Seven questions to ask before an AI tool meets a child.

Every school leader we speak to has sat through the same demo: a polished AI tool, a confident salesperson, a lesson plan generated in eleven seconds. The demo is always good. The questions that matter come afterwards, and they are unglamorous. The Department for Education's product safety expectations for generative AI in education give you a framework; here is how we'd turn it into a conversation.

The seven

• Where does what a pupil types go? Is it stored, for how long, and is it used to train anything? Ask for the sentence in the contract, not the reassurance.
• Is there a filtering layer between the model and the child, and can we see how it fails? Every filter fails somewhere. A vendor who says theirs doesn't hasn't tested it.
• Can a teacher see what students did? Logs are a safeguarding feature, not surveillance, when used well.
• What happens to a student's account when they leave? Deletion should be real and documented.
• Who can we call when it goes wrong at 9.15 on a Tuesday? A named human, not a chatbot.
• What does it cost per pupil, next year, when the launch pricing ends?
• What would you teach a pupil to check? If the vendor can't answer this, they haven't thought about learning — only about output.

Free is not free

The tools that worry school leaders most, according to Third Space Learning's 2026 report on UK schools, are free ones — because the business model has to be something, and it is often the data. That is not a reason to avoid free tools; some of the best classroom AI tools are free and Chromebook-friendly. It is a reason to ask the first question above with particular care.

One last suggestion: put a student on the evaluation panel. They will find the workaround the vendor didn't mention within four minutes, and that is exactly the kind of judgement AI literacy is supposed to build.