How it works
AI changes tasks inside a job long before it changes the job title, so a useful measure has to start from the tasks.
How JobRoute works out what AI changes in each of your roles.
We start from the tasks inside each job, check each task against evidence of how AI is used today, and turn the result into actions you can take to your board. Every number comes from public data you can check.
Read the technical methodologySteps 1 to 3 on a real role
Paralegals and Legal Assistants
66 task exposure out of 100. AI already handles 9 of 12 tasks.
- Prepare affidavits or other documents, such as legal correspondence, and organize and maintain... 20 Estimated Efficient
- Prepare, edit, or review legal documents, including legislation, briefs, pleadings, appeals... 92 Observed Efficient
- Investigate facts and law of cases and search pertinent sources, such as public records and... 75 Estimated Efficient
- Prepare for trial by performing tasks such as organizing exhibits. 75 Estimated Standard
- Meet with clients and other professionals to discuss details of cases. 75 Estimated Standard
- Gather and analyze research data, such as statutes, decisions, and legal articles, codes, and... 65 Observed Efficient
The method in six steps
-
Start with the tasks, not the job title
AI takes over tasks inside jobs, and only rarely a whole job, so we break every role into its tasks first. Each of your job titles is matched to one of the 1,016 occupations in O*NET, the U.S. Department of Labor's database of occupations and the tasks they involve. A paralegal, for example, drafts documents, researches statutes and case law, summarizes records and meets clients. If a title is hard to match, we flag it and confirm the match with you.
-
Check each task against how AI is used today
For each task, we ask how much of this work AI already does. The evidence comes from the Anthropic Economic Index (June 2026 release), which maps real, privacy-preserving AI usage to O*NET tasks. Where the index has no direct observation for a task, a documented rule based on the task's wording and the kind of skill it uses fills the gap, and we record which method was used. Today the index observes 1,147 of the 18,796 tasks directly. The result is the task's task exposure, from 0 to 100. Higher means more of that task is AI-doable today.
-
Weigh the tasks by how central they are
O*NET rates how important each task is to the occupation, so a role's exposure leans toward the work people spend their days on rather than the rare task at the edge.
-
Add three signals about the future of the role
Exposure alone does not tell you what to do. The score also looks at skills that hold their value (the role's most important skills, rated for how well they hold up as AI spreads, using the World Economic Forum Future of Jobs Report 2025), at demand for the role (the Bureau of Labor Statistics ten-year employment projection), and at nearby roles people could move into with limited retraining. These combine with task exposure into the AI Ready Score, from 0 to 100. A higher AI Ready Score means a more AI-ready role or workforce.
-
Turn scores into decisions
For a company, the scores become cohorts of people with the same exposure pattern, each with a recommended action to redeploy, retrain or retain. The assessment adds the tasks AI already does in each role with tools built for that kind of task, the model tier each role needs so AI spend matches the work, where your job descriptions differ from what the role does nationally, and a board pack that cites the source and release behind every number.
-
Same inputs, same answer
The formula is published and every source is locked to a specific release, so anyone can reproduce a score from the same inputs. When a source publishes a new release, we re-run the scores and record the new version. This site runs on data version 2026-11-02 and method v1.0.0.
What the score is not
The score describes roles, not people. It is not a performance rating and not a prediction about any one person's job. The decision about what to do always stays with you, and the board pack records the evidence behind it.
The methodology page covers the formula, the weights and the known limits.
The sources, in one place
| Source | What it tells us | Release used |
|---|---|---|
| O*NET (U.S. Department of Labor) | Occupations, their tasks, how important each task is, and the skills they need | 30.3 |
| Anthropic Economic Index | How much AI is used on each task today | June 2026 release |
| Bureau of Labor Statistics, Occupational Employment and Wage Statistics | How many people work in each occupation, and what they earn | May 2025 |
| Bureau of Labor Statistics, Employment Projections | Expected ten-year change in employment | 2024 to 2034 |
| World Economic Forum, Future of Jobs Report | Which skills employers expect to grow or decline, and industry outlook | 2025 |
The task scores, wages and state results behind this site are open: download the data, browse every occupation or every state.
Questions about the method
How is AI exposure measured?
Task by task. Each role is broken into its O*NET tasks, and each task is scored for how much of it AI already does, using observed usage from the Anthropic Economic Index (June 2026 release) or a documented rule where the index has no observation. The role's exposure is the average across its tasks, weighted by how important each task is.
What does a higher AI Ready Score mean?
A higher AI Ready Score means a more AI-ready role or workforce. Task exposure runs the other way: higher task exposure means more of the work is AI-doable today.
Where does the data come from?
Public sources: O*NET 30.3 from the U.S. Department of Labor, Bureau of Labor Statistics employment, wage and projection data, the Anthropic Economic Index, and the World Economic Forum Future of Jobs Report 2025.
Do you use our employees' personal data to score them?
No. A person's score comes from the public task data for their role. The roster gives us job title, department and location so we can map each person to a role. No names, salaries or performance data are used.
Is AI used to calculate the score?
No language model is part of the scoring. The score comes from a published formula over public data, so the same inputs always give the same result.
How do you know which AI tools fit a task?
In the paid assessment, tools are linked to the kind of work a task involves, such as drafting documents or analyzing data. They are tools built for that kind of task, not a record of what your company uses.
How often are the scores updated?
When a source publishes a new release, we re-run the scores and record the new data version. Companies on the optional annual subscription get their scores refreshed with each release.
Check the method on roles you know.
Send us up to 10 job titles. Within two business days you get a PDF in board-pack format and a 30-minute walkthrough with a founder.
A founder replies within one business day.