EEF trial finds AI lesson planner Aila saves teachers 49 minutes a week: the gap now is adoption, not evidence

The Education Endowment Foundation (EEF) has published the results of a randomised controlled trial of Aila, the free AI lesson assistant built by the publicly funded Oak National Academy. The report, dated 5 October 2026 and covered by Tes and Schools Week on 6 October, was produced by the National Foundation for Educational Research (NFER), which ran the trial in 108 primary schools in England in the autumn term of 2025 and analysed data from 416 Key Stage 2 teachers. Teachers given Aila spent 2.5 hours a week preparing lessons and resources, against 3.3 hours for those who did not: a saving of around 49 minutes a week, or 24 per cent. Independent assessors found the Aila group's resources were no worse than the control group's. The EEF gave the finding its high security rating.

The headline number answers the question leaders have asked for three years: can AI cut teacher workload without lowering the quality of what reaches pupils? On this evidence, yes. But the detail matters as much. Some 81 per cent of teachers in the Aila group opened the tool at least once, yet in a typical week it was used for only around 7 per cent of lessons, and use fell over the ten weeks. Fewer than one in five said it meaningfully changed how they taught or planned. The report describes a disconnect between a tool designed to plan lessons from scratch and teachers who mostly adapt material they already have, and notes that outputs often needed adapting before they were classroom-ready. Newer teachers, and those less confident in their subject knowledge, gained the most.

Dan Fitzpatrick, The AI Educator, says the trial settles one argument and opens another. "The evidence question is now answered: time saved, quality held. The question that remains is adoption, and that is a human question, not a technical one." He points to the control group, who were free to use other AI tools and in many cases did, as the most telling detail. Teachers are already ahead of the tools being built for them. The 7 per cent figure is not a failure of teachers; it is information about how they work.

The finding that outputs needed editing before use is, in his view, the point rather than a flaw. The teacher adapts, judges and decides; the tool drafts. That is what "outsource the doing, not the thinking" looks like in a primary school, and it is why the quality held. The risk for leaders now is not moving too fast. It is reading low usage as a reason to wait, when the trial shows the teachers who gain most are the ones who need support first.

What school leaders should do now: first, find out how your teachers actually plan, from scratch or by adapting shared resources, and choose or configure tools to match that. Second, point AI support first at early career teachers and staff teaching outside their specialism, where the trial found the largest gains. Third, write the "adapt before you use it" step into your AI guidance so professional judgement is named as the quality control. Fourth, measure what the EEF measured, time and quality, rather than counting logins. Employers rolling out AI face the same lesson: providing a tool is not the same as changing how people work.

Dan Fitzpatrick works with schools, trusts and education bodies on how teachers and leaders put AI to use in practice, and writes and speaks internationally on AI, learning and the future of work.

Dan Fitzpatrick is available for interview and comment on this story.