NeedsAHuman scores about 1,000 US jobs on what AI can do, how well it does it and when it could replace them, from open data, with the evidence grade and range on every number. You are welcome to quote, chart and republish our figures with credit.
Headline figures, 2026-Q4
- 325 of 923 scored jobs mostly need a person: they score 80 or more out of 100 on Still needs a human, where higher is safer (our top band, Nah.).
- 5% of US workers are in jobs where AI can handle more than half of the task time.
- The median job’s replacement year is 2045. 80% of scored jobs between 2038 and 2057.
- 0 of 923 jobs have direct, head-to-head evidence on AI quality (grade A).
- Biggest movers since last quarter: listed in the data changelog.
Full tables are on the rankings page, and every number is in the open dataset.
Story angles
- Which jobs moved this quarter, and why. Each release lists the jobs whose scores changed most, with the reason (new usage data, new benchmark, new evidence).
- The 2013 forecast vs today. The most-cited “robots will take your job” figures come from a 2013 study. Each job page puts that old probability next to today’s measured coverage.
- Tasks, not jobs. Most jobs are a mix of tasks AI does, tasks AI helps with, and tasks that still need a person. Every job page shows the split.
- Where the evidence is missing. Most jobs have never had a direct test of AI against people. We publish which ones.
How to cite us
In text: “according to NeedsAHuman.com, an open dataset of AI job scores”.
Link: https://needsahuman.com/data/ for the dataset, or the job page for one job.
Formal: see How to cite on the data page.
The data is CC BY 4.0. You do not need to ask permission to use it. Please credit and link.
Charts and embeds
- Charts. Ready-made charts of the latest release are on the way. Until then, every figure is in the open dataset.
- Job badges. Every job page has an embeddable “Still needs a human” badge that stays up to date.
- Custom cuts. Need a table for your region, sector or set of jobs? Ask. We will turn it around as fast as we can.
Method in one paragraph
We score each job’s O*NET task list on three kinds of evidence: observed AI use at work (Anthropic Economic Index, Microsoft Research), task-by-task capability ratings, and benchmarks against real work. We weigh tasks by how much time they take. Quality is judged against qualified professionals, with an A–D grade for the strength of the evidence. The replacement timeline comes from 10,000 simulated futures and is always shown with its 80% range. Full methodology.
What we won’t say
We never say a job “will be eliminated”, and we never give a year or a score without its range or grade. Please keep the range or grade with the number when you quote us.
Who we are
NeedsAHuman is built and run by First Ideas, a small web design and development business in Wiltshire, England. No commercial relationship changes a score. More about us.
Contact
Email Gordon Hill at support@firstideas.co.uk. For deadlines, put “Deadline” and the time in the subject line.