AI exposure by occupation, 2026
Exposure varies more between occupations than any headline average can convey — from interpreters above 98% to roles where the measured effect is close to zero. This page collects the per-occupation figures, and separates two things that get conflated constantly: the share of tasks a model can do, and the share of jobs that actually disappear.
Thank you for reading this post, don't forget to subscribe!
Every figure below is an exposure or risk estimate, not a count of jobs lost. An occupation with 46% task-automation share does not lose 46% of its jobs — it loses some headcount and reshapes the rest around whatever the model cannot do. These numbers are modelled projections from varying methodologies, they are not directly comparable with one another, and they have a consistent history of being directionally right and numerically too aggressive. Treat them as a ranking, not a forecast.
Exposure by occupation
| Occupation | Exposure estimate | What has measurably happened | Confidence |
|---|---|---|---|
| Interpreters & translators | 98%+ automation exposure | The highest exposure of any occupation measured. Machine translation quality crossed the commercial-acceptability threshold for most routine content; surviving demand is concentrated in legal, medical, literary and certified work where accuracy is liable. | MEDIUM |
| Office & administrative support | 46% task-automation share | Highest task share of any category. Data entry is the most-cited “doomed” role in 2026 listings. Scheduling, routing and document handling absorbed; exception handling and office coordination retained. | MEDIUM |
| Legal support & paralegals | 44% task share; paralegals 80% risk by 2026; legal support 75–95% by 2027 | Document review that took paralegal weeks now takes hours with tools like Harvey and Casetext. Legal researchers estimated at 65% risk by 2027. Privileged judgement, client advice and court appearance unaffected. | MEDIUM |
| Content & copywriting | Freelance postings −33%; marketing content writer roles projected −50% by 2030 | The clearest measured displacement outside translation. Bloomberry’s analysis of 5M freelance postings shows the drop since ChatGPT launched. Content mills largely replaced; original reporting, investigation, brand voice and strategy retained. | MEDIUM |
| Market research & analysis | 53% of analyst tasks by 2026 | Data gathering, cleaning and first-pass synthesis absorbed. Framing the question and defending the conclusion to a decision-maker retained. | MEDIUM |
| Bookkeeping & basic accounting | 35–50% reduction in basic bookkeeping roles by 2028 | Transaction coding, reconciliation and routine compliance automate well. Accountants doing analysis, judgement and sign-off are materially less exposed — the liability sits with a named human. | MEDIUM |
| HR & recruiting | 30–40% of HR roles automatable at low effort (Bersin, Apr 2026) | The best-documented counter-example. IBM automated ~94% of routine HR tasks, hit a residual 6% requiring ethical judgement, and moved to triple US entry-level hiring. 62% of companies expect AI to run hiring end-to-end by end-2026, but Korn Ferry finds 52% of talent leaders are adding AI agents rather than replacing recruiters. HR AI adoption doubled 26%→43% in a year. | MEDIUM |
| Customer support | Median tier-1 deflection 41.2% | Most automated and most reversed. Full analysis → | MEDIUM |
| Software engineering | Entry-level postings −67% | Output absorbed, verification became the constraint. Full analysis → | MEDIUM |
The pattern across all nine
Line the occupations up and the same structure appears every time. What gets absorbed is production; what survives is judgement, exception handling and liability. The paralegal loses document review and keeps privileged advice. The bookkeeper loses reconciliation and keeps the sign-off. The translator loses routine content and keeps certified, legally consequential work. The support agent loses password resets and keeps the complaint that could cost the account.
That is why task-automation share is a poor predictor of headcount loss. Office and administrative support has the highest measured share at 46% — and it is not the occupation with the highest observed job loss, because the residual 54% still requires people present and accountable. Meanwhile translation, which has a narrower task bundle almost entirely inside the model’s competence, shows far greater real displacement at a similar headline number.
The occupations at genuine whole-role risk share one property: nearly the entire task bundle is routine, verifiable and low-liability. Where any meaningful fraction of the work requires someone to carry consequence, the role compresses rather than disappears.
The aggregate picture
| Measure | Figure | Period | Confidence |
|---|---|---|---|
| US job losses attributed to AI — 4.5% of all US layoffs | ~55,000 | 2025 | MEDIUM |
| US job cuts citing AI, first four months | 49,135 | 2026 | MEDIUM |
| Share of US cuts citing AI in the month it became the #1 stated reason | 25% | Mar 2026 | MEDIUM |
| Tracked tech layoff events citing AI or automation (173 of 322), affecting ~170,945 workers | 54% | 2026 | MEDIUM |
| Employers who regret AI-related layoffs | 55% | 2026 | MEDIUM |
| Roles cut for AI already refilled | 32% | 2026 | MEDIUM |
One caution on all of these: “AI” is a convenient public reason for a layoff. It reads as strategic modernisation rather than weak demand or over-hiring, which gives employers an incentive to over-attribute. Read attribution figures as a ceiling. The 2026 NBER executive survey — which asks about realised effects rather than stated reasons — reports limited employment and productivity impact over the preceding three years even at high stated adoption.
Frequently asked
Which job has the highest AI exposure?
Interpreters and translators, at over 98% automation exposure — the highest of any occupation measured. By task-automation share the leader is office and administrative support at 46%, followed by legal work at 44%.
Is writing being replaced by AI?
Commodity writing, yes. Bloomberry’s analysis of 5 million freelance postings found writing listings down 33% since ChatGPT launched, and digital marketing content writer roles are projected to fall 50% by 2030. Original reporting, investigative work, brand voice development and content strategy have not shown the same decline.
Will AI replace accountants and paralegals?
It is replacing the routine layer of both. Basic bookkeeping roles are projected to fall 35–50% by 2028, and legal support roles carry a 75–95% automation risk estimate by 2027, with paralegals at around 80%. In both professions the work that survives is the part carrying personal liability — the sign-off, the privileged advice, the appearance.
Does a high task-automation share mean my job will go?
No. Task-automation share measures tasks, not headcount. Office and administrative support has the highest measured share at 46% and has not seen proportional job loss, because the remaining 54% still needs accountable people. IBM automated roughly 94% of routine HR tasks and responded by tripling entry-level hiring, because the residual 6% required judgement it could not automate.
Methodology & sources
Exposure percentages come from different studies using different methodologies — some measure task share, some model whole-role risk, some project forward to a stated year. They are not directly comparable and are presented as a ranking rather than a set of equivalent measurements. All are rated MEDIUM, and several were gathered via search-engine summaries rather than opened at source. Layoff attribution is employer-stated and treated as an upper bound. Sources include SHRM, Josh Bersin (Apr 2026), Korn Ferry, Gartner, Bloomberry, OECD Skills in the AI Age (Jul 2026) and the 2026 NBER executive survey. Quarterly cadence. Next review: December 2026.
Part of the Will AI Replace My Job? pillar. See also The AI replacement reversal and the AI Jobs Report 2026.