Teacher AI Use Doubled This Year. Their Enthusiasm Just Fell
A Bett/YouGov survey of 1,033 UK teachers finds weekly AI use has doubled, but only a third report working fewer hours — and optimism about AI is falling for the first time.
The paradox in the numbers
Every term brings a fresh adoption statistic, and the direction of travel is always up. The latest, from a Bett-commissioned YouGov survey of 1,033 UK teachers released in August, is no exception: weekly AI use has doubled in a year, from 26 percent to 52 percent. Three-quarters of users now lean on it to build lesson resources such as worksheets and plans.
But buried in the same report, titled Working It Out, is a more uncomfortable number. Among teachers who use AI, only 35 percent say they are working fewer hours as a result. Fifty-five percent say their hours are unchanged, and 4 percent say they are working more. Just over half, 51 percent, report that AI eased their workload in some sense — a gap between "felt easier" and "actually shorter" that anyone who has ever cleared an inbox only to fill it again will recognise.
And for the first time since Bett began tracking teacher sentiment in 2024, enthusiasm has dropped. Only 47 percent of teachers expect to use AI more over the coming year, down sharply from 68 percent when the same question was asked in 2025.
Why time saved doesn't become time free
This is not a new phenomenon outside education. Economists have a name for it — the Jevons paradox, or more loosely, "induced demand": when a tool makes a task cheaper or faster, people don't necessarily do less of it. They do more of it, or they redirect the saved capacity elsewhere. A teacher who can draft a worksheet in four minutes instead of forty doesn't bank the extra 36 minutes as free time by default. That capacity gets absorbed — into differentiating three more versions of the resource, replying to more parent emails, or covering a colleague's marking.
Duncan Verry, Bett's portfolio director, put it plainly: teachers are using AI more than ever, but their enthusiasm for it is cooling. The tools are being adopted at pace, but the promised dividend — evenings and weekends returned — has not materialised for most.
Time saved on one task rarely becomes time off. It becomes time spent on the next task nobody scheduled for.
This tracks with what earlier NEU polling this year found: three-quarters of teachers now use AI for day-to-day work, concentrated overwhelmingly in resource creation, lesson planning and admin — the preparatory layer, not classroom teaching itself. Only 8 percent of teachers in the Bett survey use AI to mark or grade work, suggesting the caution around high-stakes assessment tasks that other recent research has documented is holding.
The trust problem underneath the workload problem
The cooling enthusiasm isn't purely about hours. The same survey found 57 percent of secondary teachers suspected at least one student of submitting AI-assisted work without permission in the previous four weeks, and 76 percent of those marking such work admitted they had graded pieces they suspected were AI-generated anyway — because proving it, or acting on suspicion alone, remains fraught.
On coursework specifically, teachers are split on the fix: only 4 percent want extended-writing coursework left unchanged, 30 percent want better detection software, and 29 percent would rather move that writing into supervised, in-class conditions. Just over half, 52 percent, back dedicating curriculum time to teaching pupils responsible AI use — a policy gap flagged repeatedly this year as schools write acceptable-use rules for staff while leaving pupils largely untaught on the subject.
Put together, the picture is of a profession using a tool constantly while trusting it, and the systems built around it, less than it did a year ago. Adoption and confidence are moving in opposite directions.
What to do
For school leaders, the workload lesson is specific: if you want AI to actually reduce hours, someone has to deliberately protect the time it frees up, rather than let it default-fill with the next task in the queue. That means naming what gets removed from a teacher's workload when AI takes on a task — not just adding AI as an extra tool on top of an unchanged list of duties.
For staff CPD, the marking findings matter more than the planning findings. With only 8 percent using AI to grade and most schools offering little guidance on how to handle suspected AI-assisted submissions, that is the area most likely to generate inconsistent, high-stakes decisions this term.
What to watch
Whether next year's Bett survey shows enthusiasm recovering or continuing to slide will say more about how schools implemented AI in 2026-27 than any vendor claim will. A second year of falling optimism, paired with unchanged hours, would be a clear signal that the current approach to rollout — more tools, same job description — isn't working.
Sources: