EEF trial finds Oak’s Aila cut KS2 planning time by 24%

Oak National Academy’s AI lesson-planning assistant, Aila, cut Key Stage 2 teachers’ lesson and resource preparation time by 24% on average in a school-randomized trial published Tuesday, October 6, without producing lower-rated materials than the comparison group. In practical terms, teachers assigned to use Aila spent about 2.5 hours a week on planning and resource prep, compared with 3.3 hours for teachers asked not to use Aila — a saving of roughly 48.6 minutes a week. (nfer.ac.uk)

For school leaders and teachers trying to separate real workload gains from AI hype, that makes this one of the clearest recent pieces of evidence on a teacher-facing generative AI tool. But the result is narrower than the headline may suggest. The trial measured teacher time and lesson-resource quality, not pupil attainment, and both groups were allowed to use other generative AI tools such as ChatGPT or Microsoft Copilot. That means the study estimates Aila’s added value over normal practice in 2025 England, not AI versus no AI. (educationendowmentfoundation.org.uk)

What the trial actually tested

The evaluation, commissioned by the Education Endowment Foundation and carried out by the National Foundation for Educational Research, followed 464 Key Stage 2 teachers in 108 schools over 10 weeks in autumn 2025. Schools, rather than individual teachers, were randomly assigned either to an Aila group or a non-Aila group. Researchers then compared planning time and resource quality using weekly teacher diaries, surveys, and independent teacher assessments of submitted materials. The EEF says the study received a “high” security rating, its label for findings it has high confidence in. (nfer.ac.uk)

That design matters. Teachers in the Aila group were encouraged to use the tool as much or as little as they wished for Year 3 to 6 planning, while teachers in the comparison group were asked not to access Oak National Academy’s website, including Aila, during the trial. But the comparison group was not locked out of AI altogether; it could keep using other sources and other GenAI tools as part of routine planning. Both groups were also asked not to pull new resources from Oak’s main website, though they could reuse Oak materials downloaded before July 2025. (nfer.ac.uk)

Aila itself is a free product from Oak National Academy, launched in September 2024 and still labeled beta. Oak says it helps teachers create and adapt lesson plans and resources step by step, generating editable lesson plans, slide decks, quizzes, worksheets, glossaries, and other materials from Oak’s curriculum-aligned content. The U.K. government’s algorithmic transparency record says Aila currently uses OpenAI’s GPT-4o together with retrieval over Oak’s own lesson corpus. (nfer.ac.uk)

That product design helps explain — though it does not prove — why the trial found a workload gain without an obvious quality penalty. Aila is not a blank chatbot. It constrains generation around Oak’s existing curriculum structure and keeps teachers editing throughout the process. In other words, the likely gain is not “AI replaces planning,” but “AI reduces some of the drafting and formatting burden while the teacher still supplies the professional judgment.” That is a more modest claim, but it is also more believable in day-to-day school work. (support.thenational.academy)

Why the time savings did not translate into heavy use

The most revealing finding in the report may be how little Aila was actually used. Teachers in the Aila group used it for an average of 7% of lessons, and both the frequency and amount of use fell over the 10-week trial. NFER and EEF both report mixed views: some teachers found the tool accessible, curriculum-aligned, and especially useful for unfamiliar topics, while others preferred adapting existing lesson plans or using other AI tools that felt quicker or easier. NFER says the pattern may reflect a mismatch between Aila’s design for building lessons from scratch and the way many teachers really plan, which often involves modifying something that already exists. (nfer.ac.uk)

That finding should temper any rush to treat Aila as a universal answer to workload. The trial’s own eligibility rules favored teachers who expected to create at least some resources from scratch and excluded teachers who were already regular Aila users. So the evidence is most relevant to schools where teachers still do a meaningful amount of original planning and are open to trying a specialist tool, not necessarily to schools working from tightly sequenced commercial schemes or to staff who have already settled on other AI workflows. That is an inference from the trial design, but it fits the implementation pattern the evaluators describe. (nfer.ac.uk)

What schools should — and shouldn’t — change now

The subgroup findings point to a practical adoption strategy. NFER found larger workload reductions among teachers who were newer to the profession, less confident in their subject knowledge, lower in pedagogical confidence, or more confident with technology. That suggests Aila may be most useful not as a blanket mandate for every classroom teacher, but as a targeted support for newer staff, for teachers planning outside their strongest subject, or for moments when a teacher has to build from scratch rather than adapt a familiar sequence. (nfer.ac.uk)

The EEF’s implementation advice also deserves more attention than the headline number. Its project page says schools should consider giving teachers dedicated time to learn a tool like Aila if they want efficient gains and sustained use. That aligns with the EEF’s earlier 2024 ChatGPT teacher trial, which also found time savings and a decline in use, but under different conditions: that study covered Year 7 and 8 science, gave teachers a five-week learning period, and compared ChatGPT use against a group told to avoid any generative AI. The Aila trial covered KS2 across subjects and compared Aila against normal practice that could still include other AI. So schools should be careful not to treat the percentage savings from the two studies as a simple league table. (educationendowmentfoundation.org.uk)

Taken together, though, the two EEF-backed trials do point in the same direction: teacher AI appears most promising when it is attached to a specific workflow, judged against a concrete outcome, and introduced with some support rather than as a vague invitation to “use AI.” What they do not yet show is whether those time savings last once novelty wears off, whether teachers shift that time into better instruction, or whether pupils learn more as a result. With Aila still free and in beta, and with Oak continuing to revise the product, the next evidence to watch is not another claim about minutes saved. It is whether adoption stabilizes and whether any of these tools can show benefits that reach beyond the planning desk and into classroom learning. (support.thenational.academy)