The Education Endowment Foundation has published its second trial of AI for lesson planning. The first, which tested ChatGPT with Key Stage 3 science teachers, found a 31% saving in preparation time. The new one tested Aila, Oak National Academy's purpose-built lesson assistant, with primary teachers in Years 3 to 6.
The headline result is clear. Over ten weeks in autumn 2025, teachers randomly allocated to use Aila spent around 2.5 hours a week on lesson and resource preparation, compared with 3.3 hours for teachers who did not. That is a saving of roughly 49 minutes a week. The EEF gives the finding a high security rating, and independent assessment found the Aila group's resources were no worse in quality.
The more interesting results sit underneath.
Free, curriculum-aligned and still rarely used
Of the teachers given Aila, 81% used it at least once. But it was used for only around 7% of lessons in an average week, and use declined over the ten weeks. Many teachers preferred other AI tools. Fewer than one in five felt Aila had meaningfully changed their teaching or planning.
The EEF's explanation is practical. Aila is built to help plan a lesson from scratch, and teachers did not often need to do that. They had existing planning habits, and they still had to adapt what Aila produced before using it. A tool designed for one part of the job did not fit easily into the rest of it.
The lesson for schools is that giving staff a tool is not the same as giving them a better workflow. The EEF suggests schools consider whether teachers need dedicated time to learn a tool like this, because that may be what turns occasional use into lasting time savings.
Who benefited most
Benefits were greatest for teachers newer to the profession, teachers with lower confidence in their subject or pedagogical knowledge, and those more confident with technology. Teachers also found Aila particularly useful when planning lessons on unfamiliar topics.
That points to a targeted use rather than a whole-staff rollout. An early career teacher planning a unit they have never taught, or a primary teacher covering a subject outside their specialism, may get the most from a structured assistant. An experienced teacher with a well-worn scheme of work may not.
The warning in the findings
The EEF is careful about what time saved means. It says teachers still need to check and adapt AI-generated materials so they are accurate, well sequenced and right for their pupils. It also raises a broader concern: schools should consider the extent to which particular tools encourage “a complete cognitive offload of the lesson planning task”, and what that does to professional expertise over time.
That is the same question schools ask about pupils. A tool that does all the thinking can produce a good result while the person using it learns nothing. For teachers, planning is part of how subject knowledge deepens. The aim is to save time on the routine parts while keeping the judgement.
What to take into school
Target, do not mandate. Offer AI planning tools first to the teachers most likely to benefit: those new to a year group, a subject or the profession.
Protect time to learn. A short, practical session on a real upcoming lesson will do more than an email with a login.
Keep the teacher's decisions visible. Ask staff who use AI planning tools what they changed and why. That conversation keeps the professional judgement in the process, and it is good modelling for the AI literacy we want pupils to have.
The trial measured teacher workload and resource quality, not pupil outcomes. As evidence goes, it is a useful step: AI can save planning time, but how it is used decides whether that time is well spent.
Source: Education Endowment Foundation, Aila Teacher Choices trial, evaluation report uploaded 5 October 2026. Evaluated by NFER with data from 416 teachers in 108 primary schools in England. The trial did not measure pupil attainment.