Most guides to AI tools for teachers are lists of apps. A list will not help you decide what to use for what, or what to leave alone. This guide goes in order of how much you should trust the help: planning first, then feedback, then marking, then the pupil-data check that applies to all of them.

Planning: where the time savings are

The best evidence is for lesson planning. A trial funded by the Education Endowment Foundation and run by NFER tested ChatGPT for planning science lessons for Years 7 and 8. The teachers who used it cut their planning time by 31% on average, around 25 minutes a week, compared with teachers told not to use generative AI. A blinded panel of science teachers found no difference in quality between the two groups (NFER, "AI: help or hindrance?", March 2025). NFER notes that the sample of resources was limited.

What the planning trial found

31%

less planning time on average, around 25 minutes a week

No difference in quality

as judged by a blinded panel of science teachers

Source: NFER, “AI: help or hindrance?” (March 2025), describing the EEF trial of ChatGPT for science lesson planning (Years 7 and 8)

That trial used ChatGPT for around a third of lessons, according to NFER. The honest reading is that AI is a good first-draft assistant for parts of planning. It is not a lesson-planning machine.

Several tools are built for the teacher's own work. As the tools describe themselves in our Learning Tools Index:

  • MagicSchool AI generates lesson plans, levelled texts, assessments, feedback and parent communications from teacher prompts.
  • Eduaide.ai offers generative tools for lesson planning, resource creation, differentiation and assessment.
  • Diffit adapts a text to different reading levels, translates it and generates aligned activities.
  • Curipod generates an interactive lesson from a prompt, with polls and open response.
  • Twee generates texts, dialogues and questions for English-language teachers.
  • Toddle includes an assistant for planning units, assessments and rubrics, grounded in the teacher's own curriculum, in an IB and international-school platform.

We list what each tool says it does. We have not tested them here, and we do not rank them. Whatever you choose, check every fact, example and reading-level claim before it goes in front of a class. A well-formatted plan with a wrong fact in it is still a wrong fact.

Feedback: draft it, then make it yours

Tools that suggest comments on pupil writing can save time, and the same trial logic applies: a draft you edit is safer than text you send untouched. Brisk Teaching, for example, is a browser extension that generates feedback, rubrics and levelled text in place. CoGrader evaluates essays against a rubric and produces a score and per-criterion feedback.

Keep two things yours. The first is the choice of the one thing a pupil should work on next, which depends on knowing the pupil. The second is the conversation. A pupil who reads automated comments from a tool, with no teacher voice behind them, in our view learns less from them than from a teacher who talked to them.

Marking: the part to hold most tightly

A tool can sort the pile. The mark is still yours.

AI marking is the topic with the most hype and the least support. A preprint testing ChatGPT on essay and short-answer scoring against human raters found that a ChatGPT-based scorer achieved agreement close to human raters on some datasets, but its overall performance was inconsistent and often lower than human scores. The author concludes it can complement human scoring but needs further development to be reliable for high-stakes assessment (Shermis, arXiv, 2024). It is a preprint, so we weigh it lightly. It does match what a teacher would expect: a tool may handle a clear mark scheme and stumble on nuance.

Some tools work on a different principle from a chatbot. Gradescope uses machine learning to group near-identical answers so a grader scores a group once. Crowdmark is a collaborative grading workflow where pages are scanned and graded against shared rubrics, and its AI-assisted grading is described as a recent addition. In both, a human grader stays in the loop, and you should confirm any mark that counts.

A practical rule: use AI marking support where a wrong mark is cheap to correct (a low-stakes quiz), and keep your own hands on anything that feeds a grade, a report or a decision about a pupil. Exam boards have their own rules on assessment, and on what must be marked by a person, so follow them over anything here.

What to hand over and what to keep

What to hand to a tool and what to keep

  • PlanningA tool can draft: A first draft of a lesson outline, starter questions, a differentiated text.Keep it human: What the class needs next, the sequence, and checking every fact.
  • MarkingA tool can draft: Sorting and a first pass on short-answer work, if you check a sample.Keep it human: The mark that counts, anything borderline, and every extended answer.
  • FeedbackA tool can draft: A first-draft comment you edit before the pupil sees it.Keep it human: The conversation, and which one thing the pupil works on next.
  • Reports and parentsA tool can draft: A tidy first draft from your own notes, with no names in the prompt.Keep it human: What is said about a child, and how it is said.
A starting split, not a rule. Your school's policy and the exam board's rules on marking come first.

The pattern is the same in each row. The tool is useful for the first draft, and the teacher is responsible for the final one. A quick test: if a parent asked who wrote this comment, could you say "I did, with a tool's first draft"? If not, you have handed over too much.

The pupil-data checklist

This is the part that matters most, and the part marketing pages say least about. England's Department for Education recommends that personal data is not entered into generative AI tools, and expects teachers and leaders to use professional judgement with them (DfE, Generative AI in education). Its product safety expectations, aimed at the developers and suppliers of edtech, include standards on cognitive development, emotional and social development, mental health and manipulation, and schools can use them to judge products (DfE, Generative AI: product safety expectations). If you work outside England, your own education ministry and data-protection law set the rules, so use the checklist as a prompt and not a legal standard.

Before pupil work goes into any AI tool: seven questions

  1. 1

    Can I do the job with no pupil names, photos or identifiers in the prompt?

    The safest data is data that never leaves your hands.

  2. 2

    Does the school or trust have an approved list of tools?

    Using a tool that is not on it may break school policy, whatever the tool's terms say.

  3. 3

    Where is the data processed and stored, and is that in the tool's published terms?

    Look for it in the privacy notice, not the marketing page.

  4. 4

    Is pupil input used to train the tool's models, and can that be switched off?

    Ask in writing if the terms are unclear.

  5. 5

    Is there a school or education account with different terms from the free one?

    Free and school plans often have different data terms.

  6. 6

    What is the minimum age in the terms, and who is the account holder?

    Do not ask pupils under the stated age to sign in to anything.

  7. 7

    Can I delete pupil work, and is there a record of who can see it?

    If you cannot delete it, do not upload it.

A checklist, not legal advice. Your school's data protection lead has the final say.

Three habits make the checklist easier.

  • Strip identifiers. Remove names, photos, class lists and anything that could identify a child before you paste anything. Use a code, not initials, and keep the key yourself.
  • Use school accounts, not personal ones. A school or education plan often has different data terms from a free personal account. Ask your data-protection lead which is approved.
  • Never put sensitive information in. Safeguarding notes, SEN or health information, and behaviour records do not belong in a general-purpose tool.

If a tool cannot answer these questions in its published terms, that is your answer.

Choosing a tool: three questions

  1. What does it free me to do? If the answer is "more time with pupils", good. If it is "another dashboard", pause.
  2. Does it fit my system? A tool built around one country's curriculum may not match your exam board. Check before you plan around it.
  3. Who is accountable when it is wrong? You are. Pick a tool whose output you can check quickly.

Frequently asked questions

What is the best AI for teachers? It depends on the job. For planning and differentiation, tools built for teachers are the starting point. For marking, treat any tool as a first pass. Our Learning Tools Index describes what each says it does.

Can AI mark student work? It can assist with sorting and a first pass on some kinds of work. The research we found shows it is inconsistent on essays, so a teacher should confirm any mark that counts.

Is it safe to put pupil work into an AI tool? Only after the checks above. Remove identifying details, use an approved account and confirm the tool's data terms. If you cannot, do not upload it.

Will AI save me time on planning? In the trial described above, yes, by about a third on average, without a loss of quality as judged by a panel. Results will vary with how you use it.

What should I always do myself? The mark that counts, the conversation with the pupil, and every decision about a child.

Sources

  1. NFER, "AI: help or hindrance?" (March 2025), on the EEF trial of ChatGPT for lesson planning: https://www.nfer.ac.uk/blogs/ai-help-or-hindrance/
  2. Shermis, "Using ChatGPT to Score Essays and Short-Form Constructed Responses," arXiv:2408.09540: https://arxiv.org/abs/2408.09540v1
  3. UK Department for Education, "Generative artificial intelligence (AI) in education": https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education/generative-artificial-intelligence-ai-in-education
  4. UK Department for Education, "Generative AI: product safety expectations": https://www.gov.uk/government/publications/generative-ai-product-safety-expectations/generative-ai-product-safety-expectations