Nobody has yet measured it rigorously. The strongest trial, run by the Education Endowment Foundation (EEF) and NFER, found AI cut lesson-planning time by 31%, about 25 minutes a week. It deliberately excluded marking. Figures such as “up to 50%” on marking come from developers and vendors, so schools should measure their own time saved against a baseline.
What does the rigorous evidence show?
One robust trial, on planning, not marking. The EEF and NFER ran a randomised controlled trial with 259 teachers in 68 English secondary schools. Teachers using ChatGPT with a guide spent “around 56.2 minutes per week” preparing Year 7 and 8 science lessons, against “81.5 minutes” in the comparison group. That is a saving of “25.3 minutes per week on average”, with “a high security rating” (EEF/NFER evaluation report, December 2024).
The outcome measure explicitly excluded “time spent on teaching, marking, or administrative tasks”. Some teachers said they used the time saved “for marking” (14 of 68 who answered). The EEF itself noted that “marking and administration probably represent a greater burden”. A second EEF trial, of Oak National Academy’s Aila tool, also measures lesson planning; its report is due in autumn 2026.
Which marking time figures are claims rather than evidence?
Several widely quoted numbers come from developers, vendors or individuals:
| Figure | Source | Status |
|---|---|---|
| AI tools “could save time spent on formative assessment by up to 50%“ | Developers, quoted by the DfE in January 2025 | Developer estimate, not an evaluation |
| Grading time cut “from six hours to 20 minutes” | One interviewee in a DfE educator-views report (higher education) | Individual claim |
| ”At least 50% time-savings” | A vendor, quoted in the same DfE report | Vendor claim |
| Oak resources saved “4 hours per week on average” | Oak’s own impact report | Self-reported; resources generally, not AI marking |
Sources: DfE press release (13 January 2025); DfE, Generative AI in education: educator and expert views (January 2024).
None is wrong to report as a claim, but none tells a school what it will save. The DfE’s own product safety standards ask suppliers not to “exaggerate the impact or capabilities of their tools”.
How much time is there to save?
About four to six hours a week of marking per teacher, on OECD figures. TALIS 2024 found lower-secondary teachers spend 4.6 hours a week “on marking and correcting of student work” on average. The figure is 5.4 hours in the UAE and 6.4 in Singapore. Across OECD systems, 40% of teachers say too much marking causes them stress. Not all of that time can or should go to AI. Cambridge, Pearson and JCQ bar AI from being the sole marker of assessed work, so the realistic target is practice, quizzes and homework. See how to reduce marking workload.
How can a school measure its own time saved?
With a before-and-after comparison on the same tasks. A simple method:
- Baseline. For two weeks, teachers in one department log marking time by task type: homework, quizzes, mocks, coursework.
- Introduce AI for one task type only, such as weekly practice questions.
- Log again for the same two-week period the following half term.
- Compare the time on that task type, and check quality: are marks and feedback as good? See how to validate AI marking.
- Ask where the time went. Time saved on marking only helps if it goes to planning, feedback conversations or intervention.
Our marking workload calculator (classes × students × pieces per week × minutes per piece, divided by 60) gives the baseline figure.
What should leaders expect from AI on workload?
Real but specific savings, on routine practice marking, with teacher judgement kept for work that counts. Ofqual expects AI marks to become “more acceptable in contexts such as formative and low-stakes assessments”. That is where the time is likely to come from. Treat any headline figure as a hypothesis for your own pilot, not a promise.
How schools do this with AI Buddy
AI Buddy marks students’ practice answers to past-paper-style questions instantly, with AI feedback, so routine practice marking comes off teachers. Teachers see the results by class and topic, and keep their marking time for mocks and extended work. Tutopiya reports from its own school data up to 70% less marking time for teachers using AI Buddy. That figure is ours, not independently evaluated, so measure it in your own school with the method above before relying on it.
Frequently asked questions
How much time can AI save teachers on marking?
There is no rigorous published measure yet. The best trial evidence, a 31% saving, is for lesson planning. Marking figures such as “up to 50%” are developer or vendor claims.
Does AI reduce teacher workload?
For some tasks, yes: an EEF/NFER trial found AI cut lesson-planning time by about 25 minutes a week. For marking, schools should measure their own savings on practice work.
What is the evidence on AI and teacher workload?
One robust randomised trial on planning, several self-reported surveys, and developer estimates. The EEF’s evaluation of Oak’s Aila tool is due in autumn 2026.
Can AI mark coursework to save time?
Not as the sole marker. Cambridge, Pearson and JCQ require a human to decide marks for assessed work. AI can help with practice and formative work.
Discover how AI Buddy helps schools strengthen teaching, learning and evidence-informed school improvement. Or start a short consultation with our schools team using the form below — we will get back to you directly.
Sources
- EEF/NFER, ChatGPT in lesson planning: evaluation report (December 2024)
- EEF, Aila Teacher Choices trial
- Department for Education, AI teacher tools set to break down barriers to opportunity (13 January 2025) and Generative AI in education: educator and expert views (January 2024)
- OECD, Results from TALIS 2024: The demands of teaching
- Oak National Academy, Early insights on Aila (February 2025)