What relative AI exposure means

The word relative does important work. The Bureau of Labor Statistics compares detailed occupations with one another, then places them into four broad categories. A category tells you where an occupation sits within that comparison. It does not say that a fixed percentage of the job can be automated, or that every worker in the occupation has the same experience.

Exposure is best treated as a prompt for investigation: Where do current AI capabilities overlap with this occupation's tasks or work activities? Which parts of the workflow might be assisted or completed by a tool? What judgment, physical action, relationship, or accountability surrounds those parts? Those questions are narrower than "Will this job vanish?" and much more useful.

For the BLS 2025–2035 Employment Projections release, BLS combined five external measures. Three estimate theoretical exposure by comparing AI capabilities with occupational abilities or tasks. Two use observed interactions with AI systems mapped to O*NET tasks or work activities. BLS normalizes the measures as percentile ranks, combines theoretical and observed dimensions, and uses a clustering method to assign every covered detailed occupation to one of four categories.

The categories are not equal steps on an absolute scale. They depend on the occupations, source measures, crosswalks, normalization, imputation, and clustering choices in the BLS method. BLS also notes that much of the underlying capability evidence focuses on language models and reflects technology available during specific earlier periods. The measure is informative, but bounded.

BLS is explicit about these limits. A High or Very high category does not necessarily imply employment decline. A Low category does not mean an occupation will be unaffected. The classification also does not distinguish whether AI use automates an activity or augments a worker performing it. JobShift therefore never converts the category into an invented percentage or a "future-proof" score.

An occupation is a bundle of tasks, decisions, tools, relationships, and working conditions. AI may assist with drafting or search, automate part of a repeatable information step, leave a physical activity largely unchanged, and create new work in review or exception handling. These effects can coexist inside one job.

This is why JobShift pairs the BLS category with O*NET task statements, skills, work activities, and technology skills. The category supplies a comparative signal. The work profile helps you inspect what the occupation actually contains. Neither source observes the exact workflow at every employer, so local tools, policies, regulation, safety requirements, and customer or patient needs still matter.

These examples use the occupation identities and source data published in JobShift Atlas. They illustrate why the category should open a task-level inquiry rather than close it.

Use exposure as one column in a larger comparison. Projected employment change describes a national scenario; annual openings include growth and replacement needs; wages summarize a distribution rather than your likely offer; preparation fields describe typical entry routes; and O*NET describes the work and worker characteristics. Your preferences, location, access to education, and tolerance for particular working conditions are additional questions that public data cannot settle.

A sensible review asks: Do the tasks interest me? Is the preparation realistic? What do growth and openings show separately? What skills and technology are important? Which duties may be changing, and would I be willing to learn the tools and review practices around them? A universal score would hide those tradeoffs, so JobShift does not publish one.

Do not treat the label as an instruction to leave. Start by identifying the tasks that occupy your time and checking whether tools are actually changing them in your setting. Learn the relevant tool where that is useful, but pair tool fluency with occupation-specific knowledge and the ability to verify outputs, handle exceptions, communicate decisions, and take responsibility for quality.

• Inspect the occupation's tasks instead of reacting to the category alone.
• Ask how employers in your field use AI now, including policies and review requirements.
• Strengthen domain expertise that helps you recognize errors and important context.
• Practice verification, documentation, communication, and oversight around tool-assisted work.
• Identify transferable skills and compare related occupations if you are already interested in a change.
• Revisit dated evidence periodically as tools, usage, and projections change.