The data and research behind your score
- Your AI Job Risk Score isn't our opinion.
- It's built on U.S. government occupational data and peer-reviewed research.
- Three sources do the work.
Occupational data — O*NET 30.2 (U.S. Department of Labor), used under the CC BY 4.0 license.
AI task-exposure framework — Eloundou et al.
- Each task's exposure to current AI comes from the peer-reviewed framework and ratings in "GPTs are GPTs."
- Read it free (preprint): Read the preprint on arXiv
- Published version (Science, paywalled): Published version in Science
- Underlying exposure data: Underlying exposure data on GitHub
- Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2024). GPTs are GPTs: Labor market impact potential of LLMs. Science, 384(6702), 1306–1308.
AI capability trajectory — METR (2025).
Matching your job to a real occupation
- We start by matching your job to a real occupation, because AI doesn't replace job titles — it affects the specific tasks inside a job, and standardized occupations are how those tasks are defined.
- When you type your title, we match it against more than 57,000 real-world job titles that map to 1,016 standardized occupations, so a confident pick takes you straight into the assessment.
- If your title doesn't resolve cleanly, you can describe what you do in a sentence or two; a database search narrows the field to the closest candidates, a language model ranks the best three, and you choose the one that fits.
- Throughout, you only ever see the plain occupation name and how confident the match is — the underlying federal occupation codes stay behind the scenes.
Capturing how you spend your time
- Once your occupation is set, you rate each of its real tasks on a five-point scale, from "None" to "Nearly all my time."
- Those ratings become weights that capture where your week actually goes — not rigid percentages, but the relative emphasis across everything you do.
- This is why two people with the same title can land on very different scores: the number reflects your mix of work, not a generic average for the title.
Calculating your score
- Every task carries an exposure value drawn from the Eloundou framework: minimal (today's AI can't meaningfully do it), partial (AI can do it with the right tools), or full (AI can already do it on its own).
- To build your score, we weight each task's exposure by how much of your time it takes, average those together, and place the result on a 0-to-100 scale — so the tasks that fill the most of your week move your number the most.
- The score is capped at 95, because no job today is fully automatable with zero human input, and it then falls into a plain-language band: minimal (0–19), limited (20–39), moderate (40–59), significant (60–79), or severe (80–95).
- Here's the idea in miniature. Say half your week goes to a task AI can already do on its own, a quarter to a task AI can handle with the right tools, and the last quarter to a task AI can't touch. Weighting each task by its share of your time and averaging gives (0.5 × 1.0) + (0.25 × 0.5) + (0.25 × 0) = 0.625 — about 63 on the 0-to-100 scale, which lands in the "significant" band.
Score = ( Σ βiwi / Σ wi ) × 100, capped at 95
What each symbol means:
- Score — your AI Job Risk Score, on a 0-to-100 scale (capped at 95).
- Σ — "the sum of," added up across every task you rated.
- βi ("beta") — the AI-exposure value for task i, drawn from the Eloundou framework.
- wi ("weight") — how much of your time task i takes, from your five-point rating.
- i — a single task; the formula runs over all of your rated tasks.
In plain terms: multiply each task's AI exposure (βi) by the share of your time it takes (wi), add those products up, divide by your total time, and scale to 100. The tasks you spend the most time on move your number the most, and the result is capped at 95 because no job today runs with zero human input.
How βi is set — anchored to Eloundou's exposure rubric:
- Minimal — AI can't meaningfully do the task → βi = 0
- LLM-via-software — AI can do it with the right software/tools → βi = 0.5
- Direct-LLM — AI can already do it on its own → βi = 1.0
The tasks driving your score
- Alongside the number, we surface the handful of tasks doing the most to drive it — the ones that are both AI-doable today and take up the most of your time.
- These are the concrete "AI can already do this" examples behind your score, so it's never just an abstract figure.
- And if none of your top tasks turn out to be highly exposed, we tell you that plainly — for some roles, that's the honest and welcome result.
The three-year outlook
- We also give you a sense of where things are heading.
- Using METR's measured rate of AI improvement — the roughly seven-month doubling in how long a task AI can complete — we project each of your tasks forward and translate the result into a direction of travel: rising, steady, or already at the ceiling.
- We deliberately don't put a single percentage on three years from now.
- Pinning down one future figure would imply a precision no one honestly has; the useful and truthful read is the trajectory, not a false decimal.
What this is — and what it isn't
- This score measures task exposure — what current AI can do — rather than predicting whether you'll lose your job.
- It's directional by design, built on today's models and the public research behind them, and it will move as both the models and the research move.
- It also can't see the things that often matter most: your employer, your skill, your judgment, and your relationships at work.
- Treat it as an informed starting point for thinking about your career, not as career, financial, or legal advice.
- Producing it asks very little of you, too — no résumé and no work history, just your job and how you spend your time — and we collect only what's needed to generate your result; see our Privacy Policy for the details.