Anthropic Economic Index 2026: key figures
Several times a year, Anthropic analyzes anonymized conversations with Claude and counts which tasks people use AI for. The Anthropic Economic Index 2026 has three reports; according to the March report, 49% of occupations already did at least a quarter of their tasks with Claude. Below are the figures from all three reports, each with its own reference, plus the method, the limits and the conclusions for a company.
Key figures
52%
of Claude.ai conversations in November 2025 were collaboration between a person and the model, up 5 percentage points; the automation share fell by 4 percentage points to 45%
48,62%
of Claude.ai conversations in May 2026 were classed as automation and 51.38% as collaboration, by the open data of the June release
open data, release of 26 June 2026, file aei_claude_ai_2026-06-26.csv
49%
of occupations already did at least a quarter of their tasks with Claude, according to the March report
35%
of Claude.ai conversations are about tasks of computer and math occupations; programming remains the main use
42%
of Claude.ai conversations in February 2026 were personal, against 35% in November 2025; the share of study conversations fell from 19% to 12%
10%
higher share of successful conversations for users who have worked with Claude for 6 months or more
86%
of Anthropic’s respondents say they work faster; 82% see a larger volume of work, 69% higher quality and 27% savings on bought-in services
10%
of respondents think it likely they will lose their job in the next year; 38% of them link this forecast to AI
The report in brief
- Published by
- Anthropic
- Published on
- 26 June 2026
- Original title
- Anthropic Economic Index reports 2026: Economic Primitives (January), Learning curves (March), Cadences (June)
- Length and language
- three reports of 55, 20 and 33 pages and open data on Hugging Face, in English
- Checked
- 16 October 2026
- Next edition
- several times a year; Anthropic promises to announce the schedule of the next releases later
“Augmentation is once again more common than automation on Claude.ai”Anthropic, January report, p. 5
How the report was made
The index is built on anonymizeddata anonymization. Data anonymization is the processing of information about people so that nobody can tell whom it concerns without extra information. automatic analysis of conversations with Claude. The January and March reports (method sections) each took 1 million Claude.ai conversations on the Free, Pro and Max plans and 1 million API requests over one week: 13–20 November 2025 and 5–12 February 2026 (data description in the reports).
The June report uses a continuous sample of chat and Cowork conversations from 10 April to 10 June 2026 and the first Anthropic Economic Index Survey: about 9,700 respondents whose answers are linked to their usage data (June report, ch. 3). The open data is published on Hugging Face under the MIT license.
What the report does not measure
The reports describe only Claude users, not the whole economy and not all AI users. API data is not split by country. A model, not people, classifies the tasks and work modes.
The survey is self-reported and skewed toward IT staff and managers: women make up 12% of the linked sample (June report, ch. 3). The reports do not measure actual employment or wages.
What this means for a company in Russia
Russia in the report: the reports have no data on Russia. It is absent from the open data of both the March release (178 countries) and the June release (121 countries), while Kazakhstan and Ukraine are there. The January report marks countries where Claude is not available in gray on its map (p. 37, footnote 3).
- In 49% of occupations, at least a quarter of tasks were already done with Claude (March report, p. 7). The reports have no data on Russia, so assess not whole occupations but single tasks in your processes. The AI solution map helps break processes down into tasks.
- Users with 6 months of experience or more solve tasks better: their share of successful conversations is 10% higher (March report, p. 4). According to the report, the result depends on experience, not only on access to the model. Give staff basic training: you can start with free AI courses.
- In November 2025, collaboration made up 52% of conversations and automation 45% (January report, p. 5), and by the open data for May 2026 automation was already 48.62%. The more tasks the model does alone, the more important it is to check the result. Build an agent quality check into the project.
What to compare it with
- NBER: how people use ChatGPT: the same approach on ChatGPT conversations, an NBER study
- Microsoft: AI diffusiondiffusion model. A diffusion model is a neural network that creates an image, a video or a sound by removing noise from a random picture, step by step. 2026: the share of people using generative AI by country, including Russia, from Microsoft telemetry
- ILO: generative AI and jobs 2025: exposure of occupations to generative AI as estimated by the ILO
Sources and dates
- Anthropic, Anthropic Economic Index, checked 16 October 2026anthropic.com
- Anthropic, Anthropic Economic Index: Economic Primitives (PDF, January 2026), checked 16 October 2026www-cdn.anthropic.com
- Anthropic, Anthropic Economic Index: Learning curves (PDF, March 2026), checked 16 October 2026cdn.sanity.io
- Anthropic, Anthropic Economic Index: Cadences (PDF, June 2026), checked 16 October 2026cdn.sanity.io
- Anthropic, Economic Index open data on Hugging Face, checked 16 October 2026huggingface.co