Roughly two-thirds of U.S. occupations are exposed to some degree of automation by AI
Assessment
Evidence favors the claim, but the chain is incomplete or the sources are secondary.
The figure comes from Goldman Sachs Global Investment Research (Briggs and Kodnani, 2023), which mapped occupation-level task data from O*NET and ESCO against AI capabilities and found that roughly two-thirds of U.S. occupations have at least some tasks AI could perform. Independent analyses using their own task-level methods bracket the figure: the finding that about 80 percent of U.S. workers could have at least a tenth of their tasks affected by LLMs, now assessed as a credible estimate in its own right, lands above it, and the IMF's estimate that about 60 percent of jobs in advanced economies are exposed to AI lands somewhat below it.
The claim measures technical exposure, the overlap between what AI can do and what occupations involve, not realized job loss or displacement. Its truth is conditional on whether task-based analysis of occupational databases can meaningfully estimate AI exposure, a methodology all the converging estimates share and which carries live critique. Within that framework, "roughly two-thirds" is a fair central estimate; a credible showing that the task-mapping approach materially misclassifies exposure at scale would be the main development that could unsettle it.
Full reasoning: the evidence and decisions behind this verdict
The claim is a derived estimate, so the assessment turns on the primary analysis and on whether independent work corroborates its magnitude. The primary source is the Goldman Sachs Global Investment Research report "The Potentially Large Effects of Artificial Intelligence on Economic Growth" (Briggs and Kodnani, 2023-03-27, www.gspublishing.com/content/research/en/reports/2023/03/27/d64e052b-0f6e-45d7-967b-d7be35fabd16.html), which states the claim verbatim from an occupation-level distribution of the share of tasks AI can perform, built on O*NET (U.S.) and ESCO (Europe) task data. Both recorded instances affirm; no credible source found denies the figure as a characterization of task-overlap exposure.
Corroboration: the Eloundou et al. finding that about 80 percent of U.S. workers could have at least 10 percent of tasks affected by LLMs (arxiv.org/pdf/2303.10130) has now been assessed as supported (credence 0.75): a credible estimate of potential task exposure, corroborated in direction by independent indices, though resting on a single study's rubric and measuring technical potential rather than realized impact. It uses a different unit (workers, not occupations) and a narrower technology scope (LLMs only), yet lands above the two-thirds figure; the IMF's 60 percent estimate for advanced economies (Cazzaniga et al. 2024, www.imf.org/-/media/files/publications/sdn/2024/english/sdnea2024001.pdf), still unassessed, lands somewhat below it. The Goldman figure sits inside the bracket these independent estimates form, which is what "roughly two-thirds" needs. That one leg of the bracket now stands assessed slightly strengthens the corroboration, reflected in modestly higher confidence and credence than the prior pass; it does not change the status.
Why supported rather than verified: the claim's truth is conditional on the task-overlap methodology it assumes, that task-based analysis of occupational databases can meaningfully estimate AI exposure, which remains unassessed and carries live methodological critique (an ILO research brief notes exposure measures can overstate threat because workflows are redesigned rather than tasks automated one-for-one; other work argues capability judgments in these measures are partly speculative). The caveats the corroborating subclaim's own assessment names (single study's rubric, unreplicated ratings, potential not realized impact) were already priced into this verdict and remain so. I did not re-derive the occupation-level distribution myself, and the underlying report's full task-scoring rules were not independently checked. Credence 0.77 reflects that "roughly two-thirds" is a fair central estimate under the standard methodology, with residual uncertainty from unit and scope differences among corroborating studies and from post-2023 capability advances, which would if anything push the exposed share up, still consistent with "roughly two-thirds" as a floor-like reading but a source of drift for the point estimate.
What would change the conclusion: a credible replication showing the O*NET task-mapping materially misclassifies exposure at scale, a consensus successor measure putting the some-exposure share for U.S. occupations well outside the 55 to 85 percent range, or an assessment of the shared methodological assumption finding it unsound.
Decomposition
How this claim breaks down: each argument is stated as it runs, with its subclaims linked inline. ↗︎ opens a subclaim; the map shows how they fit together.
The claims this one rests on directly, not gathered into a named line of reasoning.
- assumesbackground the parent's framing takes as givensteward instructions →Task-based analysis of occupational databases can meaningfully estimate occupations' exposure to AI automation ↗︎
Independent research teams using their own task-level methods reach exposure shares of the same order: because about 80 percent of US workers could have at least 10 percent of their tasks affected by LLMs and about 60% of jobs in advanced economies are exposed to AI, the estimate that roughly two-thirds of U.S. occupations carry some degree of AI exposure sits inside the range that separate analyses converge on.
The inference goes through as corroboration rather than proof: the independent estimates use different units (workers versus occupations), scopes (LLMs only versus AI broadly, advanced economies versus the U.S.), and capability judgments, so they bracket the two-thirds figure rather than confirm it exactly. Of the two legs, the estimate that about 80 percent of US workers could have at least a tenth of their tasks affected by LLMs is now assessed as supported, which firms up one side of the bracket, while the IMF's 60 percent advanced-economies estimate awaits its own assessment. Because all such estimates share the task-overlap methodology, the convergence remains weaker evidence than it appears if that shared assumption fails.
Provenance
Where this claim has been said, linked to its canonical form.
roughly two-thirds of U.S. occupations are exposed to some degree of automation by AI
Analyzing databases detailing the task content of over 900 occupations, our economists estimate that roughly two-thirds of U.S. occupations are exposed to some degree of automation by AI.
We find that roughly two-thirds of US occupations are exposed to some degree of automation by AI
The primary Goldman Sachs research report deriving the estimate from occupation-level task data (O*NET for the U.S., ESCO for Europe), reporting the occupation-level distribution of the share of tasks AI can perform.
Assessment history
0 status changes over 2 assessments. full history →
Cite this claim: a formal citation with its evidence attached
Contribute
Every judgment on this page is open to challenge. A contribution is evaluated on its merits by the reviewer; if it succeeds the page changes, and if it does not, the reasons are stated. Either way the exchange becomes part of the claim’s public record.
The attention this claim received was paid for by a funded mandate. Funding buys only scheduling: it can make an assessment happen sooner, or reach deeper into a subtree. It has no influence on what the assessment concludes, and none on which claims enter the graph; assessments run under the same public standards whoever pays, funders never see or shape a verdict before anyone else, and mandates that attempt to steer conclusions are refused.
Created by extractor · Aug 10, 2026. Every judgment on this page is accompanied by a reasoning trace.