Google and Pearson Will Build England's AI Tutors. Evidence Is Thin

The Department for Education has awarded contracts to eight companies, including Google and exam board Pearson, to co-design AI tutoring tools under its AI Tutoring Pioneers Programme. Each successful bidder receives an equal share of a £2.88 million pot, according to Schools Week. The tools are meant to reach up to 450,000 disadvantaged pupils in Years 9 to 11 by the end of 2027, starting with English, maths, science and modern foreign languages.

This sits inside a wider £49 million package for AI in education that Education Secretary Bridget Phillipson announced this year, which also includes £45 million for school connectivity and £3 million to build AI training datasets. The tutoring pilot itself has a separate £23 million commitment.

What's actually being built

DfE and the government's AI incubator, i.AI, are working with the eight pioneer firms to co-design tools with teachers rather than simply buying off-the-shelf products. The pitch is personalisation: software that adapts to where a pupil is stuck, offers extra practice, and identifies gaps against the national curriculum, working alongside rather than instead of classroom teaching.

Tools must meet the DfE's Generative AI Product Safety Standards before rollout. The department says pupil work will not be used to train AI models without parental permission, and no identifiable pupil data will be shared publicly. Testing under teacher supervision is due to begin this autumn.

The target group is specific: pupils eligible for free school meals, on the theory that private tutoring is already available to families who can pay for it, and that AI could close some of that gap at scale. Roughly 450,000 pupils nationally fall into the eligible cohort across those three year groups.

The evidence gap ministers have already admitted to

The government's own language undercuts the confidence of the announcement. Ministers have acknowledged the evidence for AI tutoring tools is "limited in quantity, scope and evidence base," according to Tes. Researchers cited in Schools Week draw a sharp distinction between a strong evidence base for what effective human tutoring looks like, built up over more than a decade of National Tutoring Programme data, and an AI evidence base that is, in their words, still in its infancy.

That matters because this pilot is not a small trial. It is a procurement decision covering hundreds of thousands of pupils, running in parallel with the evaluation that is meant to establish whether the approach works at all.

No statutory requirement exists for any product to demonstrate in advance that it is safe, that it protects children's data, or that it produces a measurable educational benefit over the non-digital alternative.

That gap has featured in written evidence to the parliamentary committee scrutinising AI in education, and it echoes concerns already raised about other English AI procurement this year: policy and funding moving ahead of proof.

Teachers are not convinced

The National Education Union surveyed its members in English state schools on the tutoring plan. Only 14 per cent supported it; 49 per cent disagreed. Objections raised in that consultation, reported by Schools Week, ranged from the practical to the political: that AI tutoring risks weakening the pupil-teacher relationship precisely for the pupils who most need consistent adult contact, that disadvantaged pupils in particular need in-person support rather than a screen, and that some saw the programme as a way to avoid funding schools properly rather than a genuine intervention.

There are also accessibility concerns that rarely make it into ministerial press releases: pupils with English as an additional language and those with speech recognition difficulties may struggle with voice-based or text-heavy AI interfaces, and pupils with behavioural needs typically require closer adult presence than a chatbot can offer.

Why this differs from the Boston or Alpha School models

Unlike US districts betting on AI tutors as a substitute for classroom time, the DfE has been explicit that these tools are meant to supplement, not replace, face-to-face teaching. That framing matters for how schools should treat the rollout: as an additional resource with a specific target group, not a staffing solution.

It is also narrower than some coverage suggests. The current phase is co-design and testing with pioneer schools, not a mandate for every secondary in England to deploy AI tutors this term.

What to do

Schools not among the pioneer test sites should treat fanfare around this announcement with proportion. There is no requirement to adopt anything yet, and the safety standards and evaluation results that would justify wider purchasing are not yet published. If your school is invited to participate in testing, ask specifically what outcome measures will be used, who owns the data, and whether results will be published regardless of whether they are favourable.

What to watch

Watch for the first published evaluation data from the pioneer testing phase, expected as tools move from co-design into supervised trials this autumn. Watch too for whether Ofsted or the DfE sets any minimum evidence bar before the 2027 rollout target, given ministers have already conceded the evidence base is thin. If the pattern from Maryland's evidence law is any guide, procurement without proof tends to attract scrutiny sooner rather than later.

Sources: