The DfE is moving ahead with trials of personalised AI tutoring tools for disadvantaged pupils. Here is what the evidence says and what schools need to know before applying.
The Department for Education's decision to invite bids from AI tutoring pilot partners in 2026 represents one of the most significant shifts in English education technology policy for a generation. For the first time, the government is directly funding the development and trialling of personalised AI tutoring tools — and the explicit focus on disadvantaged pupils makes this not just a technology question but a social justice one.
If AI tutoring works — and the early evidence from international pilots suggests it can — it could democratise access to high-quality personalised learning that has historically been available only to children whose families can afford private tuition. That would be genuinely transformative.
The DfE has invited bids from EdTech providers, schools and research organisations to participate in a structured pilot of AI tutoring tools specifically designed for disadvantaged pupils in key stage 3 and key stage 4. Priority subjects include English, mathematics and science — the core GCSE subjects where the disadvantage gap is most pronounced and most consequential.
Pilot partners will be expected to provide robust evaluation data, work within existing school timetables and demonstrate clear mechanisms for teacher oversight of AI tutoring interactions. The DfE has been explicit that AI tutoring is intended to supplement rather than replace teacher-led instruction — a position supported by the available evidence on what makes AI tutoring effective.
The evidence base for AI tutoring is growing rapidly, though it remains uneven. The most rigorous studies — including those from the Education Endowment Foundation and the Rand Corporation — consistently find that AI tutoring tools are most effective when they provide immediate, specific feedback on student responses, adapt the difficulty of tasks in real time, maintain detailed records of student progress that teachers can access and review, and are embedded within a broader pedagogical framework rather than used in isolation (EEF, 2025).
The evidence also consistently shows that the quality of implementation matters as much as the quality of the tool. AI tutoring in schools where teachers are trained, engaged and able to integrate the data into their planning produces significantly better outcomes than the same tool used in a more disconnected way.
The focus on disadvantaged pupils is the right instinct. The disadvantage gap in English education — the difference in outcomes between pupils eligible for free school meals and their better-off peers — has stubbornly resisted decades of policy intervention and remained essentially unchanged since 2017 (EPI, 2025).
If AI tutoring can provide disadvantaged students with the kind of personalised, patient, responsive learning support that their better-off peers access through private tuition, it could make a meaningful dent in this gap. But only if the implementation is right — and only if it does not become a substitute for the human relationships and teacher quality that remain the most powerful drivers of educational equity.
At Academica Mentoring we welcome this development. We have always believed that every child deserves access to the kind of personalised support that transforms learning outcomes. Whether that comes through human tutoring, AI tools or — ideally — a thoughtful combination of both, the goal is the same: ensuring that a child's educational outcomes are not determined by their postcode or their parents' income.