Anthropic's Economic Data: Feedback & Interest Form

We have developed a privacy-preserving dataset from millions of human-AI interactions across economic tasks. Our aggregated and anonymized dataset provides unique insights into how AI systems are being integrated into different types of work and extends the replication data from our recent paper (Handa, Tamkin, et al. 2025).

As we show in the appendix, analyzing the privacy-preserving aggregate data (produced using Clio) can yield similar results to classifying conversations directly. The aggregate dataset contains ~2,000 hierarchically-clustered groups of AI interactions on Claude.ai (Free and Pro) with the following columns:

  • Cluster hierarchy (3 levels, each containing summaries of at least several hundred unique human-AI interactions from at least several hundred unique organizations):
    • cluster_name_0, cluster_description_0: Base-level clusters (most granular), representing specific task patterns  (e.g., "review business NDAs for contractors")
    • cluster_name_1, cluster_description_1: Mid-level clusters, grouping related base-level tasks (e.g., "check business contracts for typos and errors")
    • cluster_name_2, cluster_description_2: Top-level clusters (least granular), capturing broad categories of tasks (e.g., "draft, explain, and analyze legal documents and procedures")
  • Usage metrics:
    • percent_records: percentage of total records in the cluster
    • percent_users: percentage of total users in the cluster
  • Occupational mapping:
    • onet_task: O*NET task mapping for base cluster
    • onet_occupations: Comma-separated list of related O*NET occupations
    • onet_occupational_areas: Corresponding O*NET occupational areas

We are collecting feedback about this dataset and its potential research applications to understand how it might advance research in economics, labor markets, and technological change. We aim to understand the broader research community's interests and gather feedback on our data format.

Email *
Full name *
Email address *
Affiliation (optional)
Do you have any feedback about the data format described above and whether it would be useful for your research? Is there any other information not present above that you suggest we include? *
We value your input on our dataset structure and content. We'd love your feedback in particular on:

- Whether the current data format is useful for your research needs
- What additional fields or metadata would make this dataset more valuable
- What specific data aggregations or cross-tabulations would benefit your analysis
- Whether there is anything we could pre-compute (e.g., more O*NET mappings) that would make downstream analysis easier

Your feedback will help us refine the dataset to better serve the research community.
If you're comfortable sharing, what research ideas would you be interested in exploring with this data?
Understanding the kind of research you might pursue will help us better tailor our work to the needs of the research community.

Is there anything else you'd like to share?

Submit
Clear form
Never submit passwords through Google Forms.
This form was created inside of Anthropic.

Does this form look suspicious? Report

Google Forms