The 20 least AI-exposed occupations

Why this page does not say safe

A low score is a narrow, specific finding: across the conversations this study covers, the assistant it measured was not being used for the core activities of this occupation. Three large things sit outside that finding.

Physical automation is not in frame. The measure covers generative AI, meaning language, analysis and images. It does not cover machinery. A low score for equipment operators or vehicle work says nothing about autonomous systems, and industrial robots have an employment record of their own. Hands-on and licensed jobs sets out what that record actually shows, with the caveats its authors attach.

Wages and headcount are not in frame. Nothing here is weighted by pay or by how many people hold the job. An occupation can sit at the very bottom of this list and still be shrinking, or badly paid, for reasons that have nothing to do with AI.

A zero is a floor, not a verdict. That one needs its own section.

What a zero actually means

The zeros in the table above belong to 8 of the 20: Bridge and Lock Tenders, Dredge Operators, Floor Sanders and Finishers, Foundry Mold and Coremakers, Orderlies, Pile Driver Operators, Rail-Track Laying and Maintenance Equipment Operators, and Water and Wastewater Treatment Plant and System Operators. That does not mean an AI assistant was tried on their work and failed. It means their activities fell below the study's 0.05% activity-share threshold in the sample, so they never came up often enough to be measured.

The floor is not a manual-work phenomenon either. Across the whole dataset, 205 of the 332 measured work activities score zero on the user-goal series, and they are not confined to physical work. How to read your exposure score sets out the cut-off and how much weight a zero can carry.

The bottom of the list is not a natural boundary. The next occupation up, Massage Therapists at 0.012, sits 0.0001 above the 20th on this list. Rounded to three decimals they are the same number, which is the whole reason this site leads with bands rather than decimals.

Which fields the bottom 20 come from

Construction & Extraction supplies 7 of the 20. The other 13 are spread across 7 more fields, out of 22 in total.

Of the 8 fields listed, 6 have a median occupation above the Minimal band, so putting someone in the bottom 20 is not the same as being a low-exposure field. The Minimal band holds 157 occupations, one fifth of the 785, and 6 of the 22 fields put nobody in it: Computer & Mathematical; Architecture & Engineering; Community & Social Service; Education & Library; Arts, Design, Media & Entertainment; and Sales.

Low scores do not track credentials

Within Legal, Paralegals and Legal Assistants score 0.057 and sit in the Minimal band, while Lawyers score 0.180 in the Elevated band. Healthcare splits the same way: 14 of the 17 occupations in Healthcare Support sit in the Minimal band, against 16 of 69 in Healthcare Practitioners, a field that also puts 4 occupations in the High band. Whatever this measure tracks, it is not qualification level.

What to do with a low score

On its own, not much. The band places you against 784 other occupations on one measure. It is not a career plan and it does not extend to the next decade. The useful step is the same at either end of the distribution: look at which of your own activities the data does and does not cover. Your occupation page breaks the score into exactly that list. If the occupation title does not match what you actually do, and for a lot of people it does not, the task self-assessment works from your week instead.

The other end of the same distribution is on the 20 most exposed occupations, and the full ranking holds all 785.