Quick answer: If staff in your school already use AI, the useful next step is usually not to encourage more use. Find out what is already happening, agree a short rule that staff can find and understand, and train people on real tasks, including when not to use AI. In a June 2026 Bett/YouGov survey of 1,033 UK teachers, 52% reported using AI weekly or more, 41% reported no training or help from their school, and 33% reported a staff-use policy (48% said none and 19% did not know). These are self-reports from teachers, not a census of schools.
Use is not the same as readiness. A teacher can use AI every week without knowing which account is approved, what data must stay out, or who is accountable for the result. This guide is for school leaders, business managers, IT leads and training leads who want to review existing use calmly, without starting from a ban or a rollout.
Treat the next half term as a review, not a launch. Ask what staff already do, check accounts and data boundaries, publish one page of local rules, then run one scenario-based training session. Measure the value of tasks, not how often AI is used.
What does the survey show, and what does it not show?
The figures below come from AI in Education 2026: Working it out, published by Bett. It reports an online YouGov survey of 1,033 UK teachers, weighted, with fieldwork from 16 to 29 June 2026. The survey was commissioned by Bett and supported by Lenovo. Denominators matter, so each row says who was asked.
| Self-reported finding (June 2026) | Figure | Who was asked |
|---|---|---|
| Use AI weekly or more | 52% (26% in 2025) | All teachers |
| Received no training or help from their school | 41% | All teachers |
| School has a staff-use policy / has none / do not know | 33% / 48% / 19% | All teachers |
| Say AI has cut their workload | 51% | Teachers who use AI |
| Say AI means they work fewer hours (55% say the same hours, 4% more) | 35% | Teachers who use AI |
| Use AI for lesson resources / for marking pupils’ work | 76% / 8% | Teachers who use AI (more than one answer allowed) |
| Agree that using AI for core teaching would feel like “cheating” | 45% | All teachers |
These are teachers’ own reports. The survey does not show that AI improves learning, that schools with training or policy get better results, or how small and medium-sized businesses behave. Some questions and answer options changed between yearly waves, so year-on-year comparisons should be read with care. Findings are dated June 2026 and may not describe your school.
Five checks for a school where AI is already in use
Work through these in order. Each check produces something a member of staff can use.
Find out what is already happening
Ask staff, without blame, which tasks they use AI for and which accounts they use. A short anonymous tally by task type is enough. Do not collect pupil information or ask staff to paste examples.
Check accounts and tools
Separate approved, managed school accounts from personal or free accounts that staff chose themselves. The school decides which tools are approved; a tool being popular does not approve it.
Set the data boundary
State what must never be entered, and when to ask the data protection officer (DPO) first. DfE data protection guidance for England says to seek advice from your DPO or IT lead, and to discuss personal data use with the DPO before using a tool that way.
Name who checks the output
Staff remain responsible for anything they use. DfE guidance says content generated by AI needs critical judgement to check it is appropriate and accurate. Agree what “checked” means for pupil-facing material.
Make the policy findable and testable
A policy that staff cannot find, or cannot apply to a real task, does not help. Bett reports that, even where a policy exists, only 18% of teachers say they understand it well. The report text does not state the base for that figure, so treat it as a prompt to test understanding in your own school rather than as a benchmark.
Staff may use [LOCAL APPROVED TOOLS] through their managed school account for [LOCAL APPROVED TASKS]. Staff must not enter [LOCAL PROHIBITED DATA] and must check every output before it is used. For anything involving personal data, safeguarding, SEND or assessment decisions, staff must ask [LOCAL CONTACT] first. Questions: [LOCAL CONTACT]. Reviewed: [DATE].
The square-bracket items are placeholders. Only your school can decide them. The free AI Acceptable Use Policy Template gives a longer structure to adapt. It is written for UK organisations in general, using business wording such as company and clients, so a school will need to adapt that wording; there is no school-specific version on this site yet. It does not make a policy compliant by itself.
What should staff training cover?
Bett reports that two in five teachers have had no training or help from their school. Training that only demonstrates prompts goes out of date quickly and says little about judgement. A short session built on realistic tasks works better as a starting point, in AHAI’s view.
Predict first
Staff write down what a good answer for a real task would contain before they use AI.
Compare and critique
Compare the AI output with the prediction. Look for errors, gaps, tone and anything that does not fit the class.
Decide and explain
Keep, edit or discard, and say why. Include the option of not using AI at all.
Apply the local rule
End each scenario by asking which part of the school’s own policy applies and who to ask.
The predict, compare and explain approach follows an opinion piece in the Bett report by a teacher development lead, which argues for practising with authentic classroom problems. It is one author’s view, not a survey finding and not evidence that this format works better than others.
Facilitator scenarios (fictional)
| Scenario | Suggested starting position for discussion |
|---|---|
| A teacher asks an approved tool for a worked example on fractions for a Year 6 class. | Reasonable first task if the school approves it. Check the maths, the method and whether the example fits the class before use. |
| A teacher wants to paste a pupil’s SEND notes into a tool to draft support ideas. | Stop. This is personal information about a pupil. Ask the DPO and safeguarding lead before any use, and do not rely on removing the name. |
| A teacher wants AI to give marks to a set of essays. | Not a routine use. Marks affect pupils, so the purpose, checking and accountability need a local decision first. The teacher stays responsible for the mark. |
| A teacher thinks a pupil’s homework was written by AI. | Suspicion is not proof. Do not treat a detector score or writing style as proof. Follow the school’s assessment and malpractice procedure, and the awarding organisation’s rules for qualification work. |
| A colleague prefers not to use AI for lesson planning. | A legitimate professional choice. Make sure they still know the policy and where the boundaries are. |
Not using AI is a valid decision
In the survey, 45% of teachers agreed that using AI for core teaching tasks would feel like cheating, and 49% disagreed. That feeling is not a training gap to be removed. Staff should be able to decline a task without being treated as behind, and leaders should say which tasks AI is not suitable for, such as decisions about individual pupils.
Task savings, workload and working hours are different things
Among teachers who use AI, 51% said it had cut their workload, 35% said they work fewer hours, and 55% said they work the same hours. The Bett report notes that regular users typically report saving around one to two hours a week. A task can be quicker without the working week getting shorter, because the time may go to other demands.
For one task, record the time to prepare, check and correct the AI draft, compared with doing it without AI. Ask what the time saved was used for. Do not present the result as a shorter working week, or as improved learning, unless you have evidence of that.
Assessment looks different in primary and secondary
Bett reports a large difference between phases. Among teachers who assess work, 76% of secondary and 16% of primary respondents said they had marked a piece in the past year that they suspected was AI-generated. These are suspicions, not confirmed cases, and the survey cannot say how often the suspicion was right for any one school.
Keep primary and secondary approaches separate. Secondary and post-16 settings should check current awarding organisation rules for coursework and controlled assessments. Whatever the phase, do not use an AI detector score as the sole basis for an accusation, and give pupils a fair chance to explain their work.
When to stop and ask
Scope and sources
This guide is for staff-facing use in schools. DfE guidance applies to England. Scotland, Wales and Northern Ireland have their own education policies, and the survey covers the UK as a whole. It is general information, not legal, data protection or safeguarding advice.
- Bett, AI in Education 2026: Working it out, YouGov online survey of 1,033 UK teachers, fieldwork 16 to 29 June 2026, supported by Lenovo. Report text read on 11 October 2026.
- Department for Education, Generative artificial intelligence (AI) in education, page showing an update of 12 August 2025 when read on 11 October 2026.
- Department for Education, Generative artificial intelligence (AI) and data protection in schools, page showing an update of 9 July 2026 when read on 11 October 2026.
Reviewed against these sources on 11 October 2026 by AHAI. It has not had independent review by a DPO, safeguarding lead or assessment specialist.
Start with the rule
Use the free AI Acceptable Use Policy Template as a starting structure for your own staff rule. It uses business wording that a school will need to replace, it has to be completed and approved locally, and it is not a certificate of compliance. For a first low-risk task, see What can teachers safely use Gemini for?
Get the Policy Template