Survey of 81,000 Claude users: 20% of respondents are worried about losing their jobs

null

Author: TinTinLand

Why are the people who benefit the most from AI actually the most worried about losing their jobs?

On April 22, Anthropic released a survey report covering 81,000 real users of Claude—“What 81,000 People Told Us About the Economics of AI”—aiming to reveal the true situation and mindset of ordinary people under the wave of AI.

The core conclusions mentioned in the report are as follows:

The deeper AI intervenes in a profession, the stronger the employment anxiety among practitioners, especially for newcomers;

The highest and lowest income groups experience the most significant productivity gains. And most of this improvement is not “doing things faster,” but “doing things that were previously impossible”;

Those who achieve the highest efficiency improvements through AI are actually the most deeply anxious about their career prospects.

TinTinLand has provided an in-depth translation and analysis of this latest research on AI, economics, and survival.

🤔 Who is worried about unemployment?

One-fifth express concern

“Like all white-collar workers today, I am almost constantly worried that my job will be replaced by AI.” — a software engineer

Among respondents, about one-fifth explicitly expressed concerns about economic unemployment.

One software developer said, “At this stage of AI, it’s very likely to replace entry-level positions.” Others lamented that their work content is being eroded by automation.

A market researcher stated, “Undoubtedly, AI has enhanced my capabilities. But in the future, it might replace my job.”

In some roles, the arrival of AI even makes work more difficult. One software developer observed, “Since AI appeared, project managers have been assigning us increasingly difficult tasks and bugs.”

Data validation

In this report, we use Claude to infer respondents’ attributes and emotions from their answers. For example, many respondents mention their profession or provide details about their work and life, allowing us to deduce their job categories. Similarly, we quantify “unemployment concern” by having Claude identify and interpret respondents’ direct statements about “the risk of AI replacing their position.”

Research shows that respondents’ subjective perception of AI threats is highly correlated with their job’s “observed exposure level.” Exposure level refers to the proportion of tasks in a profession that are actually handled by AI.

For example, elementary school teachers are significantly less worried about being replaced than software engineers, which aligns perfectly with the reality that programming tasks dominate Claude’s traffic.

As shown in Figure 1, the vertical axis indicates the proportion of respondents in a certain profession who believe AI has already replaced or is very likely to replace their jobs in the near future; the horizontal axis represents the “observed exposure level.”

For every 10 percentage points increase in exposure, perceived job threat rises by 1.3 percentage points. The group with the highest 25% exposure reports a worry frequency three times that of the lowest 25%.

Figure 1: Job Threats from AI and Observed Exposure Level

Younger people are more anxious

Career stage is a key variable influencing anxiety. Previous studies have observed signs of slowdown in hiring among recent graduates and early-career professionals in the U.S.

In this survey, we also found that early-career practitioners are far more panicked about unemployment than seasoned professionals.

Figure 2: Unemployment Concerns at Different Career Stages

Who benefits from AI?

Most people feel productivity has increased

We scored respondents’ self-reported productivity improvements on a scale from 1 to 7: 1 means “decreased efficiency,” 2 means “no change,” and each higher level indicates greater improvement.

A typical 7-point response: “It used to take months to build a website; now I can do it in 4-5 days”;

Score of 5: “What used to take about four hours now takes half an hour”;

Score of 2: “AI helped me fix some code, but I had to try several times to get the desired result.”

The average score is 5.1, indicating “significantly more efficient.”

Of course, these respondents are active Claude users who participated in the survey, so they are more likely to perceive productivity gains than average users. About 3% reported negative or neutral effects, and 42% did not explicitly mention productivity changes.

High earners benefit the most

This result shows some differentiation based on income levels.

The left side of Figure 3 shows that high-paying professions (such as software developers) experienced the greatest productivity boosts. This trend holds even when excluding computer and math-related jobs.

In tasks requiring higher education levels, Claude tends to significantly shorten the time needed to complete tasks (compared to not using AI).

But a detail worth noting is that the benefits for lower-wage jobs are also substantial. For example, a customer service representative quickly generated responses with AI, saving a lot of time; a courier used Claude to start an e-commerce business; a gardener developed a music app. AI is opening doors for lower-educated, lower-income individuals that were previously inaccessible.

Figure 3: Productivity Gains by Profession (Inferred)

We further break down this result on the right side of Figure 3.

Management roles rank highest, mostly entrepreneurs using Claude to start businesses. Next are computer and math professions, including software developers. The groups with the most moderate productivity gains are researchers and legal professionals.

Some lawyers worry whether AI can accurately follow complex instructions: “I’ve given very specific rules, including content placement, how to interpret legal documents, and desired operations… but it always goes off track.”

Where do the benefits flow?

As AI spreads in the economy, a key question is: who ultimately gains from these benefits—workers themselves, managers, consumers, or companies?

Overall, most believe the benefits go to themselves: tasks are completed faster, more things can be done, and they gain more free time.

However, about 10% of respondents feel that these dividends are “harvested” by employers or clients—delivering more output in the same amount of time. A small number also mention that AI companies benefit from it.

This difference is also related to career stage: only 60% of early-career workers see themselves as beneficiaries of AI dividends, compared to 80% of seasoned professionals.

Figure 4: Where Do AI Productivity Gains Flow?

Where are efficiency improvements seen?

“I achieved things I couldn’t do before”

Respondents shared in which aspects they felt productivity had improved. We categorize these into four dimensions: scope, speed, quality, and cost.

Analysis shows that among all respondents who explicitly mentioned productivity changes, the most common improvement was “expanding work scope,” accounting for 48%; 40% emphasized speed improvements.

For example, many programmers using AI said: “I wasn’t a technical person before, but now I can do full-stack development.” This is an expansion of scope—AI unlocked new capabilities for them.

Others achieved faster results on existing tasks, such as an accountant who said: “I built a tool that completes a financing task in 15 minutes, which used to take two hours.”

Quality improvements often involve more thorough and detailed checks of code, contracts, and documents. A few respondents also mentioned AI’s low-cost advantage.

Figure 5: Types of Productivity Gains Reported by Users

Faster work, greater job insecurity

Research shows a U-shaped relationship between AI’s speed enhancement and perceived job threat (see Figure 6).

The slow group (tasks slowed down): mainly creative workers (writers, artists), who believe AI’s rigidity limits creativity, but also worry that low-quality AI content flooding the market will squeeze their survival space.

The fast group (rapid improvement): when task completion time shrinks from hours to minutes, users feel a strong sense of insecurity—if work becomes so easy, what is their long-term value?

Figure 6: Relationship Between Job Threat and Changes in Speed

Conclusion: What can we learn from this?

Perceptions align closely with data

Data shows that people’s perceptions are consistent with actual usage data: the more tasks Claude can handle, the more worried people are about AI’s impact.

Additionally, early-career individuals exhibit higher economic anxiety, consistent with previous research findings.

AI empowers, but anxiety is real

At the same time, the survey reveals the other side of the coin: AI is genuinely expanding people’s capabilities.

While high-income groups are most positive about productivity gains, lower-income and less-educated groups also report significant efficiency improvements. Most respondents believe Claude has enhanced their abilities by expanding scope or speeding up execution.

But this does not eliminate anxiety. Those who benefit most are often the most uneasy—because they understand better than anyone what AI can do.

Limitations and outlook

It’s important to note some limitations of our analysis:

Respondents are active Claude users, more likely to perceive personal benefits from AI; job and career stage information is inferred from open responses, which may have errors; additionally, the survey relies on open-ended questions, so results depend on what respondents chose to mention.

Nevertheless, the 80,508 Claude users’ mention of economic anxiety in the report is a signal that cannot be ignored.