AI Tools for Teachers: What's Actually Worth Your Time

Most articles about AI for teachers are written by people who have never had thirty books to mark on a Sunday evening. They list twenty tools, describe each in marketing language, and leave you to work out which of them would survive contact with an actual week of teaching.

This guide is organised by the job instead — planning, differentiating, marking, admin — and it is honest about the two things that matter most and get discussed least: where AI marking stops being appropriate, and what happens to student data when you paste it into a chatbot.

Start here if you have twenty minutes

If you only try one thing, try differentiation. It is the single strongest genuine use of AI in teaching, and it addresses the task that eats the most time for the least professional satisfaction.

Take a resource you already use. Paste it into ChatGPT or Claude with:

Rewrite this at three reading levels — significantly below, at, and above the level of the original. Keep the same content, the same questions and the same structure. Do not simplify the concepts, only the language.

What took an hour takes about four minutes, and the output is genuinely usable after a read-through. That last clause matters — always read it. It occasionally simplifies a concept when you asked it to simplify only the language.

Lesson planning and resources

What it is good at: generating starter activities, producing worked examples, writing comprehension questions at a specified level, suggesting misconceptions to address, and turning your outline into a resource.

What it is bad at: knowing your class, your curriculum specification, or what you actually did last lesson. It will confidently produce something aligned to a different exam board.

A prompt pattern that works better than “plan me a lesson on X”:

I teach [subject] to [year group] following [specification]. They have just covered [previous topic] and struggled with [specific difficulty]. Suggest three starter activities for [new topic] that address that difficulty. Ten minutes each, no printing.

The constraints are doing the work. Vague prompts get generic lessons.

For slides, Gamma will turn a document into a deck in minutes — useful for a quick internal presentation, though you will want to rewrite every headline. For worksheets and display material, Canva remains the most practical option, and its free tier is genuinely generous.

Turning your own materials into study aids

This is where NotebookLM earns its place, and it deserves to be better known among teachers.

Upload your scheme of work, your slides and the set text. It answers only from those sources and cites the passage it used, so it will not invent content that is not in your materials. From there you can generate revision guides, question banks and — the feature students actually engage with — Audio Overviews they can listen to on the bus.

It is free, with limits generous enough for a full module. Our NotebookLM study guide walks through the workflow in more detail from the student side.

Marking and feedback: where to stop

This section is the reason we wrote the guide.

AI is genuinely good at drafting formative feedback. Give it a mark scheme and a piece of work, and it will produce specific, constructive comments faster and more consistently than you can at 9pm. Used as a first draft that you edit, it is a real reduction in workload.

AI should not decide grades. Three reasons, and they are not hedging:

1. It is inconsistent. Submit the same piece twice and you can get different marks. That is disqualifying for summative assessment.
2. It cannot see the student. It does not know this is the first time that child has written three paragraphs unaided. Marking without that context is not marking.
3. Accountability does not transfer. If a grade is challenged, “the AI decided” is not an answer you can give a parent, a head of department, or a moderator.

A workflow that respects the line:

• You read the work and decide the grade.
• AI drafts the written feedback against your mark scheme.
• You edit it, adding the one specific observation only you could make.

That final step is what makes the feedback worth reading, and it takes thirty seconds.

Admin and communication

The least glamorous and possibly highest-value category.

Report comments, parent emails, letters, risk assessments, meeting notes, newsletter copy. AI is good at all of it, because these documents are formulaic by design.

For report writing specifically:

Write a report comment for a student who has [specific observations]. Warm but honest, 60 words, no clichés, must include one concrete next step.

Then edit. The tell of AI-written reports is that every child is “a pleasure to teach who consistently demonstrates” something. Specificity is what makes a comment land, and specificity has to come from you.

Grammarly is worth having installed for everything outward-facing. The free tier covers what most people need.

By subject

A few uses that hold up in specific departments.

English. Generating model answers at different grade boundaries so students can see the difference is genuinely powerful — and far more useful than another exemplar essay. Also good for producing comprehension questions on any text at a specified level.

Maths and sciences. Ask for worked examples with deliberate errors in them and have students find the mistake. Verify every generated calculation yourself before using it; these tools still make arithmetic errors and state them confidently.

Humanities. Ask for the strongest argument on each side of a contested question, then have students evaluate which is better supported. It is good at producing balanced positions, which for once is the point.

Languages. Generating varied practice sentences and conversation scenarios at a target level. Have a fluent colleague sanity-check idiom.

Primary. Differentiated reading passages on a topic you are already covering, and simplifying instructions to an age-appropriate register.

The recurring jobs it removes

Ranked by hours returned per week, in our reading of what teachers report:

1. Differentiating an existing resource — the biggest single win.
2. Drafting formative feedback against a mark scheme you supply.
3. Report comments, edited for specificity afterwards.
4. Parent communication — first drafts of difficult emails, which take disproportionately long to write from scratch.
5. Turning a document into slides with Gamma.
6. Producing question banks from your own materials in NotebookLM.

Notice that none of these are “plan my lessons for me”. The wins are in the surrounding admin, not the teaching.

The data-protection question

Here is the part that most teacher-facing AI content skips entirely.

Student work is personal data. When you paste it into a consumer AI tool, you are sending it to a third party — and on consumer plans, that content may be used to improve the model unless you have turned that setting off.

Practical rules that keep you on the right side of this:

• Anonymise before you paste. Remove names, and remove details that identify a student even without a name. This is usually enough.
• Check whether your school has a policy. Many now do, specifically about AI. Ask before assuming.
• Prefer institutional accounts where your school provides them. Education and business plans exclude your data from training by default; consumer plans do not.
• Never paste safeguarding material. Not anonymised, not ever.

None of this makes AI unusable in a school. It just means the ten seconds it takes to strip a name is not optional.

What about students using it?

Briefly, because it is the question that always comes.

Detection does not work reliably. AI-detection tools produce both false negatives and false positives, and a false accusation of cheating does real harm to a student. Building your response around detection is building on sand.

Designing around it does work. Assessment that assumes AI access — in-class writing, oral defence of a submitted piece, work that requires reference to a specific lesson discussion, drafts submitted alongside finals — is more robust and more useful than an arms race.

And teaching students to use it well is more valuable than forbidding it. They will have it for the rest of their working lives.

What we would actually install

If you are starting from nothing:

1. ChatGPT free — planning, differentiation, feedback drafting. The workhorse.
2. NotebookLM free — your own materials, turned into study aids students will use.
3. Canva free — worksheets, displays, slides.

That is three free tools covering most of it. Add Claude if you write long documents — policies, schemes of work, reports — and want something that handles length well.

Do not pay for anything in your first month. Our guide to what you actually get on free AI plans covers where the free tiers genuinely stop, and for most teaching tasks they do not stop anywhere near where you would expect.

If you are pointing students at tools rather than using them yourself, our best free AI tools for students is written for them directly.