AI lesson planning: what the trials show teachers

The short answer

Use AI for the parts of planning that need hands, not judgement: question banks, quizzes, cover work, a reading passage at a different level. Decide first what students should learn, which mistake you expect and how you will check, then prompt with a class description that contains no names, and check every source the tool gives you. In the best trials this saved teachers roughly 25 to 49 minutes a week, so treat AI as a modest helper, not a rescue.

Does AI lesson planning actually save time?

Yes, but less than the headlines suggest. Chapter 4 of my book Teaching in the AI Era, “Planning with a Tireless Assistant”, starts from the strongest evidence available: randomised trials in which teachers recorded their planning time every week against a comparison group.

  • ChatGPT in English secondary science (2024). Teachers in 68 schools planned Year 7 and 8 lessons with ChatGPT and an online guide, or without generative AI. The ChatGPT group spent 56.2 minutes a week planning against 81.5, a saving of 25.3 minutes, and a blind expert panel found no evidence of a difference in quality (Roy et al., 2024).
  • Oak National Academy's Aila in primary schools (2025). 464 primary teachers in 108 schools took part. Teachers with Aila spent 2.5 hours a week planning against 3.3, a saving of 48.6 minutes, with resources no worse in quality (Roy et al., 2026).
  • Microsoft Copilot in Lima, Peru. Teachers wrote better prompts and became better judges of weak AI answers, but the trial found “precisely estimated null effects” on time (Glave et al., 2026).

The book's reading is plain: “The minutes are real, and they are far smaller than many of the claims teachers hear.” None of these trials measured what students learned from the lessons.

Why do surveys promise hours when trials find minutes?

In one US survey, teachers who used AI weekly estimated it saved them 5.9 hours a week (Gallup & Walton Family Foundation, 2025). Surveys ask people to estimate; the English trials asked teachers to record. Estimates drift with hope and with marketing. Oak's own newsletter promised Aila could “save five hours a week”, yet the measured saving was under fifty minutes. The book's advice: when a colleague, a vendor or a headline promises hours, ask how anyone measured them.

What did teachers actually use ChatGPT for?

Mostly small, repeated jobs. In the 2024 trial, most teachers used ChatGPT for one or two activities in a lesson, usually questions, quizzes or ideas for activities. Teachers said it wrote comprehension and recall questions faster than they could, and that most were broadly accurate. Cover work suited it well too: one teacher estimated that planning three cover lessons this way saved him “3 or 4 hours”.

For a teacher new to these tools, that suggests a sensible first step. Try AI on one routine task, such as a quiz, before trusting it with a whole lesson.

Why do some teachers save hours and others nothing?

In the same trial, one experienced teacher used ChatGPT for about nine in ten of her lessons with a low-attaining class. Her prompts described the class and what the students already knew, and she asked for refinements. Another used it for one lesson in ten and saved no time overall. What each teacher brought to the tool made the difference.

The book calls this teacher-design burden: the effort a teacher must put in before a tool helps a particular class learn, as distinct from simply having access to it. One trial teacher described how their prompts grew to carry “the year group, and their reading age level and what they should do”. That knowledge does not live in the machine. Someone has to supply it, and good prompting is itself design work.

What does an AI lesson plan get wrong?

Enough that checking is part of the job. The chapter gathers smaller studies that show what checking has to catch:

  • A sound first-grade science plan made with ChatGPT recommended a children's book that does not exist, “a fake book by a real author” (Powell & Courchesne, 2024).
  • Of 2,230 activities in 310 civics plans from ChatGPT, Gemini and Copilot, only 10% called for higher-order thinking, and nearly half asked students simply to remember (Trust et al., 2025).
  • Running the same lesson-plan prompts again and again, “simply by chance”, some plans missed over a quarter of the criteria while others were more than 90% complete (Dornburg & Davin, 2025).

This checking takes time that weekly totals may hide. A few teachers in England told researchers they had spent as long editing AI outputs as making the material themselves. A teacher who saves twenty minutes on writing may spend fifteen of them checking.

How should I use AI for lesson planning, step by step?

The chapter ends with practical routines. They draw on the research, and the book is clear that nobody has yet tested them as programmes.

  • Think first, then prompt. Before opening a tool, write three lines: what students should be able to do by the end, the mistake you expect and how you will check. Then ask for practice questions, and drop any that do not test your three lines.
  • Context cards with no names. Keep a short card for each class with the year group, typical reading age, prior learning and what worked last time. Paste it into your prompts and leave out names and anything else personal. If a name is needed, add it yourself after the AI has finished, and use tools your school has approved.
  • Check every source and one claim. Before a generated resource reaches students, check every book, website and quotation it cites, and at least one claim, against a trusted source.
  • Shared prompt banks. A department keeps the prompts that worked, each with a note on the class and topic, so one teacher's design work serves everyone.
  • One lesson from scratch. A mentor asks a trainee to plan one lesson a week without AI and talk it through, so the thinking still happens.
  • Two weeks of honest timing. Record every minute of planning, including prompting, checking and editing, and compare a week with the tool against a week without it.

The Chapter 4 page opens with the trial teachers' stories and lists the chapter's questions for reflection, ready to take to your next department or phase meeting. The full chapter gives a classroom example for each routine.

What should a teacher never hand over?

The decisions that make a lesson belong to a class. What should these students learn next, and which mistake will they make? Which example will reach them, and how will the teacher know? No tool in the studies the book reviews answered those questions on a teacher's behalf. The book puts the division of labour simply: “AI can carry part of the load, and the judgement stays with the person who knows the class.”

Differentiation and SEND paperwork need extra care. The book suggests using AI to produce versions of a text, while the teacher decides which student needs which version. For any formal plan, AI wording is a draft at most, written only after the people who know the child have gathered the observations and data.

Is AI lesson planning only for well-resourced schools?

No, but the saving does not reach everyone equally. Teacher-facing planning is the lighter route: one teacher with one phone can draft questions while the connection holds, and the students need no device at all. The design burden does not shrink, though. The chapter follows teachers in Pakistan, India and Sierra Leone who plan around power cuts, shared phones and unreliable signal, and argues that schools should budget protected time for this work rather than leave it to teachers' evenings.

Where can I read the book?

Teaching in the AI Era is free to read online, all thirteen chapters, with the full reference list. If you prefer your own copy, the ebook is available from your preferred store and the paperback is on Amazon. Chapter 4 is the place to start if planning is where your week disappears.

Questions people ask

How much time does AI lesson planning save?

In two randomised trials in England, teachers saved 25.3 minutes a week with ChatGPT and 48.6 minutes a week with Oak National Academy's Aila, with no drop in resource quality. A trial in Lima, Peru, found no time saving. Surveys report hours, but they rely on estimates rather than recorded time.

Can I use ChatGPT to write a whole lesson plan?

You can, but the evidence favours smaller jobs. Most trial teachers used it for one or two activities in a lesson, such as questions or quizzes. Generated plans can cite books that do not exist, lean on simple recall and vary in completeness from run to run, so every plan needs checking against what you know about your class.

What should I put in an AI lesson plan prompt?

A short description of the class: year group, typical reading age, prior learning and what worked last time, plus what students should learn and the mistake you expect. Leave out names and other personal details, and add those yourself after the AI has finished.

Is there a book on teaching with AI for school teachers?

Teaching in the AI Era by Dr Muhammad Tayyab Ilyas is written first for primary and secondary teachers in any subject. Chapter 4 covers lesson planning with AI, and the whole book is free to read online at tayyabcheema.com/teachingintheaiera/read.

About the author. Dr Muhammad Tayyab Ilyas is the author of Teaching in the AI Era, a book for teachers on AI in schools and universities. He wrote Teaching in the AI Era after more than a decade in Pakistan's education system, as an elementary school teacher and an assistant education officer in Gujranwala.