The UK Is Spending £4 Per Student on AI Tutors. A Human Tutor Costs £38 an Hour.

England just committed £1.8 million to build AI tutoring tools for 450,000 disadvantaged students. That's £4 per kid.

A private tutor costs £30 to £50 an hour.

Education Secretary Bridget Phillipson is betting AI can deliver what money never could — a personal tutor for every struggling child in the country.

The numbers that forced the decision

The 450,000 aren't random. They're Years 9 through 11 students on free school meals — the government's proxy for disadvantage. These kids are heading toward GCSEs, the exams that decide whether doors open or close.

England tried the human version already. The National Tutoring Programme spent hundreds of millions after COVID connecting students with real tutors. It reached millions of sessions by 2023-24. But it was expensive, hard to scale, and quality varied wildly.

Now the Department for Education wants a private sector partner for a three-year contract, with tools in schools by late 2027. Teachers start co-designing this summer. If it works, it goes national.

"AI tutoring tools could level the playing field for families who cannot afford private assistance," Phillipson said. That could is doing heavy lifting.

What the research actually says

The evidence is surprisingly strong.

A trial in Scientific Reports found AI-tutored students learned more in less time than those in traditional classrooms. They were more engaged too.

Kulik and Fletcher reviewed 50 studies and found intelligent tutoring systems can "match the success" of human tutoring. Not outperform. Match.

Brookings looked at four randomized trials in January 2026 and found "substantial learning gains across all studies" — plus better motivation and efficiency.

The Tutor CoPilot trial found the biggest gains among the least experienced tutors. The AI didn't replace the teacher. It made mediocre tutoring good.

That's the key for England's plan. The government isn't removing teachers. It's giving every classroom an invisible assistant that never sleeps, never tires, and adapts to each student's pace.

The £4 question

That £4 figure sounds absurd. It's also misleading.

The £1.8 million covers development — building, testing, iterating with teachers. Once built, each additional student costs almost nothing. That's software's promise. Build once, deploy millions of times.

Compare: at £38/hour (UK average), giving those 450,000 students just 10 hours of tutoring would cost £171 million. Twenty hours? £342 million. And you'd need to find enough tutors. You can't.

The National Tutoring Programme proved this. Rural areas got fewer tutors. Poor areas got less experienced ones. Kids who needed the most help got the weakest support.

AI doesn't have a supply problem. It scales instantly. The question isn't cost — it's quality.

What could go wrong

Plenty.

AI tutors can give wrong answers with total confidence. They can't read body language. They can't tell if a kid's struggling with fractions or struggling because their parents are splitting up.

Then there's the engineering challenge. Building AI that teaches the UK maths curriculum accurately, adapts to different learners, catches mistakes live, and doesn't hallucinate — that's hard. £1.8 million is a modest budget.

And dependency is real. Bastani et al. found in PNAS that students showed gains while using AI tutors but struggled when the tool was removed. The AI taught them to lean on the AI, not think for themselves.

The global race

England isn't alone in this bet.

Massachusetts partnered with Google last week to offer free AI training to all residents — the first US state to do so. Governor Healey wants to build the "applied AI capital" of America.

Japan launched a student-led nonprofit for low-cost online tutoring. Tokyo schools saw a surge in uptake.

Khan Academy's Khanmigo jumped from 40,000 to 700,000 students in one academic year. Sal Khan's been arguing AI could finally deliver Bloom's "two sigma" effect — the 1984 finding that one-on-one tutoring produces learning gains two standard deviations above classroom teaching.

Bloom's problem: one-on-one tutoring for every student was impossibly expensive.

AI might solve it. Or it might be an expensive way to discover that teaching is harder than pattern recognition.

What happens next

The pilot runs through next academic year. If results hold, nationwide rollout starts by 2027.

Contract closes in May. Teachers co-design this summer. First students could be on AI tutors by autumn.

450,000 kids are waiting. Most have never had a personal tutor. Most never will — unless this works.

The price of finding out: £4 per student. The cost of not trying: another generation where your postcode determines your grades.