What is official and what is our judgment?
Task wording, importance ratings and skills come from O*NET. Pay, employment, openings and typical entry categories come from BLS. We add the task categories, explanations, hypothetical workflows, skill exercises and career comparisons. These are editorial inferences; they have not been independently validated against measured workplace adoption or employment outcomes.
Selecting the task sample
For each of 36 occupations, we use the corresponding O*NET-SOC code ending in .00 in the O*NET 31.0 Database . We select tasks marked “Core,” order them by the task Importance (IM) rating descending, break ties by numeric task ID ascending, and keep up to eight. If there are fewer than eight, we keep all core tasks. Each profile shows the selected count and total core count.
Importance is a 1–5 rating, not a measure of time. Selection favors highly rated tasks but omits lower-ranked duties, non-core tasks and work specific to an employer. A task can combine several activities. Occupations with differently broad task statements are therefore not perfectly comparable. We do not infer task frequency or work hours from importance.
Categories and calculation
- Routine automation (100): a bounded, repeatable output can be handled largely by configured software under stated conditions. This includes established workflow automation, not just generative AI. Exceptions and accountability remain.
- AI-assisted (50): documented tools can support a meaningful part of the work, while direction, verification or other substantive work remains with a person.
- Human-led (0): direct interaction, physical action or accountable judgment is central to the complete task. Supporting software may still be useful. This is not a permanent “human-only” claim.
Two routine tasks, five assisted tasks and one human-led task produce 56/100: (200 + 250) ÷ 8 = 56.25, rounded. We do not apply statistical calibration or label score ranges as measured high or low risk.
How much could a judgment change the result?
Moving one task by one category step changes an eight-task mean by 6.25 points before rounding; with six tasks, it changes the mean by about 8.33 points. Each profile shows the rounded mean minus and plus 50 divided by its sample size, bounded to 0–100. This sensitivity example is not a confidence interval. Other reasonable task selections, weights or judgments could change the result by more.
A small index gap does not establish a meaningful advantage. Even a larger difference cannot predict job loss: costs, adoption, regulation, demand, new tasks and organizational decisions intervene between a capability and an employment outcome. The ILO's occupational exposure research provides broader context using a different method; it does not validate this site's index.
Capability evidence and source age
Task rows link either to a relevant product capability reference or to the occupational description supporting the human or physical responsibility. A vendor page establishes what the vendor documents, not that the whole O*NET task is automated, that an employer uses the product or that the result is independently reliable. The inference connecting that capability to the task is stated separately in our explanation.
O*NET 31.0 was released in August 2026. Individual task and rating records can be older; both dates are shown in each matrix. A database release date is not a new survey date for every occupation. Source data were retrieved and checked on September 29, 2026. The publication process uses AI assistance and does not claim an independent occupational expert review; see the editorial policy.
BLS economic context
BLS table 1.2, 2025–35 projections supplies 2025 median annual pay and employment, projected percentage change, average annual openings, typical entry education, related experience and on-the-job training. Source employment and openings are in thousands; we multiply by 1,000 for displayed job counts. The projections were released August 27, 2026 .
Openings include replacement needs as well as growth; they are not today's vacancies. A national pay median is not an entry salary or an individual forecast. Wage estimates exclude self-employed earnings. Education and experience categories describe typical entry, not every credential or local requirement. BLS projections cover broader economic and occupational demand; they do not isolate this site's assessment of AI.
Related careers and skill overlap
Career links identify plausible work to investigate and a concrete exercise or gap. We do not require the destination to have a lower index: that would turn an uncertain number into an unsupported recommendation. Larger retraining steps are described as such, and entry categories appear beside each suggestion.
For each occupation, we combine O*NET Essential and Transferable Skills, select IM ratings, sort by importance descending and break ties by element ID ascending. The top twelve form the displayed overlap calculation. A shared skill is a matching element ID in both sets. Many are broad skills such as reading comprehension and active listening; overlap measures neither skill level nor hiring readiness.
Earnings scenarios
Stay = current annual pay × years. Move = preparation annual pay × (preparation years − unpaid months ÷ 12) + new annual pay × (years − preparation years) − entered one-time costs. Unpaid months occur at the start of preparation, with costs charged at the start. Preparation must fit within the selected period, and unpaid months must fit within preparation. Invalid inputs produce an error instead of being silently adjusted.
The calculator checks cumulative earnings each month and reports a recovery point only if the move remains at least level through the selected endpoint. Pay is fixed in nominal dollars. Taxes, benefits, raises, inflation, investment returns, hiring uncertainty and later career events are excluded. Both annual pay defaults are equal; the visitor must enter a plausible alternative. Inputs are not stored or transmitted by the calculator.
Reproduce and challenge the analysis
Download all profiles and references as JSON or download the task assessments as CSV . Count the labels and apply the formula. The JSON includes the selected skills, sample denominator, entry data and comparison rationale. Source snapshots and a deterministic data-build script are retained with the site's project so a future update can be checked against the previous version.
Report a disputed fact or category through contact, including the profile, task ID, source and what should change. A capability reference cannot settle every borderline category; a better explanation of a task boundary is useful evidence too.
Attribution
O*NET® data are from the O*NET 31.0 Database, U.S. Department of Labor, Employment and Training Administration (USDOL/ETA), under CC BY 4.0. O*NET is a USDOL/ETA trademark. We selected and adapted task and skill data and added classifications and commentary. USDOL/ETA has not approved, endorsed or tested these changes. Economic statistics are attributed to BLS; the task index is not a BLS measure.