How we measure
Every one of the 1,840careers on this site carries a number: the share of that job's day-to-day tasks AI is already used for. This page is the full account of where that number comes from, what keeps it honest, and how it relates to the official UK index.
What goes into the number
The score is a blend, not a single source. Published AI-usage research provides the observed foundation: Anthropic's 2026 labour market research measures what AI is actually used for by occupation in real workplaces, and OpenAI's "The AI Jobs Transition Framework" (Richmond 2026, CC BY 4.0) underpins the entry-difficulty grades and the 5, 10 and 20 year horizons. On top of that sit our own UK reviews, because the UK labour market is not the US labour market: a review of structural protections (statutory licensing, reserved legal activities and other legal protections that shape who may do what work here), a per-career assessment of robotics evolution (whether AI alone, or AI with robotics, could take on the work, and when), and relevant scientific developments.
The result is a measurement of now, not a prediction. When the landscape moves, the measurement moves with it: the score is maintained to constantly reflect the developing picture across AI, technology, robotics and scientific research, and every revision is recorded permanently in the open change ledger, career by career. Nothing is silently re-stated. The ledger currently holds 4 dated entries, including corrections to our own earlier statements, because a measurement you cannot audit is just an opinion.
How this relates to the official UK index
The Department for Education published an AI Occupational Exposure index in November 2023, and Skills England's 2026 skills assessments build on the same methodology. It is serious work and we encourage anyone comparing sources to read it. It answers a different question from ours, and both answers are useful.
The official index measures ability overlap: it maps occupations to the abilities they rely on, and scores how related those abilities are to what AI systems have demonstrated. It produces a relative ranking of occupations against each other (a standardised score, so a negative value still means some exposure, and there are no official Low or High bands). Our score measures observed task usage: the share of a job's actual day-to-day tasks that AI is already used for in real workplaces today, adjusted for the UK's own structural protections.
Across a matched sample of occupations the two approaches order the labour market in strong agreement: the trades, hands-on and physical work sit at the low end of both, and software, finance and marketing work sit high on both. Where they differ most is the relational, licensed and caring professions: an abilities lens sees high overlap with what AI can demonstrate, while a usage-and-protection lens observes that the day-to-day work of, say, a psychologist or a social worker still runs through a person, and that UK licensing and accountability structures shape how that changes. The official report itself is careful on this point: it measures exposure rather than distinguishing augmentation from substitution. Our moat axis is one attempt to add exactly that distinction, and we publish it as our own editorial classification with an honest "not yet classified" state where we have made no judgment.
In short: the official index is a valuable relative map of ability exposure. Ours is a living measurement of present-day usage, kept current as AI, robotics and science move, with every change on the record. Read together, they are more useful than either alone.
Read further
Questions about the method, or spotted something we should look at again? The ledger exists because we expect to be checked. Contact us via the schools page and we will answer.