Announcing the Anthropic Economic Index Survey

economic indexsurveysanthropic interviewerlabor marketqualitative datamethods

A short announcement of a monthly survey, run through Anthropic Interviewer with a rotating random sample of Claude users, that asks how AI is changing their work and what they expect, intended to add first-hand accounts to the Economic Index's usage data and catch changes before they appear in labor statistics.

Anthropic (Economic Research team) · 2026-04-22 · original

By April 2026 Anthropic's Economic Index had reported for over a year on what people use Claude for, drawn from conversation data. This announcement, from Anthropic's Economic Research team, launches a different kind of data source: a recurring survey that asks Claude users directly how AI is affecting their work. It was published on 22 April 2026, the same day as a companion report on economic themes in an earlier survey of about 81,000 users, What 81,000 people told us about the economics of AI. The post is short and reports no results; this note describes what it says the survey is for and how it will run.

Why a survey

The post's argument is that the existing quantitative sources leave a gap. Usage and diffusion metrics show how AI is being deployed, and labor market indicators such as employment, wages and layoffs show what has already happened, "often with meaningful delay". Neither, the post says, captures how people experience the change or what they expect as capabilities advance. Given what it calls "substantial uncertainty" about AI's effects on jobs, productivity and unemployment and their timing, the team wants to hear from people "on a cadence that can identify changes as they emerge".

The stated aim is "a rich new corpus of qualitative data". Repeating the survey monthly is meant to measure not only what people report but how quickly their views shift. Combined with Claude usage data, analyzed through the privacy-preserving methods of Clio (to which the post links), the first-hand accounts are expected to "surface change before it shows up in aggregate labor market data".

How the survey works

  • Instrument. The survey is conducted through Anthropic Interviewer, Anthropic's tool for running interviews with Claude users, introduced in Introducing Anthropic Interviewer (December 2025).
  • Sample. Each month a small, randomly selected group of Claude users is invited. Anyone with a personal account at least two weeks old is eligible, and the sample rotates each month to reach as broad a range of people as possible over time.
  • Invitation. Invited users see a banner on claude.ai or in the Cowork desktop app, or receive an email if they use Claude only on mobile (FAQ).
  • Questions. The interview asks about the respondent's work, changes they have noticed in their role or organization, how they expect that to shift over the next year, and how they hope an AI-shaped economy will look in ten years (FAQ). The body of the post lists the topics as tasks being handed off, productivity gains, shifts in hiring and roles, expectations, and what a well-handled transition would look like.
  • Use of the data. Findings will appear in future Economic Index reports and research briefs, and the data will also be used to improve Anthropic's models and services. Processing follows a supplemental privacy policy for Anthropic Interviewer, and de-identified responses may be quoted from users who opt in (FAQ).

Limitations and open questions

The post does not describe sample sizes, response rates, weighting, or how open-ended answers will be coded. Two limitations follow from its design as described. The sample is drawn only from personal Claude accounts, so it excludes non-users and enterprise accounts; the companion 81,000-person report names this restriction as a caveat for its own findings. And the interviews are run inside Claude's own apps by the company whose product's economic effects are being studied; the post does not discuss whether this setting affects answers.

Research questions

  1. How representative is a rotating sample of personal Claude accounts? Comparing respondents' reported occupations and demographics with national labor force data would show which groups the survey over- or under-represents.
  2. Do survey responses lead or lag official labor statistics? The post's central claim is that first-hand accounts can show change early. Once a year of monthly data exists, comparing trends in reported job threat or hiring changes by occupation with later official employment data would test it.
  3. Does an AI interviewer change what people say about AI? Running the same questions through Anthropic Interviewer and through a conventional web form with randomized assignment would measure any mode effect.
  4. How stable is the coding of open-ended answers? The post does not say how answers will be coded; the companion report coded its interviews with Claude-powered classifiers. If the survey does the same, comparing classifier labels with human coding on a sample, month by month, would show whether measured shifts reflect respondents or the coding.

How it connects

References

  • Anthropic (2026). Announcing the Anthropic Economic Index Survey. Economic Research, 22 April 2026. https://www.anthropic.com/research/economic-index-survey-announcement