A lesson begins before pupils enter the classroom. A teacher selects explanations, anticipates misunderstandings and adapts material to a particular class. Generative AI can help prepare that material. Whether the saved time improves working conditions depends on what happens to checking, professional discretion and the rest of the working day.

Sector profile: school teaching
The focus is teachers’ preparation work in primary and secondary education. It excludes university teaching and does not treat administrative staff, teaching assistants or educational technology suppliers as the same workforce.
European Union
Eurostat reports 5.30 million teachers in primary, lower secondary and upper secondary education in 2024 (ISCED levels 1–3). These teachers work across different national curricula and employment systems. The total establishes the scale of school teaching; it does not measure AI adoption or exposure to job loss.[1]
United States
The National Center for Education Statistics records 99,073 operating public schools in 2024–25 across the 50 states and Washington, DC. Its table covers prekindergarten through grade 13. A national teacher total is unavailable because Tennessee’s teacher data are missing. Teacher counts in this source use full-time equivalents.[2]
These indicators have different units and coverage and cannot be ranked against one another. The English trials below are evidence from England, outside the EU; they are not estimates for European teachers generally.
What changes in the work?
Drafting a worksheet and deciding whether it will help a struggling pupil are different activities. An assistant may accelerate the first while leaving the teacher responsible for accuracy, curriculum fit and accessibility. Checking and adaptation therefore belong inside an assessment of preparation time. A faster initial draft alone is an incomplete measure.
Two trials, bounded results
An independent NFER evaluation published in October 2026 examined Aila, an AI lesson-planning assistant. It randomised 108 English primary schools, involving 464 teachers of Years 3–6, during autumn 2025. The primary analysis included 416 teachers. Over ten weeks, the group encouraged to use Aila reported 24% less lesson and resource preparation time: adjusted weekly means were 2.51 and 3.32 hours. Both groups could use other generative AI tools.[3]
The estimated time ratio was 0.76, with a 95% confidence interval of 0.64–0.89. An independent resource review supported non-inferiority, but resource ratings do not establish classroom learning gains. Aila accounted for an average of 7% of lessons, use declined, and teachers’ views were mixed. Some valued its accessibility and curriculum alignment; others preferred existing resources or alternative tools. The unblinded trial relied on self-reported time. It did not establish effects on staffing, pay or long-term retention.
An earlier NFER trial, published in 2024, randomised 259 teachers in 68 English secondary schools. For specified Year 7–8 science preparation tasks, the ChatGPT group averaged 56.2 minutes per week against 81.5 minutes in the comparison group during the five measurement weeks: 31% less time. The primary analysis included 211 teachers. Reviewers found no evidence of a resource-quality difference, but submitted materials could be selective, and pupil attainment was not measured.[4] The two percentages concern different interventions and tasks; they are not a ranking of products.
A workforce survey: who receives guidance?
School arrangements matter alongside tool performance. A Walton Family Foundation–Gallup survey of 2,069 US public K–12 teachers, conducted from 9 February to 2 March 2026 through the probability-based RAND American Teacher Panel, found that only 18% reported formal AI guidance in at least one of ten work areas. Responses were weighted to national teacher and school demographics; the full-sample margin of sampling error was ±2.5 percentage points at 95% confidence.[5]

Figure 1: accessible data
| Guidance | Teachers (%) |
|---|---|
| Formal in at least one area | 18 |
| Informal only | 48 |
| None | 34 |
A separate Gallup survey in July–August 2026 covered 1,952 K–12 educators, including teachers and other classroom staff. Among AI users, 64% said it gave them more time for activities they considered most meaningful. This is a reported experience among users, not a measured causal effect or a percentage of all respondents.[6]
Who gains from the time saved?
Less preparation time could reduce evening work, create space for individual feedback or support collaboration. It could also be absorbed into additional assignments. Neither outcome follows automatically from the trials. Evaluating a school’s implementation requires observing total working time and the allocation of duties, as well as the time spent generating materials.
Professional development is another part of that calculation. Teachers need opportunities to recognise incorrect explanations, unsuitable examples and inaccessible tasks. Subscription access, training and checking all have costs. A school should distinguish a reduction in one task from a net reduction in workload, and a change in workload from a change in employment.
What does the ILO perspective add?
The UN Secretary-General’s High-Level Panel on the Teaching Profession, supported by a joint ILO–UNESCO secretariat, places technology within teacher-led education. Its 2024 recommendations emphasise professional autonomy, participation in technology decisions and protection of the human relationship in teaching. These are sector-specific recommendations, not a forecast of employment effects.[7]
After the plan, the work continues
The evidence supports a practical question: can teachers retain the benefit of faster preparation while maintaining educational judgement and time with pupils? Answering it requires following working conditions and learning over time. An employment conclusion needs employment evidence.
Evidence and method
This focused narrative review closes on 21 November 2026. Official sector statistics, two school-randomised trials, workforce surveys and professional recommendations answer different questions. No pooled effect is calculated. Trial findings are specific to the tested settings; survey associations and perceptions are not causal estimates. The Aila evaluation was commissioned by the Education Endowment Foundation and conducted by NFER; the 2024 trial was co-funded by the Foundation and the Hg Foundation. Gallup’s surveys were produced with the Walton Family Foundation.
References
- Who teaches in the EU’s schools? Eurostat. (2026, 5 October). Dataset: educ_uoe_perp01, 2024 observations.
- Table 2. Number of operating public elementary and secondary schools and districts, student membership, teachers, and pupil/teacher ratio, by state or jurisdiction: School year 2024–25. National Center for Education Statistics. (n.d.). Common Core of Data, final version 1a. Retrieved 21 November 2026.
- Lesson planning using AI lesson assistant, Aila: A Teacher Choices Trial. Roy, P., Poet, H., Staunton, R., Aston, K., Thomas, D., & Styles, B. (2026). NFER; Education Endowment Foundation.
- ChatGPT in lesson preparation: A Teacher Choices Trial. Evaluation report. Roy, P., Poet, H., Staunton, R., Aston, K., & Thomas, D. (2024). NFER; Education Endowment Foundation.
- Most teachers receive no formal guidance on AI use. Malek Ash, A., & Senseman, K. (2026, 26 May). Gallup. Survey methodology: Walton Family Foundation–Gallup, Teaching for tomorrow: Closing the expectations gap, p. 17.
- Teachers using AI say it frees up time for what matters most. Brenan, M. (2026, 14 October). Gallup.
- Transforming the teaching profession: Recommendations and summary of deliberations. United Nations Secretary-General’s High-Level Panel on the Teaching Profession. (2024). International Labour Organization. Recommendations 43–47.
Liked this post? Follow this blog to get more.