Can you trust an AI tutor? Five checks for parents and schools

    11 November 2026
    8 min read
    Listen to this post
    Can you trust an AI tutor?  Five checks for parents and schools

    Imagine your child is doing a maths quiz. The AI asks which fraction is equal to ½. Two of the choices are 2/4 and 3/6. Your child picks 3/6 and is marked wrong. The quiz expected 2/4.


    Both answers are correct. The question is the problem. But if your child is already unsure about fractions, they may come away thinking their reasoning was wrong.


    So, can you trust an AI tutor? Look for evidence of five things. It should teach the right material for your child, explain it accurately, help them think it through, ask fair questions and handle mistakes openly. Here’s what each of those checks looks like in practice.


    1. Curriculum relevance - does it teach what your child is learning?


    Curriculum-grounded AI uses relevant teaching material to inform its responses. For secondary learners in England, that means knowing what belongs at Key Stage 3 (KS3), what’s expected at GCSE and, where it matters, which exam-board specification a pupil is following. It also means knowing when to go back a step.


    SideCog is built by experienced teachers for 11–16-year-olds studying KS3 and GCSE subjects. We draw on Oak National Academy’s curriculum structure and resources where relevant. Oak describes how its subject leads review lesson material for curriculum, pedagogy and content. The sequence matters to us because it helps us identify the earlier ideas a learner may need to revisit. Why SideCog draws on Oak’s curriculum


    Say a Year 10 pupil gets stuck on a chemistry calculation involving moles. SideCog can go back and check an earlier skill, such as whether the pupil can scale a simple ratio from 2:1 to 6:3. If they struggle with the ratio, Cogsy can teach that first. If they can scale it, Cogsy can ask what the numbers in the balanced equation tell them about the amounts that react. Those two answers point to different places to start teaching. 


    Which curriculum materials does the tool use, and how does it choose what fits my child’s level and course?


    2. Factual accuracy - how are explanations checked?


    Even with a good source in front of it, generative AI can add an unsupported detail or contradict the material it was given. Researchers documented both problems in RAGTruth, a study of AI systems that retrieve source material before generating a response. An explanation can read fluently while including a mistake a pupil has little reason to spot. 


    Grounding gives the AI relevant material to work from. Checking whether the output agrees with that material is a separate job. A quotation can be compared with the original text, a calculation can be checked independently, and a subject teacher can review whether a science explanation makes sense at the level being taught.


    Can the provider show what gets checked, who checks it and what happens when a check fails?


    3. Teaching quality - can your child use the idea on their own?


    In a randomised study involving nearly 1,000 pupils at a Turkish high school, students using a standard GPT-4 chat interface did better during maths practice but scored lower on a later test taken without AI than pupils who had practised without it. With a version designed with teacher input to guide pupils using hints, that drop largely disappeared. The study found no clear improvement on the unaided test for that version. The result illustrates why it’s worth looking at what pupils can do after the help is taken away.


    That question shapes how we design Cogsy, SideCog’s learning companion. If a pupil says 1/8 is bigger than 1/4, Cogsy can ask how they decided. If the pupil says eight is bigger than four, Cogsy can ask them to picture two identical pizzas. Cut one into four equal slices and the other into eight. Which pizza gives you the bigger slice? Cogsy can then ask the pupil to compare 1/5 and 1/7 and explain their choice. The follow-up gives them a chance to use the idea in a fresh example.  How SideCog diagnoses gaps in understanding


    Can your child explain how they’d tackle a similar question with one thing changed? Try it together after a session.


    4. Question validity - does the quiz measure what it says it measures?


    A question can contain correct facts and still be poorly designed. In the opening fractions example, the marking would misread the pupil’s understanding. A better single-answer question would say, “Which fraction is equal to ½ and has a denominator of 6?” If the aim is to see whether the pupil understands equivalent fractions, a stronger follow-up is, “How could you show that 2/4 and 3/6 have the same value?”


    Researchers reviewing AI-generated multiple-choice questions in computer science and medicine found problems including ambiguous answer keys and implausible wrong options. The subjects differ from school maths, but the same review questions apply. Is the wording clear? Is there exactly one defensible answer when the question asks for one? Would the pupil’s answer tell you what they understand? 


    SideCog’s  multiple-choice questions use wrong options that reflect common misunderstandings. Take a KS3 biology question about whether plants respire while they’re photosynthesising. A pupil who chooses “No, plants photosynthesise instead” may think that one process takes the place of the other. That gives Cogsy a reason to ask what plants use respiration for. It can then explain that photosynthesis makes glucose and respiration releases energy from glucose, and that both processes can happen at the same time.


    Can the provider show a sample question, its answer key and the feedback a pupil gets for different responses?


    5. Safety and transparency - what happens when the tool is unsure or makes a mistake?


    An AI learning tool should be able to say when it’s unsure. Families and schools should be able to find out what the tool is designed to do and how to flag a questionable response. The Department for Education’s generative AI product safety standards for England call for testing with a realistic range of users and use cases, and further safety testing before new versions or models are released. Department for Education: generative AI product safety standards


    Cogsy works in a closed learning environment without access to the open web. SideCog’s parent dashboard gives families a view of what their child has understood and which gaps still need attention. These are features a family can see. For schools evaluating the quality of the AI’s responses, useful evidence would include teacher-reviewed examples across subjects, how the provider tested them and what it did with any errors it found. Explore the SideCog parent dashboard


    If a pupil thinks the AI is wrong, how can they flag it, who reviews it and how is the fix checked?


    Does curriculum-grounded mean the AI was trained on the curriculum?


    Grounding means giving the AI relevant curriculum material to draw on when it responds. Training or fine-tuning changes the AI model itself.


    SideCog combines KS3 and GCSE curriculum content with teaching principles and more than 50 years of combined classroom experience. For example, Bloom’s taxonomy informs how we help pupils move from remembering a fact to explaining it and applying it in a new situation. Rosenshine’s Principles of Instruction inform how Cogsy introduces ideas in small steps, models worked examples and checks understanding as pupils go.


    Growth mindset principles shape how Cogsy gives feedback. When a pupil gets stuck, Cogsy recognises what they’ve understood so far and offers a specific hint or a different strategy to try.


    Cogsy uses AI to personalise the next step based on what a pupil says and does. If they can remember a rule but struggle to use it, Cogsy can guide them through a worked example and then give them a fresh question to try on their own. That gives the pupil a chance to show what they can do independently.


    See how SideCog teaches


    Pick a KS3 or GCSE topic your child finds tricky. See how Cogsy works out where they’re stuck, guides them through it and gives them a fresh question to try on their own.

    Start your free 7-day trial.




    Source references

    • Education Endowment Foundation. Feedback. Teaching and Learning Toolkit. EEF


    J

    Written by

    Julian Dance

    Rate this article
    Was this helpful?

    Sign in for free to react to this article.

    • Can an AI Tutor Replace Private 1:1 GCSE Tuition?

      Could an AI tutor help your child with GCSE revision, or will it simply give them answers? Julian Dance explains what good tutoring actually does, where AI can help, where a human tutor still matters, and four questions parents should ask before choosing a learning tool.

      6 November 2026

    • AI Can Answer a Question, But Can It Teach?

      AI can answer questions, but can it teach? Discover why SideCog puts pedagogy, adaptive teaching and lasting learning at the heart of educational AI.

      22 September 2026

    • Flashcards - Are You Revising or Just Copying?

      Flashcards are one of the most popular revision tools, but using them badly can create the illusion of learning. Julian explores how to make them work harder.

      10 August 2026

    • Romeo and Juliet, AI and Safeguarding - It's All About Context

      When Oak National Academy's CEO raised AI safeguarding in Romeo and Juliet at the House of Commons, it highlighted a challenge every education AI company must solve - understanding context. Here's how SideCog's learning companion, Cogsy, was designed to respond.

      14 July 2026

    Browse all SideCog articles