Does AI save teachers time? What the evidence actually says

25.3 minutes a week. That is how much planning time science teachers in England saved when they used ChatGPT with a short guide, in the largest controlled trial so far (EEF, December 2024). So does AI save teachers time? In a trial, yes, a little. In a normal school week the answer is less clear, and the gap between the two is what this post is about.

What the trials actually measured

The Education Endowment Foundation ran the trial with 259 science teachers in 68 secondary schools in England. Half used ChatGPT to plan lessons for classes of 11 to 13 year olds. Their weekly planning time fell from 81.5 to 56.2 minutes, a cut of 31 percent.

An independent panel of teachers compared a sample of the lessons and found no noticeable difference in quality. The evaluators say that finding should be treated with caution, because the sample was small.

A second trial tested Aila, the planning assistant from Oak National Academy, in 86 primary schools in England in autumn 2025, evaluated by NFER. Teachers using it spent about 2.5 hours a week on preparation against 3.3 hours for those who did not, roughly 49 minutes saved (EEF, Aila trial, 2025).

Then the awkward number. Aila was used for only about 7 percent of lessons in an average week, and use fell over the ten weeks. The saving was real. The habit was not.

What teachers report in a normal week

Surveys tell a different story from trials. In a YouGov poll of 1,033 UK teachers for Bett, 52 percent used AI every week, double the year before. Among AI users, 51 percent said it eased their workload. Only 35 percent said they worked fewer hours (Tes, August 2026).

Half feel lighter. A third actually go home earlier.

Across Europe the picture is similar. Sanoma Learning surveyed more than 20,000 teachers in 14 European countries, including Finland and Sweden. 63 percent now intend to use AI to prepare learning materials, up from 49 percent in earlier rounds, but only 16 percent believe general-purpose AI improves learning outcomes (Sanoma Learning, September 2026).

The OECD's TALIS 2024 survey of lower secondary teachers found that about 41 percent used AI in their teaching, on average across OECD systems. Of those, 64 percent used it to generate lesson plans or activities. 66 percent of teachers believed AI can make inappropriate or incorrect recommendations (Education International on TALIS 2024, December 2025).

Why the saved minutes disappear

We usually blame the tool, or a lack of training. I think it is mostly something else: where the saved time goes.

In Denmark, the teachers' union asked 1,168 teachers in Danish public schools about their preparation in spring 2026. 78 percent said their preparation time goes to other tasks at least once a week. 72 percent said that at least once a week they cannot prepare teaching matched to individual kids' level (Danmarks Lærerforening, February to March 2026).

So a teacher who saves 25 minutes on Tuesday's plan does not get 25 minutes back. The minutes go to the team meeting, the parent email, the kid who needs a different version of the worksheet. When I taught in a Danish public school, the planning I did at the kitchen table at 21:00 was rarely the lesson itself. It was the second and third version of it.

There is a second leak, and this one is on us, the people who build the tools. A teacher who uses one tool for a quiz, another for slides, a chatbot for ideas and a fourth for differentiation spends time logging in, copying, reformatting and checking each output. I know of no trial that measures this. I have watched it happen in staffrooms, and I think it eats a large part of the saving.

Third, checking takes time. Two in three teachers in TALIS expect AI to get things wrong sometimes, and they are right to check. A plan you read line by line is faster than a plan you write from scratch, but not by as much as the marketing says.

AI can shorten the writing of a lesson, but only the school decides whether the teacher gets those minutes back.

So does AI save teachers time? A fair summary

• In controlled trials in England, AI planning support saved about 25 to 49 minutes a week, with no clear drop in quality (EEF, 2024 and 2025).
• In surveys, about half of the teachers who use AI feel less workload, but only about a third work fewer hours (Bett and YouGov, 2026).
• Neither trial measured what kids learned. That is the question that matters most, and it is still open.

What a school can do

1. Measure before you buy. Ask three teachers to log their planning minutes for two normal weeks. About 5 minutes a week each. Without a baseline, any saving is a guess.

2. Pick one tool per job, not five. Fewer logins, fewer formats, less copying. One leadership meeting to agree, then it holds.

3. Decide in advance where saved time goes. If a team saves 30 minutes a week, protect it on the timetable, for differentiation or for the kids who are falling behind. Otherwise it disappears into admin.

4. Keep the teacher in the loop. Every AI-made plan gets read before it reaches a class. Put those 5 to 10 minutes into the estimate, honestly.

5. Review after six weeks with the same log. If use has dropped, as it did in the Aila trial, ask the teachers why before you renew anything. About 30 minutes for one conversation.

Where GradeAid fits

GradeAid is a Danish platform for teachers. The teacher types in the topic and the level and gets a lesson plan, a worksheet, a quiz, a presentation and screen-free activities in seconds, which you can also differentiate. The material can be printed. We have not measured how much time it saves our teachers, and we will not claim a number until we have; what we do know is that 47.6 percent of worksheets get edited after generation (September 2026), so the teacher's check is real work and we count it.

If you work with teachers and have measured what AI does to your planning week, we would like to hear from you. Write to us at founders@gradeaid.ai.