AI for teachers: what's actually useful

The short answer

The AI uses that have survived contact with real classrooms are the teacher-facing ones: drafting feedback on completed work for the teacher to review, generating planning materials and model texts, adapting resources, and drafting routine communication. That matches the DfE's own position — it sees “more immediate benefits and fewer risks from teacher-facing use of generative AI” than from tools put in front of pupils.

What AI is not for: making summative judgements about a child's attainment (the DfE's training materials rule this out explicitly), knowing your children, or teaching. The useful mental model is a competent assistant who drafts and never decides. Everything below follows from it.

Start with marking and feedback

If AI is going to give a teacher real time back, it has to attack the biggest workload item, and that is marking. UCL Institute of Education research across 9,405 teachers found marking is a key driver of work stress; one in four teachers in England report working more than 60 hours a week. A tool that drafts specific, criteria-linked feedback on each child's piece — for the teacher to edit rather than write from scratch — is aimed at exactly that load. The workload evidence in full is on why Howay exists.

Marking is also where the questions get sharpest: accuracy, data protection, moderation evidence, chatbot or purpose-built tool. All of it is covered in AI marking: the complete guide for teachers — the companion page to this one.

Know the rules before you start

Three sets of rules cover a teacher using AI in England, and none of them is long.

• The DfE's two conditions. Check every AI output for accuracy and appropriateness — responsibility stays with you — and never use AI for summative judgements about attainment. The documents, quoted and linked, are on the DfE guidance explainer.
• Data protection. The DfE recommends personal data is not put into generative AI tools; where it is strictly necessary, the product must comply with data protection legislation and school policy. Pupils also own the copyright to their own work. What that means in practice is on is AI marking safe?
• The STA independence rules at KS2. Using AI to inform your own assessment of a completed piece leaves that piece independent — but work a pupil edits or rewrites after seeing AI feedback is not independent evidence, and moderators must be told where AI feedback was used. Keep originals in the collection. The exact wording is on AI marking and independent writing.

If your school needs all of this written down as policy, there's a free AI policy template for primary schools with every clause linked to the DfE, ICO and STA guidance it comes from.

A chatbot, or something purpose-built?

For planning and drafting, a general chatbot is often enough — the stakes are low and you are the only person in the loop. For anything involving pupils' work, the calculation changes. The DfE's guidance notes that general-purpose LLMs are designed to vary their answers, so the same work can get different feedback on different days; you supply and re-supply your own criteria; and every paste of a child's writing is a data protection decision you are making personally.

Purpose-built tools exist to close those three gaps — consistency, curriculum criteria, compliant data handling. The direct comparison is on Howay vs ChatGPT for marking, and the honest survey of the market — including tools that aren't Howay — is on the best AI writing feedback tools for UK primary.

A sensible way to start

• Pick one class set of completed writing — work that is finished and would be marked anyway, not work in progress.
• Choose the criteria first. Feedback is only as useful as what it is measured against; use the curriculum objectives you are actually teaching.
• Review every word before a child sees it. This is the DfE condition, and it is also where you find out quickly whether the tool is any good.
• Tell your school what you're doing. If there's an AI policy, work within it; if there isn't, the template is a faster start than a blank page.
• At Year 6, protect the evidence trail — keep original independent versions for moderation and flag where AI feedback was used, per the STA rules.

Where Howay fits

Sources

1. DfE policy paper, “Generative artificial intelligence (AI) in education” (updated August 2025) — the DfE's formal position on teacher-facing AI use, oversight, and pupil personal data.
2. DfE support materials, “Use cases of generative AI in education” Module 4 — marking and feedback use cases, the summative-judgement rule, and the personal-data and copyright warnings.
3. Jerrim & Sims (2020), “Teacher workload and wellbeing”, UCL Institute of Education / Nuffield Foundation — marking as a key driver of work stress (9,405 teachers, TALIS 2018).
4. STA, “Key stage 2 teacher assessment guidance” (updated March 2026) — the independence rules for writing evidence, including writing edited or rewritten following AI feedback.

Common questions

How are teachers actually using AI in 2026?

The uses that have stuck are teacher-facing: drafting feedback on completed work for the teacher to review, generating planning materials and model texts, adapting resources for different needs, and drafting routine communication. The DfE says it sees more immediate benefits and fewer risks from teacher-facing use of generative AI than from putting tools in front of pupils — which matches how most schools have adopted it.

Is using AI allowed for teachers in England?

Yes. The DfE's policy paper on generative AI in education supports teachers using AI for tasks including feedback and planning, on two conditions: outputs must be checked for accuracy and appropriateness by the professional, and AI must not be used to make summative judgements or predictions about student attainment. Schools are expected to set their own policy on top — many are still writing one.

Will AI replace teachers?

No — and the design of the official guidance assumes it won't. Every DfE document on the subject places final responsibility with the teacher: AI can draft, suggest, and inform, but the professional judgement about a child's learning stays human. The realistic framing is that AI replaces some of teachers' Sunday-evening tasks, not teachers. The judgement calls — what this child needs next, whether this piece shows real progress — remain the teacher's job.

What should I check before putting pupils' work into an AI tool?

Four things. Whether the tool processes pupil data lawfully under UK GDPR and your school's privacy policies — the DfE recommends personal data is not put into generative AI tools unless the product is compliant and the school has approved it. Whether pupils' work is used to train the model. Whether you can review and edit everything before a child sees it. And, at Year 6, what the STA independence rules mean for moderation evidence: work a pupil redrafts after seeing AI feedback is not independent, so keep originals and flag AI-feedback use.

Does AI actually reduce teacher workload?

It can, on the tasks that eat evenings — and marking is the biggest of them. UCL Institute of Education research across 9,405 teachers found marking is a key driver of work stress, and one in four teachers in England report working over 60 hours a week. AI that drafts feedback for the teacher to review attacks that load directly. The honest caveat from the DfE: formative assessment is integral to planning next steps, so reducing workload here needs care — which is why review-and-edit, not fire-and-forget, is the model that works.

Is AI for teachers different at primary?

Meaningfully, yes. Primary writing is handwritten in exercise books, so a tool that only accepts typed text is useless for everyday marking — reading handwriting from a photograph is the entry requirement. Feedback has to land at a child's reading level. And at the end of KS2, the statutory teacher assessment and moderation rules shape what AI use is safe. Most general AI advice for teachers is written for secondary essays and typed work; primary needs its own answers.