Methodology

An occupation is "exposed" when a large share of its tasks are ones generative AI (large language models and the tools built on them) can perform or substantially speed up, or is observed being used for. The index places every occupation we cover on a 0 to 100 scale relative to the others: a higher figure means more of the work is of that kind.

Exposure is not job loss. The BLS, the International Labour Organization (ILO), Microsoft and Anthropic — the publishers of the data we use — all state this explicitly. Exposure can mean that a job is automated, that it is augmented and changes shape, or that it grows because AI makes the work cheaper. The US projections shown next to the index make this visible: several of the most exposed occupations are projected to grow.

Two sources describe what AI could do with an occupation's tasks (theoretical exposure), and two describe what people actually use AI for (observed exposure):

Each source uses its own scale, so each is first converted to percentiles across all the occupations that source covers (not only ours); tied values share the average rank. The theoretical component is the mean of the Eloundou and ILO percentiles; the observed component is the mean of the Microsoft and Anthropic percentiles. The index is the mean of the two components. When a source has no value for an occupation, the mean uses the sources that do; every occupation page shows both components separately.

Our occupations are matched to the US Standard Occupational Classification (SOC 2018) and to ISCO-08. Where official statistics have no exact category for one of our job titles, the page says so and uses the closest official category; where the exposure datasets lack a code (for example surgeons, professors, miners), the value averages the closely related codes they do cover. Military occupations are not covered by any source, so those pages show no figure.

The rank shown on each page compares the occupation with the others we cover. Industry figures are the mean index of an industry's occupations, weighted by US employment (BLS 2025) when every occupation in it has BLS data. Country figures are calculated separately: each country's employment by ISCO-08 major occupation group (ILOSTAT, ILO modelled estimates) multiplied by the average ILO Working Paper 140 score of the occupations in that group. They use the ILO's 0–1 scale, not this site's 0–100 index.

The "US labour market" box on each occupation page quotes the BLS Employment Projections 2025–35 (published 27 August 2026) without change: employment in 2025, the projected change to 2035, average annual job openings, the median annual wage and the typical entry education. It also shows the BLS's own AI exposure category (Low, Moderate, High, Very high), which BLS derives from five studies including three of our sources.

BLS projections assume technology continues to develop at its historical pace and adjust for AI only where BLS has evidence for a specific occupation; they are not a forecast of AI's effect alone. They describe the United States. Pages in other languages show the same US figures, labelled as US data.

No. It is not the chance that a job will disappear, and a high figure does not mean a high percentage risk of losing the job. No published method can honestly put a single probability on a whole occupation, because the outcome depends on employers, regulation, prices and how the work is reorganised. For a question like "what is the probability that accountants are replaced by AI", the occupation page is the better answer: it shows which tasks are exposed, which stay with people, and what official projections say.

The task definitions behind three of the four sources are American (O*NET), and the usage data comes from users of two products who do not represent the whole workforce. Theoretical ratings reflect AI capabilities at the time each study was made. Percentiles are relative: a middle figure means "middle among these occupations", not "half the work". Physical automation — self-driving vehicles, warehouse robots — is outside what these generative-AI measures capture; where BLS names such factors for an occupation we report them, and BLS currently makes no autonomous-vehicle adjustment for drivers.

Each occupation page also carries a hand-written guide: which tasks are changing, what remains human work, which skills to build and which career moves make sense. Guides are researched from public sources, reviewed on a fixed schedule and never written by an automated pipeline. Suggested career moves are hidden when the target occupation is both more exposed and projected to shrink.