JRIJob Risk Index

Methodology / 2026 edition

How the index is built.

Evidence before anxiety: every score in the index is produced by our own data pipeline — official research feeds plus a weekly intake of live job postings that we collect, classify and verify in-house. The derivation rules are documented on this page; the dataset behind them is built every week, not downloaded once.

01 / The score

A 0–100 automation risk score per profession.

Each profession carries a single risk score from 0 (resistant to automation) to 100 (most of the day-to-day output can already be automated). Scores are grounded in published evidence. The launch collection was editorially reviewed across its original 33 profiles. New data-qualified profiles use the versioned AEI automation-share v1 rule: the Anthropic Economic Index automation share is rounded to the nearest integer after demand and source gates pass. This measures observed task automation exposure, not the probability that a job disappears. The current catalog contains 84 professions.

02 / Risk bands

Four bands, derived — never stored.

The level shown next to a score is always derived from it with fixed thresholds: 0–25 Low, 26–50 Medium, 51–75 High, 76–100 Critical. Storing only the score means a level can never contradict the number it summarises.

03 / Timelines

When the change is expected to bite.

Every profession also carries an impact window — 0–2, 2–5, 5–10 or 10+ years — describing when automation is expected to change a meaningful share of its tasks, based on the same sources as the score. A high score with a long window means the direction is clear but the transition is gradual. Every week the pipeline pulls live job postings from employer career boards, classifies them against the profession set, and stores the raw records — the demand signals, salary figures and skill mentions on this site come from that weekly intake, not from a static file.

04 / Market signals

Search-interest trends, point in time.

Where a profession shows a vacancies or interest trend, it reflects point-in-time, research-captured search-volume growth for that role — not a live feed. Trend coverage is partial today and clearly dated wherever it appears. Live job-market evidence from aggregators is on the roadmap and will be documented here when it ships.

05 / Limitations

What this index is not.

The index measures task exposure to current AI capabilities, not the certainty of job losses: adoption speed, regulation and demand shifts all shape real outcomes. Scores are updated as new research is published, and profiles show their last update date. Treat the index as a structured starting point for career decisions, not a verdict.

06 / Editorial process

Who curates and reviews what gets published.

List rankings always come from the dataset through a versioned rule — never from a hand-written list. The editorial copy (intros, FAQs, entries) is drafted with AI assistance and then read and checked by a human editor before publication: G. Roman Aaron. If something does not hold up, it is changed; an audit that passes is recorded with its date on the page. Everything you see is data first, editorial review second.