Press and data

Key figures

Three findings worth writing about

1. Almost nothing scores as high risk

Of 423 ranked occupations, 1 occupation clears the high risk threshold of 70. 66 sit in the transitioning band (40 to 69) and 356 are in the lower risk band (below 40). The mean is 26. The single most exposed occupation in the index is Market Research Analysts and Marketing Specialists at 70.

Read carefully, the aggregate research is not saying any occupation is close to fully automatable today. It says a meaningful slice of tasks in a minority of occupations is now within reach of the tools.

2. The studies disagree far more than the coverage suggests

The median gap between the lowest and highest published estimate for the same occupation is 24 points on a 100 point scale, and 151 occupations have a gap of 30 points or more. This is the least reported thing in the field and, we would argue, the most important.

Restricted to occupations with at least four independent sources, so the gap reflects real disagreement rather than two studies being far apart by chance.

3. Sector is a weak predictor, task mix is a strong one

By sector mean, administration is the most exposed at 40 and construction & trades the least at 13. But the range inside a sector dwarfs the range between sectors. Healthcare contains both some of the most exposed occupations in the index and some of the least, tens of points apart, because the exposure follows the task content of a role rather than the industry it sits in.

How to cite this

Quoting the figures in articles, research and presentations is free, with attribution and a link. A suggested citation:

Two requests, both about accuracy rather than credit. Please date the figures, because scores change as new research is ingested and the live index is always the current version. And please describe a score as exposure of an occupation to current AI capability, not as a probability that a job disappears, because that is what the underlying sources measure.

Need a cut of the data we do not publish, a sector breakdown, or a comment for a piece? Email willaitakemyjob@proton.me.

What the data supports, and what it does not

Stating the limits plainly, because these numbers are easy to overclaim.

It supports

• Comparing occupations against each other on task level AI exposure
• Showing which kinds of work the research consistently places above or below average
• Quantifying how much the published research disagrees about a given occupation

It does not support

• Predicting how many jobs will exist in a given occupation in future
• Claims about a specific employer, region or individual
• Reading a score as a probability, a percentage of tasks automated, or a timeline
• Treating a score built on two sources as equal in confidence to one built on twelve, which is why the source count is published on every occupation page

The weighting scheme is ours: more recent and higher confidence estimates count for more. A different scheme would move individual numbers. The full method is on the methodology page.

The sources

29 sources sit behind the index, ranging from peer reviewed academic work to government statistics and consultancy analysis. By number of occupations covered, the largest contributors are:

Coverage is uneven: a small number of large datasets account for most of the index, and depth of independent corroboration varies by occupation. Every occupation page lists its own sources.

About the project

AI Job Risk is an independent project built and maintained by one person. It is not funded by any of the organisations whose research it aggregates. The public index and every occupation page are free with no account required. A personalised report is sold as a paid product, which is how the project is funded.

More on the background and motivation is on the about page.