How we calculate salary, AI exposure & growth outlook

Our sources

We tier our sources by how directly they touch a number shown to you. Only the first tier sets a value attached to a specific occupation; the others provide context and a cross-check.

Tier 1 — the figures you see

• Salary — U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics (OEWS, May 2025). Every salary figure is keyed to a federal Standard Occupational Classification (SOC) code and, where relevant, adjusted for your region and for union wage patterns. Salary figures are never AI-generated.
• Career-growth outlook — U.S. Bureau of Labor Statistics, Employment Projections, 2024–34 (released 2025). The federal government’s authoritative 10-year occupational forecast. We use each occupation’s projected percent change in employment, keyed to its SOC code. Growth figures are never AI-generated.
• Potential task exposure — Tufts University Digital Planet, “AI and the Emerging Geography of American Job Risk” (March 2026). An independent academic estimate, by occupation, of how much of the work is theoretically exposable to AI. This is the lead source for our exposure band.
• Observed AI use — the Anthropic Economic Index (March 2026). A measure, by occupation, of how much AI is actually being used in a field today, based on real usage data. We disclose that Anthropic is an AI vendor and lead with the independent academic source above.
• Pattern corroboration — Microsoft Research, “Working with AI” (July 2025). An independent measure of where generative AI applies across occupations. Its authors note their scores are meaningful only for relative comparison, so we use it as a cross-check on the overall pattern (information work highest; hands-on and physical work lowest), not as a per-job number.

Tier 2 — context and framing

• Stanford — “Canaries in the Coal Mine” finds early signs of softening in some entry-level roles.
• The Yale Budget Lab finds broad labor-market stability so far.

These two careful analyses disagree about how much AI has changed employment to date. That disagreement is exactly why we treat exposure as a signal rather than a forecast, and why we show you the underlying lenses instead of a single confident number.

What we don’t do

We do not base any number shown to you on pre-2023 task-exposure predictions or on a single consulting-firm forecast. Where sources diverge, we show the divergence rather than averaging it away.

From sources to the rating you see

• Everything is keyed to SOC. Each career is matched to its BLS SOC code, which lets us join salary and exposure to the same occupation consistently.
• Two lenses, shown side by side. “Potential task exposure” (Tufts) and “AI use in this field today” (Anthropic) are each shown on a 0–100 scale. We show both rather than averaging them, because they measure different things and often diverge — a field can be theoretically exposable yet see little AI use in practice today (skilled trades and hands-on healthcare are common examples).
• The band. The Low / Moderate / High headline reflects potential task exposure, grouped into thirds across all occupations in the dataset. The two lenses beneath it give the fuller picture.
• Growth bands follow BLS’s own categories. We show each occupation’s BLS-projected employment change for 2024–34 and band it against the 3.1% all-occupation average that BLS uses as its benchmark: Declining (a projected decrease), Stable (roughly at or below average, 0–3%), Strong (above average, 4–9%), and Very Strong (10% or more, which BLS calls “much faster than average”).
• The cross-check. Microsoft’s independent ranking corroborates the overall pattern; it does not move an individual job’s number.

Signal, not destiny

AI exposure tells you where AI is most likely to change how work is done. It is not a prediction that a job will disappear, and it is not a measure of whether a career is a good choice. A field with high potential exposure but low current AI use is often a field where there is time to build the skills — judgment, relationships, hands-on problem-solving — that AI does not replicate well.

An equity note

The research consistently finds that measured AI exposure skews toward higher-paid, more-educated, and information-heavy roles. Higher exposure therefore does not mean lower pay, lower security, or lower value — in many cases the opposite. We show exposure as one input for planning, never as a verdict on a person or a path.

Limits & honesty

• Usage-based measures reflect one slice of overall AI activity and a particular moment in a fast-moving field.
• Employment projections are 10-year forecasts that assume a full-employment economy; BLS revises them annually (the 2025–35 set is due in August 2026). Treat them as direction and relative size, not precise guarantees.
• Reasonable researchers disagree about the pace and scale of AI’s labor-market effects; our framing reflects that uncertainty on purpose.
• We update these sources as newer research is published, and we revise the version and date at the top of this page when we do.
• We only describe data sources and methods that are actually live in the product. We do not claim grounding, certifications, or partnerships we do not have.

Questions

Questions about how a figure was produced, or a correction:

Grant Salmon, Founder
North Jersey Creative Services LLC
Email: grant@newpaths.ai