What Parents Should Ask Before Using an AI Tutor

Before your child opens an AI tutoring app, run it through six checks, in this order: does it own up when it gets things wrong; where does your child's data go; is it actually built for a young user; will it show its reasoning or just hand over a final answer; what happens the moment it hits a question it can't handle; and is there a real, verified adult standing behind it when that moment arrives. Used well, an AI tutor is a genuinely useful practice partner — patient, available at 9pm before a test, happy to explain the same method five different ways. Used carelessly, it's a fluent-sounding tool nobody has actually vetted. The difference isn't the AI. It's whether the service behind it can answer these six questions without flinching.

Marketing tells you nothing; the questions do

Nearly every AI tutoring product on the market is a wrapper around the same handful of large language models, so a smooth demo proves almost nothing. Two apps can produce an identical, articulate answer to the same maths question while differing completely in what happens when that answer is wrong, who can see your child's typed questions, and whether a real person is reachable when the bot runs out of road. Those are the questions worth your time, and none of them show up in a screenshot. Run through the six below before your child spends real hours with any tool — it takes less time than reading the app's own marketing page, and it tells you far more.

The Tutorwise view: a score you can check beats a bio you have to trust

This is the standard we hold ourselves to, and it's a fair standard to hold any AI tutoring service to as well. On Tutorwise, a human tutor doesn't get to write their own reputation. Their credibility is a number the platform computes from things that can actually be verified — a completed DBS check and confirmed identity, the qualifications they hold, the outcomes families report, and reviews left after real sessions. Each signal is weighted, the score recalculates as new evidence lands, and a parent looking at a profile sees that number rather than a paragraph of self-description. Nobody is asking you to take a tutor's word for it. Hold an AI tutor to the same bar: don't ask whether it sounds capable, ask what, specifically, you can go and check for yourself.

1. Accuracy — what does it do when it's wrong?

Every large language model will occasionally produce an answer that is fluent, confident and wrong — the industry term is hallucination — and the risk is highest precisely in the subjects a parent is least equipped to fact-check on the spot. So don't ask whether the tool can be wrong; assume it can be, and ask what it does about it instead. Does it cite where an answer came from? Does it invite your child to check the working against a textbook or the exam board's own specification, or does it present itself as the final word? UK government guidance on generative AI in education is explicit that responsibility for checking output sits with the person using the tool, not the tool itself. Treat any AI tutor the way you'd treat a study partner rather than an examiner: useful, worth listening to, and always worth checking. A well-built tool makes that checking easy; a weak one just asks you to trust it.

2. Data — where does my child's information actually go?

What your child types into an AI tutor — their mistakes, their questions, their pace of progress — is personal data, and some of it is sensitive by any reasonable definition. Four questions cut through most of the marketing copy: what exactly is collected, is any of it used to train the underlying model, how long is it retained, and does the service follow the rules the UK has specifically set for children online. The Information Commissioner's Office's Age Appropriate Design Code — usually called the Children's code — sets out how digital services should handle the data of under-18 users by default, not as an opt-in extra. A service worth using will answer these four questions in plain language, unprompted, somewhere you can find it. A service that goes vague, or routes you to a wall of legal text with no clear answer, has told you something worth acting on.

3. Safeguarding — was this actually built with a child in mind?

Ask three things: was the product designed for children specifically, does it have real content controls, and — the one that matters most — what happens the moment a conversation starts drifting toward human contact. On a properly built platform, any handoff to a human tutor stays inside the platform itself, where that tutor's verification status is visible on their profile, rather than a bot passing a child a phone number or steering them into a private chat somewhere the service can't see. Keeping Children Safe in Education, the statutory guidance UK schools operate under, treats unsupervised contact between a child and an unchecked adult as a core risk to design against — and that same logic holds for any tool your child uses from a bedroom at home, not just inside a classroom. This isn't about assuming the worst of every product. It's a design question with a clean answer either way: can the service see and check what happens next, or has it built a route it can't?

4. Explainability — can it show its working, or just its answer?

A final answer with no reasoning attached teaches a child nothing and makes it trivially easy to submit work they don't actually understand. Ask whether the tool walks through its steps, whether it explains why a method works and not only what the method is, and whether it's built to build understanding rather than just close tonight's homework as fast as possible. That's the real line between a study aid and a shortcut. The stronger AI tutors are built to model a worked example first and then hand your child a similar problem to attempt themselves — which is close to how a good human tutor actually teaches, rather than how a search engine answers a query. Our guide on using ChatGPT for GCSE revision safely covers the specifics of keeping that line intact.

5. Escalation — what happens when it runs out of road?

No AI tutor can go the full distance on every problem. It might be a piece of exam technique that needs a human eye, a child who's anxious rather than simply stuck, or a question that genuinely calls for judgement a model doesn't have. Ask, directly, what the fallback looks like. A weak product just keeps generating plausible-sounding text regardless of whether it's actually helping. A well-designed one recognises the edge of its own competence and hands off cleanly to a real person. How a service handles its own limits is one of the clearest signals of how seriously it takes the rest of the checklist — a product that's honest about where it stops is usually honest about the parts before that too.

6. Human backup — is there a checked adult behind the interface?

This is the question the other five ultimately point back to. An AI tutor is a tool; accountability for a child's learning and welfare sits with a person. Ask whether the service connects your child to verified humans, and — this is the part worth pressing on — exactly how "verified" is established. "We have tutors available" is a marketing line. A completed DBS check, confirmed identity, real evidenced qualifications, and a credibility score you can actually inspect is proof. It's the identical principle behind the Tutorwise credibility score: don't take a claim, take a check. An AI tutor that's upfront about being a tool, and that hands off smoothly to a verified adult the moment one is genuinely needed, is doing its job properly. For a longer look at exactly where the technology's limits sit, we've written a direct answer to whether an AI tutor can replace a human tutor.

A five-minute way to actually test this

You don't have to take any of this on trust either — most of it is checkable in one short session before your child ever uses the tool unsupervised. Ask it a question one year-group above where your child sits and see whether it flags the answer as uncertain or states it with total, unearned confidence. Ask it, directly, "who checks the humans on this platform, and how" — a service with a real answer will name the check (DBS, identity verification) rather than deflect. Try steering the conversation toward something personal or off-topic and see whether it redirects toward a supervised, on-platform path or simply follows along. And look for the fallback: does the interface ever say, in effect, "this needs a person," or does it just keep talking regardless of the question? Five minutes of this tells you more than any amount of marketing copy, and it costs nothing to run before you decide the tool is one you're comfortable leaving your child alone with.

Where Sage fits into this

Sage, our AI tutor, was built to sit inside this framework rather than argue its way around it. It supports GCSE maths practice, walks through problems step by step, and is available whenever your child is actually studying, not just during business hours. It's also designed to be straightforward about its own limits: it shows its working rather than asserting an answer, it doesn't present itself as a person, and it points toward verified human tutors when a child needs real teaching judgement or a conversation that genuinely should involve a checked adult. The AI covers the repetition; a verified human stays accountable for the learning. That's the pairing worth looking for in any AI tutoring product, ours included.

Using an AI tutor and a human tutor together

For most families the better setup isn't choosing between AI and a human tutor — it's using each for what it's actually good at. Hand the AI the high-volume, low-stakes work: extra practice sets, quick clarifications, revision drills, the last-minute "I don't get this one" moment before a test. Keep a verified human for the work that needs judgement: working out why the same mistake keeps recurring, exam strategy, confidence, and the pastoral side of teaching that no model should be trusted with. Set up this way, the AI extends what a good tutor can cover rather than competing with them, and your child gets more support without losing the accountability only a checked adult can provide. If you're still weighing the two directly, our guide on AI versus human tutoring breaks down which one your child actually needs right now.

The short version

A service that answers all six questions cleanly — honest about its own mistakes, careful and transparent with data, genuinely built for a young user, able to show its reasoning, upfront about its limits, and backed by real, checkable humans — has earned a place in your child's routine. One that dodges even one of them hasn't, no matter how polished the chat interface looks. Judge it on what you can verify, not on how confident it sounds.

If you'd like an AI tutor that's honest about what it is and backed by verified human tutors when your child needs one, meet Sage, our free AI tutor, and browse tutors on Tutorwise whose credibility you can check for yourself rather than take on faith.

More in this series

• Meet Sage: An AI Tutor Built on Verified Credibility
• Can an AI Tutor Replace a Human Tutor? An Honest Answer
• AI vs Human Tutoring: Which Does Your Child Need?
• How to Use ChatGPT for GCSE Revision Safely and Well