Explainer, updated November 5, 2026
On August 27, 2026, BLS began rating every occupation it projects for relative AI exposure. A category places an occupation against the others, without naming the tasks behind it or what to do about them.
The BLS AI exposure categories, explained.
BLS combined five published measures into four groups, Low, Moderate, High and Very high, for 831 occupations. JobRoute scores the tasks inside each role, so you can see what sits behind a category and act on it role by role.
Book a 15-minute callWhat the four categories mean
BLS describes the two ends of the scale. Moderate and High sit between them, and every category is relative: it places an occupation against the other 831, with no absolute level of exposure implied.
- Low
- BLS: "the requirements of an occupation do not match well with the current capabilities of AI models, and that LLMs have not been observed performing many of the occupation's tasks."
- Moderate and High
- The two middle groups. BLS publishes no separate wording for them; the clustering step places each occupation by its two median scores.
- Very high
- BLS: "compared to other occupations, a larger fraction of an occupation's tasks can be completed or assisted by AI technology, and that LLMs have been observed performing some of the occupation's tasks."
How BLS builds the categories
- Five sources
Three theoretical measures of what AI could do, and two built from observed AI use mapped to tasks or work activities.
- Rank within each source
Each occupation's score becomes a percentile rank from 0 to 1. Of 4,155 occupation and source pairs, BLS imputed 211, affecting 75 occupations.
- Two medians
The median rank across the three theoretical sources, and the median across the two observed sources.
- Four groups
A clustering algorithm groups the occupations into Low, Moderate, High and Very high on those two medians.
| Authors | Study | What it measures | Type |
|---|---|---|---|
| Felten, Raj and Seamans | Occupational, industry, and geographic exposure to artificial intelligence | Whether AI capabilities relate to the abilities an occupation needs | Theoretical |
| Eloundou, Manning, Mishkin and Rock | GPTs are GPTs: Labor market impact potential of LLMs | Whether a language model could cut the time a task takes | Theoretical |
| Eisfeldt, Schubert, Taska and Zhang | Generative AI and Firm Values | The same question, rated by GPT-3.5 Turbo | Theoretical |
| Massenkoff and McCrory (Anthropic) | Labor market impacts of AI: A new measure and early evidence | Observed Claude use, mapped to occupational tasks | Observed |
| Tomlinson, Jaffe, Wang, Counts and Suri (Microsoft) | Working with AI: Measuring the Applicability of Generative AI to Occupations | Observed Copilot use, mapped to work activities | Observed |
Source list and method from BLS, Artificial Intelligence (AI) exposure categories, Methodology and Calculation sections. BLS maps every source to its SOC-based occupation list before ranking.
What BLS says the categories do not measure
BLS lists these limits on the same page. They matter most when a category is quoted as if it were a forecast.
- A category is not a forecast of employment growth or decline, and High or Very high does not mean employment will fall.
- Low does not mean an occupation will be unaffected by AI or is safe from future technological change.
- It is not a wage forecast, a worker replacement estimate, a productivity forecast or a probability of AI adoption.
- It does not separate automation from augmentation.
- The theoretical sources reflect AI available no later than mid-2023, and most sources cover language models only, not image or video generation.
- All five sources treat an occupation as a bundle of independent tasks, abilities or work activities, with no account of bottlenecks between them.
How JobRoute's scores relate to the BLS categories
JobRoute scores each task in a role from 0 to 100 and weights the tasks by how important they are to the role. The BLS category ranks the whole occupation. Across the 749 occupations both cover, the two put occupations in much the same order, with a rank correlation of 0.74.
JobRoute task exposure by BLS category
Low202 occupations
204% of tasks at 60+
Moderate183 occupations
2910% of tasks at 60+
High178 occupations
4425% of tasks at 60+
Very high186 occupations
5440% of tasks at 60+
| Aspect | BLS category | JobRoute exposure |
|---|---|---|
| What it rates | A whole occupation, placed in one of four groups | Each O*NET task, from 0 to 100, then the role as the importance-weighted average |
| Scale | Relative: each occupation is ranked against the others | A fixed 0 to 100 scale, so a task at 60 or more counts as highly exposed in every role |
| Inputs | Five published measures: three theoretical, two built from observed Claude and Copilot use | O*NET 31.0 tasks and importance ratings, observed use by task from the Anthropic Economic Index (June 2026 release), and a documented estimate where a task has no observation |
| AI capability it reflects | Theoretical sources: AI available no later than mid-2023. Observed sources: 2024 and 2025 use | Observed use in the June 2026 Anthropic Economic Index release |
| What you can do with it | Compare an occupation with others, alongside the BLS ten-year projection | See which tasks in each role AI already does, the model tier the work needs, tools built for that kind of task and nearby roles |
Where the two disagree
3 occupations that BLS rates Low score 60 or more on JobRoute, and 22 occupations that BLS rates Very high score under 40 on JobRoute. Of those 25, 19 have one task or none observed in the Anthropic Economic Index, so their JobRoute scores rest mostly on the estimate from each task's O*NET work activity.
BLS also draws on observed Microsoft Copilot use, which JobRoute does not use, and ranks occupations against each other rather than on a fixed scale. When the two disagree on a role that matters to you, read the task list on its job page and judge the tasks yourself.
| Occupation | BLS | JobRoute | Tasks observed |
|---|---|---|---|
| Farm Labor Contractors | Low | 75 | 0 of 7 |
| Embalmers | Low | 71 | 0 of 26 |
| Postal Service Mail Sorters, Processors, and Processing Machine Operators | Low | 68 | 0 of 14 |
| Travel Agents | Very high | 31 | 0 of 8 |
| Computer Numerically Controlled Tool Programmers | Very high | 32 | 1 of 16 |
| Concierges | Very high | 32 | 0 of 18 |
| Dispatchers, Except Police, Fire, and Ambulance | Very high | 32 | 1 of 12 |
Largest gaps in each direction. JobRoute exposure is the role's importance-weighted task exposure, 0 to 100. Across all 1,016 occupations, 1,145 of 18,838 task scores are observed in the Anthropic Economic Index (June 2026 release).
How to use both
- Use the BLS category
- To see where an occupation sits against all the others, read next to the BLS ten-year projection for the same occupation.
- Use JobRoute task exposure
- To see which tasks in a role AI already does, how many score 60 or more, the model tier the work needs, and the nearby roles people could move into.
- For your own company
- Job titles map to O*NET occupations, so both readings apply to your roles. The assessment scores every role and person and turns the result into a board pack.
On every JobRoute job page
Each job page names the occupation's BLS category next to the BLS wage and projection figures, for 982 of the 1,016 O*NET occupations. Specialties inside a BLS occupation show that occupation's category. Military occupations, which BLS does not project, and a few occupations BLS groups differently show none.
Questions about the BLS categories
What are the BLS AI exposure categories?
A four-level rating of relative AI exposure, Low, Moderate, High or Very high, for each of the 831 detailed occupations the Bureau of Labor Statistics projects. BLS built it from five published exposure measures and released it with the 2025 to 2035 Employment Projections on August 27, 2026.
Does a High or Very high category mean jobs will be lost?
No. BLS states that an exposure category is not a forecast of employment growth or decline, a wage forecast, a worker replacement estimate or a productivity forecast. It also says a Low category does not mean an occupation is safe from future technological change.
How does BLS decide which category an occupation is in?
BLS ranks each occupation within each of the five sources on a scale of 0 to 1, takes the median rank across the three theoretical sources and the median across the two observed sources, and then uses a clustering algorithm to group occupations into four categories on those two numbers.
How does JobRoute's AI exposure score differ from the BLS category?
JobRoute scores every O*NET task in a role from 0 to 100 and weights the tasks by importance, so it shows which tasks drive the number and how many score 60 or more. The BLS category ranks whole occupations against each other. Across 749 occupations both cover, the two put occupations in much the same order (rank correlation 0.74). See the comparison on this page.
Why would the two disagree for an occupation?
The sources differ. BLS draws on five measures, including observed Microsoft Copilot use that JobRoute does not use. JobRoute scores each task from observed use in the Anthropic Economic Index where it exists and from a documented estimate where it does not. Of the 25 occupations where the two differ most, 19 have one task or none observed in the Anthropic Economic Index, so their JobRoute scores rest mostly on the estimate.
Where can I find the BLS category for a job?
Each JobRoute job page names the BLS category next to the BLS wage and projection figures, for 982 of the 1,016 O*NET occupations. Browse every occupation. BLS also publishes the full list as a spreadsheet with its 2025 to 2035 projections.
Sources
- Artificial Intelligence (AI) exposure categories. U.S. Bureau of Labor Statistics, Employment Projections, last modified August 27, 2026.
- AI exposure categories and 2025-35 employment projections (XLSX). U.S. Bureau of Labor Statistics, retrieved November 5, 2026.
- Employment Projections: 2025-2035 Summary, news release USDL-26-1422. U.S. Bureau of Labor Statistics, August 27, 2026.
- Artificial Intelligence (AI) impacts on employment projections, factsheet. U.S. Bureau of Labor Statistics.
- JobRoute task exposure: data version 2026-11-03, method v1.0.0, built on O*NET 31.0 and the Anthropic Economic Index (June 2026 release). Methodology, data downloads.
See the tasks behind the category, for your own roles.
Send us up to 10 job titles, with no employee data. Within two business days you get a PDF in board-pack format, with each role's tasks scored, and a 30-minute walkthrough with a founder.
A founder replies within one business day.