of time by drafting replies from examples.
A driver used Claude to start an ecommerce business; a landscaper built a musical app thanks to AI.
These testimonials show AI isn’t just a direct accelerator: often it expands what you can do (scope), and other times it speeds up how long tasks take (speed).
Where gains are concentrated
The largest group reporting benefits was management (many entrepreneurs and solopreneurs use Claude to start businesses).
Computer & math (including developers) show substantial improvements: not only for coding, but also for tasks that require advanced training.
Surprisingly, there are also big gains in some lower-income jobs: customer support assistants, operators who use AI for side technical projects, etc.
By type of gain: 48% mentioned expanded scope, 40% speed, and the rest cited quality improvements or cost reductions.
Interpretation: why gains and fears coexist
The evidence suggests people notice where AI is already working harder, and that's where anxiety concentrates. At the same time, many users report personal benefits: faster tasks, new capabilities, freed-up time.
Two key points:
The biggest productivity jumps are associated with greater worry about a role's future. That makes economic sense: if the same task takes less time, demand for labor can fall.
Benefits aren’t evenly distributed: there are signals that high earners gain large advantages, but there are also clear examples of low-wage workers benefiting significantly.
Important limitations (methodological)
Sample bias: only users with personal Claude.ai accounts who opted in responded. That can skew results toward more favorable or self-selected experiences.
Automated inferences: occupation and career stage were inferred from free text; labeling errors are possible.
Open responses: findings depend on what people chose to mention spontaneously, not on structured questions about every topic.
Lack of enterprise coverage: business users may experience a different distribution of benefits between employees and employers.
Practical implications for professionals and organizations
If you’re a manager or product leader: measure observed exposure of critical tasks. Where AI takes a larger share, plan upskilling and job redesign.
If you’re early in your career: recognize the advantage AI can give you, but also the need to develop skills that won’t be easily automated (management, strategic thinking, supervising AI).
For companies and policymakers: use structured surveys alongside usage traffic to estimate where displacement risk accumulates; fund training programs aimed at tasks, not just roles.
Recommended metric: combine percentage of tasks automated (task share) with periodic surveys on speed/scope to anticipate changes in labor demand.
Final reflection
This survey links usage data with real voices and shows a complex picture: AI is empowering many people while generating anxiety where its use is higher. It’s not a single prediction about the future of work; it’s a snapshot you can use to design concrete responses: education, policy, and job redesign.
