The best-documented advantage of AI in education is teacher time: in a randomized trial of 259 secondary teachers in England, ChatGPT cut weekly lesson planning from 81.5 minutes to 56.2 minutes. The best-documented disadvantage is that students can finish a task faster or better with a chatbot and still learn no more, a pattern the OECD's Digital Education Outlook 2026 reports from several controlled studies. Outcomes depend on how a tool is designed and used, not on whether AI is present.
Key takeaways
- Teachers in the English trial saved an average of 25.3 minutes per week on Year 7 and 8 science planning, a 31 percent reduction, with no noticeable quality difference in the sample of materials reviewed.
- In a randomized essay-revision study cited by the OECD, students who used a general-purpose chatbot scored highest on the task but showed no improvement on a knowledge test.
- The OECD concludes that tools aligned with learning science can lead to better learning, and that teachers can use general-purpose tools well when they embed them in a clear teaching plan.
- Most results come from short interventions. The OECD notes that few studies follow repeated use over longer periods.
Teacher planning time: the advantage with the clearest trial
The Education Endowment Foundation published an independent trial, led by the National Foundation for Educational Research, on December 12, 2024. It covered 259 teachers in 68 secondary schools across England. A total of 129 teachers in 34 schools were randomly assigned to use ChatGPT, together with an implementation guide, to prepare Year 7 and 8 science lessons. The comparison group was asked not to use generative AI.
The ChatGPT group averaged 56.2 minutes of weekly planning, against 81.5 minutes for the comparison group, a difference of 25.3 minutes. Teachers used the tool to support about one-third of their lessons, most often to plan a single activity. An EEF press release says an independent panel of teachers found no noticeable quality difference, and it warns that the finding rests on a limited sample of submitted resources and should be treated with caution. The trial was in England, in one subject and age range, so it does not measure Canadian or U.S. classrooms.
Student learning: gains depend on the design
The OECD's Digital Education Outlook 2026 reviews the evidence and reaches a conditional verdict. Educational tools aligned with learning science can lead to better learning, and teachers can use general-purpose tools effectively if they embed them in a clear teaching strategy. On feedback, the report says AI-generated comments often match or surpass human feedback despite being less reliable and consistent, yet most studies find that students rate human feedback as more meaningful, trustworthy, and motivating.
The same report lists system-level uses, such as drafting standardized assessment items and checking curriculum alignment, as possibilities. These are described as potential uses, not as measured gains.
Performance without learning: the disadvantage with the most support
The OECD describes a randomized trial by Fan and colleagues with Chinese university students who revised an essay in four ways: alone, with expert advice, with a checklist, or with a general-purpose chatbot. The chatbot group scored highest on the essay, but its scores on a test of knowledge did not improve, and it carried out fewer metacognitive tasks such as evaluating its own work. A related study by Chen and colleagues found that some students asked the chatbot for help and applied the answer at once, skipping diagnosing, evaluating, and iterating. The authors call this metacognitive laziness. A field experiment with Turkish high school students found a similar split in mathematics, with a closed-book penalty for the general-purpose version.
The OECD adds that teachers face the same risk. Reliance on AI for marking, feedback, or planning could wear down professional skills and autonomy.
What students say they use it for
Survey data in the same report show why the learning risk is plausible. In Estonia, students in grades 6 to 12 most often said they used generative AI to get better scores, make tasks easier, and save time. In a 2024 survey of 7,000 respondents in seven European countries, asked about learning outside school without teacher instruction, 56 percent used AI to obtain information and 45 percent to have concepts explained, while 31 percent used it to produce complete solutions and 20 percent to plan or track their own learning.
Advantages and disadvantages, matched to evidence
| Claim | Best evidence found | Limit |
|---|---|---|
| Saves teacher time | Randomized trial, 259 teachers, 31 percent less planning time | One subject, one country, short trial |
| Improves student learning | OECD review: possible with tools built for learning | Depends on design; many studies are short |
| Improves feedback | OECD review: matches or surpasses human feedback in some respects | Students rate human feedback higher |
| Raises scores without learning | Essay-revision trial: top task score, no knowledge gain | University students in China |
| Replaces effort | Student surveys: 31 percent use AI for full solutions | Self-reported use, not outcomes |
Questions to ask about any pro or con
- Was there a comparison group, and were people assigned at random?
- Did the study test learning later without the tool, or only performance with it?
- Was the tool built for teaching or a general-purpose chatbot?
- How long did the trial run, and in which country and grade?
FAQ
Does AI improve grades?
It can improve scores on tasks done with the tool. The studies reviewed by the OECD show that this does not reliably carry over to later tests taken without it.
Is the teacher time saving large enough to matter?
The English trial measured 25.3 minutes per week for one subject and age group. How that scales to a full teaching load was not tested in the trial.
Are the disadvantages caused by AI itself?
The evidence points to use patterns. In the study by Chen and colleagues, some students asked the chatbot for answers and applied them at once, skipping the steps of diagnosing and evaluating, and the OECD review reports better results for tools built around learning science.
Do these findings apply in Canada and the United States?
The trials cited were run in England, China, and Türkiye. No North American trial is included here, so the results are evidence about mechanisms, not a forecast for a specific school.
Related reading
- Documented limits of AI in education
- AI and students' academic writing
- AI activities for ESL students
- Wider effects of automation and AI
More in AI in education.
Last verified: October 2026
Sources
- OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education, OECD, January 2026
- Teachers using ChatGPT, alongside a guide to support them to use it effectively, can cut lesson planning time by over 30 per cent, Education Endowment Foundation press release, December 12, 2024