How this is calculated

Every figure on this site comes from published India rows. Here is how they become your exposure and risk score.

Where the data comes from

The Anthropic Economic Index (release_2026_06_26, CC-BY) publishes how people use Claude, one AI assistant, by country and by work task. We use only the rows for India, and the latest month, May 2026. Figures for Indian states are also from this release. Pulled 2026-10-15.

Each of our 50 job titles is matched to one job in the U.S. O*NET 31.0 Database (version 31.0), which lists that job's tasks. Anthropic labels conversations with the same task IDs, so the two can be joined.

How tasks become work groups

Anthropic publishes figures for 31 groups of work activity for India. O*NET links every task to detailed work activities, and those activities sit inside these groups. A job's default work mix weights each task by its O*NET importance rating, split equally across the groups the task belongs to. Groups India does not publish figures for are left out, and the rest are re-scaled so the mix adds up to 100%. If you change your mix in the report, your weights replace the default ones.

So a work mix is a share of a job's tasks weighted by how important they are, not a measure of time spent. Where a title fits its O*NET job only loosely, its page names that O*NET job and says it is a rough match.

The numbers, in plain words

For each work group in your mix we use four India figures: its weight in your mix, how much of it is measured, how much of the measured use is Claude doing the work (automation), and how much is Claude helping a person (augmentation). We also use how much freedom Claude is given on the group (autonomy, from 1 to 5).

• AI does it adds up, across your groups, weight x measured share x automation share.
• AI helps adds up weight x measured share x augmentation share.
• Exposure is AI does it plus AI helps. Because automation and augmentation add up to 100% in every India row, this is the weighted share of your work that falls on tasks with published AI use.
• Risk score takes weight x measured share x automation share x (autonomy / 5) for each group and adds them up. That total is divided by the highest total among our 50 jobs' default mixes and scaled to 100, capped at 100. Bands: Low 0 to 24, Moderate 25 to 49, High 50 to 74, Very high 75 to 100.
• What stays human lists your work groups with the lowest measured automation, using the share of tasks only a person can do to break ties.

The same inputs always give the same score. The score is relative to our 50 jobs, not a probability.

States

State figures are share of use only: how big a work group is in a state's Claude use, compared with India's. The release publishes no automation, autonomy or time figures for states. Where a state has no published share for a group, the report says so and shows India's share.

Limits you should know

• It measures use of Claude only, not all AI, and not what AI does on its own.
• It uses India rows only. No other country fills a gap.
• It reflects one data month, May 2026. Small moves may be noise.
• It is not a forecast of hiring, layoffs or the future of any job.
• The risk score is a fixed formula. It has not been checked against real job outcomes.
• Claude users in India are not a cross-section of Indian workers.
• O*NET task lists describe U.S. jobs. Some Indian titles fit them loosely, and some of our titles share one O*NET job, so they show the same numbers.
• A task or group with no published India row is shown as not published, never as zero. Anthropic does not publish figures below its privacy and sample-size thresholds.

How each title was matched

Sources and licences

• Anthropic Economic Index, release_2026_06_26, month 2026-05, India rows. CC-BY. Pulled 2026-10-15.
• This site includes information from the O*NET 31.0 Database by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA). Used under the CC-BY 4.0 license. O*NET® is a trademark of USDOL/ETA. We joined it with India usage data; USDOL/ETA has not approved, endorsed or tested these changes.