Every learning app and platform claims to work. Scroll through any app store listing or supplier pitch and the language is remarkably consistent: personalised, evidence-based, proven to boost results. Very few of these claims come with anything a parent or a school leader could actually check. This is a practical checklist for closing that gap, built around what independent evidence, UK data protection law, and the Department for Education's own guidance actually require, not what a product's marketing page says.

The criteria below apply whether you're a parent deciding whether to subscribe to a learning app for your child, or a school leader evaluating a platform for procurement. Where the requirements differ, particularly around AI-specific safety standards that apply formally to schools, that's flagged clearly.

Start with evidence, not marketing claims

The single most useful question to ask any EdTech provider is also the one most avoid answering directly: what is the actual evidence that this works, and who checked it? There's a meaningful difference between three tiers of evidence, and knowing which tier you're being shown matters more than the headline claim itself.

Independent, peer-reviewed or third-party evaluated evidence sits at the top: a study conducted or verified by a research body with no financial stake in the product's success, ideally with a real control group and a sample size large enough to mean something (dozens of pupils, not three case studies).

Supplier-commissioned but methodologically transparent studies sit in the middle: research the company paid for, but conducted with a clear, checkable methodology and a real comparison group, disclosed as supplier-funded rather than presented as independent.

Marketing claims and testimonials sit at the bottom: individual success stories, before-and-after screenshots, or vague statements like ‘proven to boost results’ with no study, sample size, or comparison group behind them at all. This is the tier most EdTech marketing operates in, and it's the easiest to produce, which is exactly why it's worth being sceptical of on its own.

A provider confident in its evidence will tell you, unprompted, which tier its claims sit in. A provider that changes the subject or points only to testimonials when asked directly is telling you something too.

For schools: the DfE's product safety standards

If a tool your school is considering uses generative AI in any form, speaks to pupils directly, generates explanations, or produces open-ended responses, it now falls under a formal government benchmark, not just informal good practice. The Department for Education published Generative AI Product Safety Standards on 19 January 2026, thirteen specific requirements a product must meet to be considered safe for use with pupils. This replaced the earlier, less formal guidance and expectations published in 2025, and it widened what's covered well beyond the basics of content filtering.

The thirteen standards group into several practical clusters worth understanding, rather than treating as a single undifferentiated checklist:

Purpose and evidence. A product should state clearly what it's for, which ages it's built for, and back its claims with real evidence, not marketing language dressed up as data.

Content safety. Harmful or inappropriate content should be prevented from arising in the first place through how the product is built, not filtered out afterward as an add-on. This is the standard most worth probing directly: a closed-loop product, one that can only draw on pre-approved material rather than generating open-ended responses, removes entire categories of risk that a general-purpose chatbot has to manage after the fact.

Data protection and security. Clear answers on whether pupil data is ever used to train a model, where data is hosted (UK or EU hosting is the standard to look for), what security certification the supplier holds, and a genuine data protection impact assessment you can review, not just a policy document that asserts compliance.

Governance and accountability. A named person responsible for safety, published policies, and a real complaints process, not a support email that goes unanswered.

Pupil wellbeing. This is where the January 2026 update went furthest beyond earlier guidance: standards now explicitly cover whether a product scaffolds thinking rather than handing over answers (protecting against what the DfE calls cognitive deskilling), whether it presents itself as a tool rather than a companion a child might form an emotional attachment to, whether it can recognise and appropriately escalate signs of pupil distress, and whether it avoids manipulative design, flattery, guilt, streaks and engagement-maximising tricks aimed at children.

For parents: what to check before you buy or sign up

Parents don't carry the same formal procurement obligations schools do, but the underlying questions are largely the same, just asked more informally.

Ask what evidence exists, and check it yourself. A quick search for the company name plus ‘study’ or ‘evaluation’ takes a few minutes and tells you whether independent evidence exists at all, or whether every result you find traces back to the company's own marketing.

Read the actual privacy policy, not just the summary. Specifically check what happens to your child's data if you cancel, whether it's shared with third parties, and whether the company can use your child's interactions to train or improve an AI model. Under UK GDPR and the ICO's Children's Code, a company should be able to answer this in plain, specific language, not buried in generic legal text.

Check whether progress is visible to you, not just to the platform. A tool that shows you, concretely, what your child has and hasn't mastered is fundamentally more useful, and more accountable, than one that simply reports time spent or a vague progress percentage.

Notice whether it's designed to be put down. Streaks, daily notification nudges, and rewards that maximise time in the app rather than time learning are a design choice, not a neutral feature. The DfE's manipulation standard exists because this pattern specifically targets children's still-developing self-regulation, and it's worth noticing in a consumer app even where no formal standard applies.

Check curriculum alignment. A generic ‘educational’ app is a different thing from one mapped to the UK national curriculum and the specific exam board specifications (AQA, Edexcel, OCR, WJEC) your child's school actually uses. Ask directly which curriculum and exam boards a product is built around before assuming it matches what your child needs.

Questions to ask any provider directly

• What specific evidence shows this works, and who conducted or verified it?

• Can this product generate open-ended content, or is it limited to pre-approved material?

• Is my child's or my pupils' data ever used to train an AI model, and where is data hosted?

• What happens to the data if we stop using the product?

• How does the product respond if a child shows signs of distress, and who gets notified?

• Is progress visible to a parent or teacher, in specific, checkable detail?

• What curriculum and exam boards is this built around?

• Who is accountable for safety at your organisation, and how do we raise a concern?

Red flags worth walking away from

Certain patterns in a provider's answers are worth treating as a stop signal rather than a minor gap to work around. Vague or evasive answers about evidence, especially a pivot to testimonials when asked for a study, is one. An unwillingness to say plainly whether pupil or child data trains an AI model is another; this is a direct, answerable question, and hesitation around it is itself informative. A support or safeguarding contact that goes unanswered during evaluation, before you've committed to anything, is a reasonable preview of what happens after you have. And design that leans on streaks, urgency, or guilt-based messaging to keep a child using the product is worth treating as a genuine warning sign, not a minor UX quirk, given what the DfE's own standards now say about manipulation aimed at children specifically.

BrainStrata was built against this exact bar, not as an afterthought bolted on for compliance, but as the starting design brief: adaptive practice that scaffolds toward an answer rather than handing one over, progress that's genuinely visible to parents and teachers rather than hidden behind a vague score, UK-hosted data that is never used to train models on children's work, and a clear, named path for any safeguarding concern. Every question in this checklist is one we'd expect to be asked, and expect to answer plainly.

Sources and further reading

[1] Department for Education. Generative AI: Product Safety Standards (published 19 January 2026). The thirteen standards covering purpose, filtering, data protection, security, governance and pupil wellbeing for generative AI products used in education.

[2] Department for Education. Generative artificial intelligence (AI) in education, policy paper (2025), and Keeping Children Safe in Education 2025, the underlying statutory safeguarding guidance the AI standards connect to.

[3] Information Commissioner's Office. UK GDPR guidance and the Children's Code (Age Appropriate Design Code), covering data protection standards for services likely to be accessed by children.

[4] Online Safety Act 2023. Statutory framework covering content moderation and age assurance duties relevant to online educational products accessible to children.

[5] EdTech Impact and similar independent review platforms, cited as examples of third-party evaluation sources schools and parents can cross-reference against supplier claims.