The best tool makes the child think before it explains

Best AI Maths Tutors for Children in the UK: 5 Compared

“AI maths tutor” now describes at least three different products: a sequenced adaptive course, school-set personalised homework and a generative system that writes a new explanation. A camera solver that reveals working is different again. The right choice depends on the child’s age, curriculum, exact difficulty and whether the software teaches before it tells.

This comparison uses current official product information, UK prices and user-rating surfaces, plus the June 2026 independent randomised trial of Maths-Whizz. Provider claims are reported as claims. The verdicts turn on curriculum fit, teaching behaviour, independent evidence, value, privacy and the likelihood that understanding transfers away from the screen.

The best maths tools by job

First decide which type of help is missing

Five current maths tools compared in depth

1. Maths-Whizz: best evidence, highest home price

Best for: primary-age children who need a broad adaptive learning path and whose family can support several short sessions a week. The current home route begins with a 30–60 minute assessment, then selects lessons across maths topics. The provider recommends roughly 30–90 minutes a week, broken into 10–20 minute sessions.

Current price: £19.99/month or £149.99/year for one child, with additional-child prices of £14.99/month or £99.99/year and a seven-day trial. The provider says the trial becomes a subscription unless cancelled within the period, so note the deadline before the assessment.

Evidence: the 2026 EEF/NFER efficacy trial involved 10,369 pupils in Years 2–5 across 63 English state schools. Pupils allocated Maths-Whizz made the equivalent of one additional month of progress on average. Pupils eligible for free school meals made two months’ additional progress, but that subgroup finding has more uncertainty. Security was rated high.

Important limitation: this was a one-year, whole-school implementation with staff training, termly reviews and expected use of 45–60 minutes a week. Average recorded use was about 32 minutes. A parent buying a login does not automatically reproduce the implementation. The current Trustpilot surface was also only 3.1/5 from 16 reviews, with half one-star—too small to overturn the trial, but enough to read recent reliability, usability and service complaints before paying annually.

Verdict: the most defensible choice when independent UK evidence is the priority, provided the child accepts the interface and the household can sustain the routine. Trial the teaching, not the provider’s stronger “guaranteed progress” marketing.

2. DoodleMaths: best-value paid UK primary practice

Best for: an EYFS, KS1 or KS2 child who benefits from a short daily routine and parent visibility. It uses placement questions to build a personalised sequence across UK primary topics and is designed around roughly ten minutes a day.

Current price: one child costs £7.99/month or £69.99/year; a family maths plan for up to five children is £12.99/month or £119.99/year. DoodleTables is included. The wider DoodleBundle costs more, so do not buy English and spelling access unless those jobs are needed.

User evidence: the UK iOS listing showed 4.1/5 from about 6,700 ratings. That is a useful large-sample usability signal, and recent reviews mention personalised questions and parent/teacher targets. It is not independent proof of the provider’s claim that confidence or ability improves within six weeks.

Verdict: the strongest paid value for ordinary UK-primary consolidation here. Choose it over Maths-Whizz when price and a lighter daily habit matter more than having the strongest available independent trial. If the child needs direct explanation of a misconception rather than practice selection, test that carefully during the trial.

3. Khan Academy Kids: best free early option

Best for: ages 2–8 who need broad, playful early number, shape and problem-solving activities without adverts or a subscription. The provider states it is 100% free and mixes maths with books, language and creative activities.

User evidence: its UK iOS listing showed 4.7/5 from about 5,500 ratings, the strongest star score in this comparison with a meaningful sample. That supports the experience and value.

Limitation: it is aligned to US early-learning standards, not sequenced to England’s national curriculum or a particular UK school method. Use it for broad practice and curiosity, not as evidence that a Reception or Year 2 curriculum gap has been covered. Older pupils should use the main Khan Academy resources selectively with adult or teacher guidance; the main iOS app carries a 13+ age rating.

Verdict: download this before paying for a general early-maths entertainment app. Move to a UK-aligned programme only when curriculum matching solves a real need.

4. Sparx Maths: best when the school already uses it

Best for: pupils aged 11–16 whose school sets Sparx homework. The teacher selects topics already taught; Sparx then builds a personalised set at the pupil’s current level and expects complete working, including bookwork codes.

Cost: it is a school product, not an additional home subscription. If the school uses it, the most useful parent action is to understand its weekly expectation, make a workable time and place, and use the teacher’s support route when the child is stuck.

Limitation: personalised does not mean optional or conversational. Some children find the completion requirement and repetition stressful. Do not give answers to get the percentage to 100; tell school when the time, question level or anxiety is disproportionate, because that is implementation data the teacher needs.

Verdict: use the school system consistently before adding a second platform with different notation and incentives. A private tool should address a separate diagnosed gap, not create parallel homework.

5. Photomath: best checker after an attempt, not a tutor replacement

Best for: an older pupil checking algebra, arithmetic or a worked method after trying on paper. The app can scan a printed or handwritten problem and return steps and alternative methods.

User evidence: the UK iOS listing showed 4.6/5 from about 22,000 ratings. Its useful core step-by-step explanations are free, with optional paid features. That enormous sample supports convenience and satisfaction, not educational transfer.

Risk: the camera removes nearly all friction between homework and a complete answer. A correct-looking solution can also use unfamiliar notation, skip the child’s misconception or misread the image. Photos may include names, school details or teacher comments.

Verdict: use an “attempt–scan–explain–close–retry” rule: complete a first attempt, scan only the problem, compare the first different step, close the app, then solve a near-twin problem unaided. If the child scans first, uninstalling it is more educational than adding another lecture.

What the Maths-Whizz trial really proves

The trial supports a causal claim for the programme as implemented in those schools: allocation improved standardised maths outcomes by about one month on average. Randomisation, a large sample and independent delivery make that far stronger than testimonials or before-and-after app scores.

It does not prove that every child improves, that a home subscription has the same effect, that “AI” caused the impact, or that it beats every alternative. Schools received implementation support, and the trial measured a programme rather than isolating each adaptive feature. The effect is promising rather than magical.

DoodleMaths, Khan Academy Kids and Photomath have much larger public rating samples than Maths-Whizz’s Trustpilot page. Those samples help answer whether many people like the apps; they do not measure attainment against a control group. Use each evidence source for the question it can answer.

The five-minute teach-before-tell test

1. Give it a real current topic. Use a problem the child has been taught, not a novelty puzzle.
2. Ask for a hint. A tutor should start with one useful question, representation or partial cue—not dump a full solution.
3. Enter a plausible wrong step. See whether it identifies the misconception or only says “try again”.
4. Ask for the school method. Test the actual language: number line, bar model, column exchange, ratio table or another method the class uses.
5. Close it and transfer. Give a similar problem on paper. If the child cannot start, the software completed a task rather than taught a skill.

The Department for Education’s January 2026 generative-AI safety standards say learner-facing products should normally reveal a full solution only after a genuine attempt and should track cognitive offloading. That is a much sharper product test than animated rewards or the word “intelligent”.

Compare features by the problem they solve

Child safety, privacy and generative AI

An adaptive question bank and a general chatbot do not create the same risk. A child-facing generative tool can produce novel mistakes or harmful content, invite personal disclosures and retain conversations. The DfE standards expect age-appropriate filtering, human safeguarding routes, minimal data, transparency, testing with children and safeguards against emotional dependence or manipulation.

• Use the stated age route. Do not bypass a 13+ account requirement with a false birthday.
• Keep personal material out. Remove names, school logos, teacher comments, faces, diagnoses and support-plan details from uploads.
• Check model training. Find whether prompts, photos, voice or children’s work are stored or used to improve a system and how to opt out or delete them.
• Avoid companion behaviour. A maths tool should not claim feelings, ask for secrets, prolong conversation or imply it replaces a trusted adult.
• Retain a report route. The child should know how to stop, flag a wrong or upsetting response and show it to an adult.

Use our online-safety guide to configure the device and account, and our parental-control comparison to avoid collecting more child data than the problem requires.

A one-week home trial that produces a real answer

1. Name one goal. “Explain equivalent fractions using a bar model” is testable; “get better at maths” is not.
2. Ask school about method. Note the language and examples currently used.
3. Record the starting point. Keep two or three unaided paper questions—not the app’s proprietary “maths age”.
4. Run four short sessions. Sit beside the first two. Watch for hint quality, errors, frustration, reward-chasing and copying.
5. Test transfer. At the end of the week, use fresh problems on paper and ask the child to explain one method.
6. Check the family cost. Include reminders, arguments and adult supervision, not just pounds per month.
7. Keep, switch or cancel. Keep only if the defined skill or confidence improved and the child can work without the tool.

When a human tutor or school is the better choice

Go back to the teacher, maths lead, SENCO or a suitably experienced tutor when gaps span several topics, the child’s school method and app repeatedly conflict, basic number sense remains fragile, anxiety is increasing, or the child needs someone to interpret language, attention, working memory and confidence together.

A human tutor is also the better comparison when the need is explanation and responsive dialogue rather than independent practice. Ask what qualifications and safeguarding arrangements apply, how progress will be measured and how the tutor will coordinate with school. A cheaper app that never reaches the misconception is not better value.

For broader learning choices, compare the stage-first approach in our UK reading programme guide and the principles in our personalised learning explainer.

Questions parents ask about AI maths tutors

Which AI maths tutor is best for UK primary children?

Maths-Whizz has the strongest independent UK attainment study; DoodleMaths offers better-value curriculum-aligned daily practice. Choose Maths-Whizz when evidence and a broad adaptive route justify the price, and DoodleMaths when a lighter, cheaper routine meets the need.

Is Maths-Whizz worth £19.99 a month?

It can be if the child uses it consistently and transfers learning to unaided work. Its 2026 school trial is encouraging, but a home trial must test the interface, teaching and routine. Do not buy annually merely because the provider converts the price to an hourly comparison.

Is DoodleMaths actually free?

The app is free to download, but its full personalised programme uses subscriptions. The current UK listing shows £7.99 monthly or £69.99 annual for one child, with family and bundle plans. School access may follow different terms.

Can Khan Academy Kids replace a UK maths programme?

No. It is excellent free broad early learning, but it is not sequenced to the English national curriculum. Use it for playful practice, then use school material or a UK-aligned resource for a named curriculum gap.

Is Photomath cheating?

Scanning first and submitting the result as the child’s work is cheating and teaches little. Using it after a real attempt to locate the first wrong step can be legitimate if school rules allow it. The difference is the sequence and whether the child retries unaided.

Is adaptive learning the same as generative AI?

No. Adaptive software can select from pre-written questions using performance data. Generative AI writes new content. A product may use both. Ask what is generated, how it is checked and what data drives the next task.

Can an AI tutor replace a private tutor?

It may replace some routine practice at lower cost. It does not automatically replace a good tutor’s ability to notice the exact misconception, emotional state, school context and language needed. Compare against the actual job, not the label “tutor”.

How much screen time should maths practice take?

Use the shortest dose that reaches the goal and leaves the child able to explain it. Provider targets range from roughly ten minutes a day to 45–60 minutes a week. School workload, age, attention and transfer matter more than maintaining a streak.

Independent evidence, current products, prices and safety standards

1. EEF: Maths-Whizz randomised trial findings and implementation.
2. NFER: independent evaluation summary for Maths-Whizz.
3. Maths-Whizz: current UK home prices and trial terms.
4. Maths-Whizz: assessment and weekly-use information.
5. Trustpilot: current Maths-Whizz public review surface.
6. UK Apple App Store: DoodleMaths prices, scope and ratings.
7. UK Apple App Store: Khan Academy Kids scope and ratings.
8. Sparx Maths: parent guide to personalised school homework.
9. Sparx Maths: current age range and school product model.
10. UK Apple App Store: Photomath features and ratings.
11. Department for Education: 2026 generative-AI product safety standards.
12. Department for Education: safe use of generative AI in education.
13. Department for Education: AI and school data protection.
14. Information Commissioner: Children’s Code design guidance.
15. EEF: using digital technology to improve learning.

KB

By Kristine Bowman and KidsCo Editorial Team

The comparison was updated with June 2026 independent attainment evidence, live UK pricing, UK app-rating samples and current child-facing AI standards.

KidsCo editorial standards