At Digital Learning Week 2026, UNESCO and 25+ education ministers agreed on the principles that should govern AI in education: public accountability, deliberative governance, teacher agency, learner rights, safety. Useful words — but not something you can hold up against the app your child just downloaded. Here's that same thinking turned into ten questions you actually can ask.

Use this on anything: a subscription AI tutor, a free chatbot your child is using for homework help, a school-recommended tool. It's written so you can apply it to aitutors.me exactly the way you'd apply it to a competitor — we've tried to answer honestly where we can in the notes, without pretending every question resolves cleanly in our favour.

1. Does it ever hand over the final answer?

The single most important question. Watch a real session. Does the tool eventually just state the answer to the problem your child brought it — or does it stop one step short and make your child do the last bit of thinking? A tool that always gives the answer is doing the homework, not teaching. This is the one UNESCO's "learner rights" language is ultimately protecting: the right to actually learn, not just to get a correct-looking page of homework.

2. Can you see what happened in a session?

Not necessarily every word, but something. A tool with zero visibility — no session summary, no way to review a conversation — makes it impossible to exercise the same oversight a teacher has in a classroom. This is the "teacher agency" principle, applied at home: a parent needs the same visibility a teacher would have.

3. What happens to the data?

Ask three sub-questions: How long is it kept? Is it used to train a model that isn't yours? Is it sold or shared with third parties for anything beyond running the service? A vague privacy policy, or one that reserves broad rights to "improve our products and services" without defining what that means, deserves a second look.

4. Does it know when to say "I'm not sure"?

An AI tool that's always confident, on everything, is more dangerous than one that occasionally defers. Ask what it does when a question is outside its subject, or when the honest answer is "this needs a person" — a safeguarding concern, a question a subject teacher should really answer, a topic it genuinely doesn't have good information on.

5. Is there a hard rule for safeguarding indicators?

If a child expresses something concerning in a session — not just "I'm stressed about this test" but something that suggests real distress — does the tool have a fixed, non-negotiable response (pointing to a trusted adult, a helpline like Childline on 0800 1111), or does it just carry on with the tutoring conversation as if nothing happened?

6. Who decided the pedagogy, and can you see the reasoning?

"Personalised" is doing a lot of work in most AI-education marketing. Ask what it's actually personalising: pacing? difficulty? subject choice? tone? A tool that can explain, in plain terms, how it adapts is more trustworthy than one that just claims to.

7. Does the cost structure create a bad incentive?

Free tools are often monetised through data or advertising rather than subscription — worth knowing which. Paid tools should be transparent about what you're actually paying for (a subscription per child? per subject?) rather than a structure designed to be confusing.

8. Is there a human path when the AI gets something wrong?

Every AI tool will occasionally be wrong, confusing, or unhelpful. Is there a way to report that, or a human review path — or is the system a closed box with no feedback loop?

9. Does it respect your child's energy, or just their time?

A tool that pushes for maximum engagement — more sessions, longer streaks, notification nudges — is optimising for something other than your child's wellbeing. One that's honest about rest, that doesn't punish a quiet week, is a healthier design, and closer to what "public-interest" actually implies in practice.

10. Would you be comfortable reading a full transcript?

The simplest gut-check of all. If the idea of reading exactly what your child said to the tool, and exactly what it said back, makes you uneasy — that's information. A tool that would pass that read is usually the one doing right by the other nine questions too.

Using this checklist honestly

No product will score a clean ten, including ours — and a company grading its own homework is exactly the reason we tried to write our own honest comparison with specifics rather than adjectives. Use the checklist to notice patterns, not to find a perfect answer: a tool that fails on the final-answer question (#1), the data question (#3), and the safeguarding question (#5) together is worth real scrutiny, however polished its marketing looks.

FAQ

Is this checklist official UNESCO guidance?

No. It's our own translation of the principles UNESCO discussed at Digital Learning Week 2026 (public accountability, deliberative governance, teacher agency, learner rights, safety) into ten practical questions a parent can actually ask about a specific product. UNESCO did not publish this checklist itself.

Do I need to check all ten questions before letting my child use an AI tool?

No single question is disqualifying on its own, and few products will score perfectly. Treat it as a way to notice patterns — a tool that fails three or four of these, especially around data, answers, and safeguarding, is worth more scrutiny than one that only misses a minor point.

What's the single most important question on this list?

Whether it gives your child the final answer outright, or works with them until they get there themselves. Everything else — data handling, safeguarding, transparency — matters, but this one determines whether the tool is actually teaching or just doing the homework.

Duke Harewood built aitutors.me for his own KS3 daughter, and wrote this checklist so it works exactly as well pointed at a competitor's product as at his own. Published 11 September 2026.