On this page
You don't need to understand how AI works to judge whether an AI tutor helps your child. You need twenty minutes, one real homework problem and a clear idea of what to watch. This guide gives you a session plan and a checklist, so you can sit beside your child and see for yourself.
What does research say about AI tutors so far?
Two recent studies point in different directions, and the difference is instructive.
In a field experiment with nearly 1,000 high school students in Turkey, Bastani and colleagues (2025) compared a standard GPT-4 chat interface with a version that had teacher-designed safeguards, mainly hints instead of answers. During practice sessions, students using the standard interface improved their grades by 48% compared with a control group, and those using the safeguarded tutor by 127%. But on the exam, taken without AI, the standard-interface group did 17% worse than the control group, while the safeguarded group showed no significant difference. The authors describe students using the plain chatbot as a "crutch".
In a randomised controlled trial with college students, Kestin and colleagues (2025) compared an AI tutor, designed around the same teaching practices as an in-class active-learning lesson, with that lesson itself. Students learned significantly more in less time with the AI tutor, and reported feeling more engaged and motivated.
The common thread: what matters is how the tool is designed and used. A tutor that holds back answers and makes the student think can help. A chatbot that hands over solutions can make practice look better while real learning gets worse.
Before you start (2 minutes)
Pick one problem your child is genuinely stuck on, from this week's homework or a textbook exercise. Not one they can already do.
Agree the rule: your child types; you watch. You can suggest a question, never the answer.
Check what the app asks for. Look at sign-up details and privacy settings. Our post on what parents should check about privacy in learning apps lists what to look for.
The 20-minute session
Minutes 0-5, ask for help, not an answer. Your child types the problem and adds: "I'm stuck. Don't give me the answer, help me figure it out." Watch the first reply closely.
Minutes 5-12, let it teach. Your child answers the tutor's questions honestly, including "I don't know". Watch whether the tutor adjusts to those answers or carries on with a script.
Minutes 12-16, test it with a mistake. Your child gives a deliberately wrong step. Does the tutor notice and explain why it's wrong, or accept it?
Minutes 16-20, check understanding without the tool. Close the app. Give your child a similar problem on paper with different numbers. Can they do it alone?
That last step is the one most families skip, and it is the one the Bastani study suggests matters most.
What to watch for: a checklist
Good signs
Asks what your child already knows before explaining.
Gives a hint or a question first, and a full solution only after your child has tried.
Notices the deliberate mistake and explains the reasoning behind the correction.
Uses a diagram or worked steps for a physics or geometry problem.
Checks understanding with a follow-up question.
Warning signs
Gives the full answer in the first reply.
Accepts the wrong step, or agrees with whatever your child says.
Long explanations your child scrolls past without reading.
Confident errors in arithmetic or units. Our post on checking an AI's physics solution shows how to spot them.
Your child cannot do the paper problem at the end.
After the session: three questions for your child
"What do you understand now that you didn't before?"
"What helped most: the hint, the question or the explanation?"
"Would you use it again on your own? For what kind of problem?"
If you are comparing tools, run the same twenty minutes with each, using the same problem. For a longer list of checks, see how to evaluate an AI tutor: 8 tests you can run in 20 minutes.
How Fermi's AI tutor is designed
We built Fermi's AI tutor to teach rather than answer: it holds a line of reasoning across a session, notices where a student is stuck, and decides whether to explain, ask or show. It writes on a board, draws the figure, runs the simulation and checks understanding as it goes. The checklist above applies to it too: run the session, do the paper test and judge for yourself. For practice on specific problems, the Fermi Problem Library has worked problems with a step-by-step solver that explains the reasoning rather than printing an answer.
