Executive Summary

​

Artificial intelligence (AI) is already present in England’s schools. Teachers use it to mark work, plan lessons, prepare materials, support pupils with special educational needs, and streamline administration. Students use it for research, writing, summarising, revision, language drills, and mathematics practice. PISA 2025, the first international survey of students’ use of AI for schoolwork, found that a majority of 15-year-olds in most participating systems now use AI chatbots for their schoolwork: across OECD countries, only 14% of students had never or almost never done so, and about one in five reported almost daily use (OECD, 2026b). Yet using AI tools does not necessarily mean learning with or from them. In the same data, students who used AI frequently scored higher where their schools taught them to assess the quality of AI output, an opportunity that reaches advantaged students more often than disadvantaged ones, and therefore expanding access to AI tools is unlikely to produce benefits without parallel investment in guidance on appropriate and effective use (OECD, 2026b). Adoption in England has been rapid, but it has also been uneven and largely unsupported by evidence. The Shape the Future Leaders Coalition research programme was established to change that.

​

This report presents the findings of the coalition’s first research cohort: nine practitioner-led action research projects conducted over the 2025/26 school year by school and trust leaders from across England, in settings ranging from a preparatory school to a 48-school multi-academy trust, with pupils aged 4 to 16. The projects were designed separately, around different tools and questions, but read together they converge on six findings.

​

1. AI tools worked as part of pedagogical routines, not as standalone interventions. The same tool produced different results depending on the routines built around it, and gains concentrated wherever implementation was consistent and structured (Caterham Prep, All Saints Academy, Woodland Academy Trust, Beauchamp College, Stretford High, Canons High).

2. Independence and agency are different outcomes, and can move in opposite directions. Pupils can be equipped to work without a teacher while losing choices over their own learning, or gain choices without the ability to use them well (Caterham Prep, Warwick Schools Foundation, Stretford High).

3. Older students are warier of AI than younger ones, and the teacher relationship remains key. Where pupils could compare, most preferred teacher feedback, trusting the AI’s competence on bounded tasks while placing more confidence in the integrity of their teachers’ feedback (Thames Christian School, Warwick Schools Foundation, Canons High).

4. The reliability of AI marking is set by the instructions it is given, not by the model alone. Accuracy on short, structured items reached levels close to expert agreement only where marking guidance was carefully constructed, detailed, and refined where needed, and long-form marking needs further development of tools and processes before deployment (Ark Schools, Beauchamp College, Thames Christian School).

5. AI is moving assessment closer to teaching, and changing what assessment measures. Fast, automated feedback shortened the loop between evidence and reteaching from days to minutes (Ark Schools, Beauchamp College, Canons High), while digitisation changed pupils’ writing itself and interaction logs made the process of thinking visible for the first time; both shifts raise validity questions schools are only beginning to ask.

6. Delegating work to AI has costs as well as gains. Teachers saved substantial time and redirected it into teaching and better planning, but projects also recorded what delegation displaced, from a teacher’s contact with pupils’ work, to pupils’ focus on process over marks. Every successful project iterated its tool mid-course; few produced their best results as first deployed.

 

This full report is intended for three audiences.

 

For educators and school leaders, it offers practical insight into the opportunities and pitfalls of using AI in settings not unlike their own, with both recommendations and cautionary tales.

 

For policymakers, it provides grassroots evidence of the conditions schools need to adopt AI responsibly and equitably, including a specific gap in current guidance around bespoke tools built or commissioned by schools themselves.

 

For educational technology developers, the case studies are candid accounts of how AI tools survived, and sometimes thrived, amid the pressures of everyday school life, and of what schools now check for before they trust a product.

 

This report addresses Dimension 2 in EVR’s 4D AI Strategy Framework: iterative unit applications of AI use cases (Luckin and Bagshaw, 2026), and speaks predominantly through the first of our three lenses on AI in education: AI tools, with deliberate reaches into the second, increasing human intelligence, and the third, learning about AI (Luckin, 2018). Section 1.6 explains this positioning and the conclusion draws out the interconnections.

​

The chapters that follow set out the coalition’s methodology, present each project and its results, draw out the six cross-cutting themes in full, and close with implications for practice, policy, and product development.