DfE Guidance on AI in Schools: A Practical Checklist

DfE Guidance on AI in Schools: A Practical Checklist: practical strategies for teachers

Updated on

June 19, 2026

DfE Guidance on AI in Schools: A Practical Checklist

DfE guidance on AI in schools explained for teachers and leaders: policy, safeguarding, data protection, classroom use and workload checks.

Build your next lesson free Explore the toolkit

Copy citation

In this article

DfE guidance on AI in schools is clear on the core principle, and Ofsted's 2025 early-adopter research shows why schools need visible checks: staff may use generative AI, but professional judgement, safeguarding, data protection and curriculum quality stay with the school. AI can help teachers plan, draft, adapt and review materials. It should not make unsupervised decisions about learners, replace teacher assessment, or handle personal data without a lawful and secure process.

This guide turns the latest Department for Education materials into a working checklist for teachers, senior leaders, SENCOs and trust teams. It links the policy questions to classroom practice, so AI use becomes careful, useful and visible rather than hidden, inconsistent or risky.

‍

Key Takeaways

DfE expects schools and colleges to treat AI as a managed education tool, not as a shortcut around teacher responsibility.
Staff should check AI output for accuracy, bias, age suitability and curriculum fit before learners see it.
Personal data, learner work, safeguarding notes and special category data should not be pasted into public AI tools without an approved data protection route.
Homework, assessment, marking and feedback policies need updating because learners and staff can now produce AI-generated content quickly.
The safest first use cases are low-risk teacher workload tasks, such as adapting explanations, generating quiz questions, comparing examples and producing first drafts of non-sensitive resources.
Leaders should train staff through the DfE support modules, record local decisions and review impact through a small number of classroom pilots.

‍

What The DfE Says

The Department for Education's guidance on generative AI in education applies to schools and colleges in England. It recognises that AI tools can save time and widen access to resources, but it also points to risks around accuracy, intellectual property, cyber security, data protection and assessment integrity.

The guidance should not be read as a ban. It is a governance framework. Teachers can use AI to produce a first draft of a worksheet, suggest retrieval questions, rewrite an explanation for a younger reading age, or create alternatives for learners who need more scaffolding. The teacher still decides what is accurate, fair and suitable.

For leaders, the practical message is simple: write down where AI is allowed, where it is restricted, who checks the output and what staff must never upload. This connects directly to your AI policy for schools and the wider AI in education guide.

‍

Teacher and learners use an AI-supported tool under clear teacher guidance in a UK classroom. — Human-Led AI in the Classroom in practice: the teacher keeps the tool in service of the lesson rather than letting it dominate.

‍

Use The DfE Modules

DfE has also published support materials for using AI in education settings. These include staff modules on understanding AI, interacting with generative AI, safe use and practical use cases. The modules were updated on 19 May 2026, so they are a better staff training base than a slide deck copied from a vendor webinar.

A school can use the modules in three short staff sessions.

First, give all staff a common language. Explain what generative AI does, why it can produce convincing errors and why prompts matter. This reduces fear without pretending that the tool is reliable on its own.

Second, move to controlled examples. Ask staff to compare an AI-generated explanation with the curriculum source they already trust. Teachers can mark inaccuracies, missing steps and vocabulary that would confuse learners. This makes evaluation a normal part of AI use.

Third, agree local red lines. Staff should know which tools are approved, what data can be entered, who to ask before using learner information and how AI use is recorded. This is where AI literacy for teachers becomes a practical safeguarding and quality issue.

‍

Set Clear Guardrails

Every school needs a short set of AI guardrails that staff can remember. A long policy is useful for governance, but daily use depends on visible rules.

Use this five-part check before any AI task.

Purpose: What teaching or workload problem are we solving?
Data: Does the prompt include personal, sensitive or identifiable information?
Accuracy: What source will the teacher use to check the answer?
Bias: Could the output disadvantage a learner group, SEND learner or EAL learner?
Ownership: Who is responsible for the final resource, judgement or feedback?

The data question matters most. The Information Commissioner's Office explains that AI and data protection still sits inside UK data protection law when personal data is used. If a teacher pastes a learner's name, behaviour record, SEND profile or assessment notes into an unapproved public tool, the risk is not theoretical. The school may lose control of information it is required to protect.

Cyber security also belongs in the same conversation. The National Cyber Security Centre's guidance on AI and cyber security explains why large language models can create new security risks as well as useful outputs. School leaders should involve the data protection officer, IT lead and safeguarding lead before approving tools at scale.

‍

Classroom Use Cases

The safest classroom uses start with teacher preparation, not learner-facing automation. A Year 5 teacher can ask an AI tool to produce three alternative explanations of equivalent fractions, then compare them with the school's calculation policy. A science teacher can generate ten retrieval questions on particle models, then check each question against the scheme of work. An English teacher can ask for model paragraphs at three levels, then remove examples that overclaim or use vocabulary not yet taught.

These uses are close to normal teacher work. They can save time without handing over curriculum decisions. They also give staff a chance to practise prompt writing and output checking before moving into higher-risk uses.

For learner use, start with structured, transparent tasks. Learners can critique an AI answer, improve a poor explanation, compare two examples or spot unsupported claims. The learning goal is not to get the tool to do the work. It is to build judgement. This connects to teaching with an AI co-pilot, where AI is positioned as a draft partner, not as the source of truth.

SEND use needs extra care. AI can help teachers create accessible wording, visual prompts or alternative examples, but it should not diagnose need, write a support plan from scratch, or replace professional review. For more detail, see AI in special education and AI SEND administration.

‍

Assessment And Homework

DfE guidance asks schools to review homework and unsupervised study policies because AI changes what learners can produce outside class. The answer is not to abandon homework or treat every polished paragraph as cheating. The answer is to design tasks where the evidence of learning is visible.

For example, a history homework task can ask learners to submit a source annotation, a short plan and a final paragraph. A maths homework task can ask for the worked method, not just the answer. An English task can ask learners to highlight where they changed a draft after feedback. These steps make it easier to see thinking and harder for AI output to hide weak understanding.

Assessment policy should also separate low-stakes practice from formal judgement. AI-generated quiz questions can help revision when checked by the teacher. AI-generated grades should not be used as assessment evidence unless the school has a validated process and a human review. For marking and feedback, see AI marking and feedback and AI academic integrity.

‍

Leadership Checklist

Leaders can make AI safer by treating it as implementation work rather than a one-off announcement.

Start with a short approved-tools list. Name the tools staff can use, the account type they should use and the information they must not enter. Include a route for staff to request a new tool.

Update the acceptable use policy for staff and learners. Include AI-generated content, homework, image generation, plagiarism checks, parent communication and record keeping.

Create a DPIA screening route. Not every AI task requires a full data protection impact assessment, but new systems that process personal data, profile learners or influence significant decisions need proper review. When in doubt, screen the use case before purchase.

Run two or three pilots. Choose low-risk, high-need use cases such as lesson resource adaptation, retrieval quiz generation or admin drafting. Capture time saved, quality issues, staff confidence and any learner impact.

Train staff with examples from their own subjects. Generic AI training is quickly forgotten. A geography department needs different checks from a reception team, a SENCO or a safeguarding lead.

Review after one term. Ask what saved time, what created extra checking work, what learners noticed and whether any policy gaps appeared.

‍

Evidence And Limits

Ofsted's report on AI in schools and further education describes how early adopters are already using AI, but it also cautions that reliable evidence on impact is still limited. Many uses are recent, short term and dependent on local implementation.

The Education Endowment Foundation's ChatGPT in lesson preparation trial is useful because it looks at a concrete workload use case rather than broad claims about AI changing education. Schools should take the same approach: test a defined task, compare the result with normal practice and check quality before scaling.

The risk is not that teachers try AI. The risk is that AI use becomes informal and uneven. One teacher may use it carefully to adapt a text. Another may paste learner information into a public chatbot. A third may use AI feedback without checking whether it matches the success criteria. DfE guidance gives leaders a route to bring those behaviours into the open.

‍

Staff Meeting Script

Use this ten-minute script to start a department conversation.

Show one AI-generated resource linked to next week's lesson.
Ask staff to mark it for accuracy, reading level, curriculum fit and bias.
Ask what prompt would improve it.
Ask what information must not be entered into the tool.
Agree one safe use case for the next fortnight.
Agree how staff will record problems and examples.

This keeps the focus on teacher judgement. It also gives leaders concrete evidence for future policy decisions.

‍

Related AI in education guides: AI classroom implementation, human-led AI classroom practice.

‍

Research Evidence Check

Evidence Synthesis

Does current evidence support cautious, teacher-supervised use of generative AI in schools?

Mixed evidence: The Consensus search returns a mixed or indirect evidence base, so claims should be framed around the underlying teaching principle rather than the branded programme alone.

50% Yes from 8 studiesstrong evidence

Yes50%
Possibly50%
Mixed0%
No0%

Teacher takeaway

Use the approach as a structured support, not a guarantee: identify the target skill, teach it explicitly, and monitor whether it transfers into classroom language, reading or writing.

View the evidence behind this answer8 studies

1Ethical and regulatory challenges of Generative AI in education: a systematic reviewI. García-López et al. (2025) · Frontiers in Education

systematic reviewpossibly202572 citations

Generative Artificial Intelligence (GenAI) is changing education by enabling personalised learning and more efficient teaching practices.