In June, 68 percent of early-career teachers in England said they had used AI for lesson planning or preparation in the previous month. Among other classroom teachers the figure was 54 percent. The data comes from Teacher Tapp, a daily teacher survey, as reported by EdTech Innovation Hub, and it describes the one group in a school whose job is to learn how to plan a lesson.
An induction tutor should neither ban this nor wave it through. Change what the weekly mentor meeting looks at: the plan the new teacher made before a machine saw it, and the lesson afterward. Planning is where a new teacher learns what children find hard, and a draft that arrives finished takes away the part that taught her.
How many early-career teachers use AI, and what does the data not tell us?
About two in three early-career teachers in England used AI to plan lessons in June, ahead of their more experienced colleagues by 14 points. Teacher Tapp's figures, as reported by EdTech Innovation Hub on July 31, cover the 2025-26 school year and compare 359 early-career teachers with 2,739 others on the lesson-planning question. The gaps elsewhere are smaller: 18 percent of early-career teachers used AI to write to parents about individual pupils against 11 percent of the rest, 12 percent against 7 percent for marking or feedback, and 6 percent against 2 percent for timetables and seating plans. For writing pupil reports and developing curriculum, the two groups look alike.
Three limits matter before anyone builds a policy on those numbers. The sample of early-career teachers is a few hundred. The report does not define the term, though it describes the comparison group as teachers not completing the Early Career Framework. And I could not find the original Teacher Tapp release, only the coverage of it.
The wider picture is steadier. In a Chartered College of Teaching article published May 18, 2026, Karen Wespieser of Teacher Tapp reported that only 12 percent of English teachers had never used AI for schoolwork, while only 14 percent said they had used it to plan their most recent lesson, 13 percent partially and 1 percent completely. That article asked a different question of a different period, so do not set its figures beside the 68 percent. Read together, they say that AI in planning is ordinary, and that most planning still starts without it.
Why does lesson planning matter more for a new teacher than for a veteran?
Planning matters more for a new teacher because, for her, it is how she learns the job. A veteran plans from a store of pupils' wrong answers; a new teacher is building that store, one prediction at a time.
Watch what planning teaches. She picks an example and finds it confuses her own tutor group. She puts a question first and discovers that nobody knows how to start. She guesses what a struggling pupil will say, teaches the lesson, and finds out that the pupil said something else. Each of those misses is the curriculum of the first two years. The Department for Education (DfE) gives those two years a structure: in its guidance on setting up and managing the entitlement, last updated August 27, 2025, early-career teachers are "entitled to support from a dedicated mentor" over the first two years of their career. The statutory Induction for early career teachers (England), which carries a June 2026 date on the edition I read, caps a new teacher's timetable at 90 percent of an existing teacher's in the first year and 95 percent in the second, and says the time is to be used for her training and mentoring. In the parts I read, neither document mentions AI. That is not a complaint. It marks where the gap sits.
I offer one way of seeing the gap, and I offer it as a suggested filter, not a validated instrument. I call it the Practice Question, and I set it out first in what happens to entry-level work when AI does it:
The Practice Question is three questions I ask before handing any junior task to AI: What did doing this badly, and being corrected, teach the person who used to do it? Where will they learn that now? And who would notice if they never did? A task that survives all three can be automated with a clear conscience. A task that fails any of them is not a chore. It is a classroom, and you are deciding to close it.
Every junior task does three jobs: it produces the work, it produces the worker, and it produces the evidence that the worker can be trusted. A first-year teacher's lesson plan does all three. Put it through the Practice Question and it fails the first test at once. Doing it badly, and being corrected, taught her where pupils get stuck. If AI drafts the plan, she has no obvious place to learn that. Checking is not the same activity as doing, and it does not build the same person.
Is there a case for letting new teachers plan with AI?
There is a real case, and it rests on two facts: a new teacher's week is overloaded, and a blank page is a poor teacher.
The strongest evidence I found is small and I hold it loosely. A study in the journal Contemporary Issues in Technology and Teacher Education (volume 26, issue 2), by Demetrice Smith-Mutegi, Monica Miles and Olayinka Mohorn-Mintah, followed nine early-career teachers in a science and mathematics methods course in the southeastern United States. The design was exploratory and the data self-reported, and the first author both taught the course and ran the research. Even so, the finding is worth keeping. The teachers used the tools "as brainstorming aids rather than as sources of finalized content," and one said: "It got me thinking and unstuck." Nine people cannot tell you what happens in your school. They can tell you what to look for: AI that starts a teacher's thinking is a different thing from AI that finishes it.
The Chartered College article gives the practical side. Among teachers who had not used AI, 35 percent said it would not save time because the output needs checking. A new teacher is the person least equipped to check a lesson plan, since the checking needs exactly the knowledge she is there to acquire.
So the case holds for work that does not train her: a first draft of a letter home, a seating plan, the formatting of a resource she has already designed. It weakens as the task moves toward deciding what to teach and predicting how it will land. That boundary is what an induction tutor has to draw, and what AI guidance teachers need, task by task shows how to draw it for the whole staff.
What should the weekly mentor meeting look at now?
The weekly meeting should look at three things in order: the plan she made alone, the comparison with what AI produced, and the defense of her choices. I call the routine attempt, compare, defend. It is my suggestion, drawn from the Practice Question and untested in any trial.
Attempt. Once a week, the new teacher plans the skeleton of one lesson without AI. Not the slides or the worksheet: the question that opens the lesson, the wrong answer she expects, and what a struggling pupil will say first. It can run to a page. What matters is that it exists before the machine's version does.
Compare. She then asks AI for its own plan and lays the two side by side. She marks what the AI plan has that hers lacks, and what hers has that the AI plan missed, such as the misconception from Tuesday's lesson, the pupil who will not volunteer, the example that will land in this school and no other.
Defend. In the mentor meeting she explains her choices in a sentence each: why this example, why in this order. The mentor asks one question: where will they get stuck? After the lesson they ask the second: which wrong turn did you not predict?
Here is a hypothetical case. A Year 8 mathematics teacher in her first term is planning a lesson on adding fractions. Her own skeleton predicts that pupils will add the numerators and the denominators, because that is what happened in last week's starter. The AI plan opens with a clean worked example and never mentions the error. In the meeting, the comparison takes her five minutes and teaches her more than the plan itself did. The AI plan was fine. The skeleton was hers.
The Mistake I See Most Often
The mistake I see most often is a school that asks whether its new teachers are using AI and never asks what the use is replacing.
I was a secondary teacher and an assistant headteacher before I started training educators, and in the leadership conversations I have about new staff, the question that arrives first is nearly always whether to allow it. It is the wrong question, because "allow" draws one line around everything from a parent letter to a scheme of work. The better question is task by task: does this one train her? A school that cannot answer it for six tasks will find that the answer arrives anyway, in the form of new teachers who plan faster in October and cannot say in March why their lessons are built as they are. I would treat that last symptom as the thing to watch for, and my guess is that most schools will not see it until they look for it.
What should an induction tutor decide this term?
The induction tutor should make four decisions this term, and write each one down in plain sentences.
First, name what stays hers. For the first term, the lesson skeleton is planned unaided, once a week. Everything else gets a clear yes or no, in language a first-year teacher can apply at 9 p.m.
Second, brief the mentors. A mentor who has never compared an AI plan with a human one cannot run the meeting above. Give them the same exercise first. The wider case for giving staff more than tool training is in AI skills for staff: why basic training stops short of judgment.
Third, decide where the time goes. If AI saves a new teacher two hours on a Sunday, that time needs a destination before it is released: observation, another mentor conversation, a visit to a colleague's classroom. Otherwise it fills with more work. What happens after the initial AI training argues the same about any hours AI returns.
Fourth, decide what you will look for at Christmas. Three things a leader can see without a survey. Can she predict, before teaching, where pupils will stumble, and is she right some of the time? Can she explain, in a sentence per choice, why a lesson is shaped as it is? Can she say what she changed in an AI draft, and why? If the answer to all three is yes, the tool is helping. If the answer to any is no, you have found the classroom that was closed.
This argument was written about England, where induction runs two years and carries a statutory mentor. In the United States, new-teacher induction varies by state and district. The structure differs; the question of what a first-year plan teaches does not.
The next step
If your leadership team is working out what its newest teachers should do with AI, and what your mentors need to be ready for, this is the kind of work I support through AI strategy sessions for schools, alongside the capability questions that sit underneath it. For the wider case on building capability without creating dependence, start with is your AI training building capability, or dependence?
Sources and further reading
- Emma Thompson, "England's early-career teachers lead experienced peers in classroom AI use, Teacher Tapp data shows," EdTech Innovation Hub, July 31 (the page gives no year; the data runs through July 2026). edtechinnovationhub.com
- Karen Wespieser, "From novelty to normal: How teachers are using AI in 2026," Impact, Chartered College of Teaching, May 18, 2026. my.chartered.college
- Department for Education, "Set up and manage the early career teacher entitlement," GOV.UK, published April 22, 2025, last updated August 27, 2025. gov.uk
- Department for Education, "Induction for early career teachers (England): statutory guidance for appropriate bodies, headteachers, school staff and governing bodies," June 2026 edition. gov.uk
- Demetrice Smith-Mutegi, Monica Miles and Olayinka Mohorn-Mintah, "The Perceptions of GenAI Tools in Lesson Planning: Implications for Science Teacher Education," Contemporary Issues in Technology and Teacher Education, 26(2), 2026. citejournal.org
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and the author of bestselling books on AI in education. Read more about Dan.
