Tutor your kid with AI
What it does: Turn AI into a patient tutor that never gives away the answer: it asks where your kid is stuck, works a similar example, and lets the kid solve the real problem themselves. The worry every parent has (“AI will do their homework for them”) is exactly what the ground-rules prompt prevents. How long: 15–30 minutes per session. Does it cost money: No.
| Effort | Setup | Risk |
|---|---|---|
| ★★★☆☆ Ground rules once, then easy | ★☆☆☆☆ Just the app | ★★☆☆☆ Needs a parent nearby at first |
How to do it
Using ChatGPT (or Google Gemini)
- Open a new chat and send the ground-rules prompt above. This one-time setup is the whole trick. Without it, AI defaults to just answering, which is exactly what you don’t want
- Snap a photo of the homework problem and send it. AI should respond by asking which step is confusing, not by solving it
- Sit with your kid for the first few sessions. Your job isn’t to explain math. It’s to notice when AI starts giving too much away and say “don’t tell them, ask them a question instead”
Why this works better than just asking AI the answer
A kid who gets answers learns to fetch answers. A kid who gets guided through a similar example learns the method, and the method transfers to the next problem. The ground-rules prompt forces the second pattern: similar example first, original problem second, hints one small step at a time.
Watch out
- AI can be wrong, especially on poorly photographed problems. If your kid says “but the AI said…”, that’s a reason to check the work, not to trust it
- Wrong level: if the “similar example” is way too hard or too easy, just say so (“make the example easier”) and it adjusts
- Kids drifting into chat: keep the session on the homework app screen. The tutor prompt keeps AI on task, but a curious 10-year-old can derail anything
- The goal is a kid who doesn’t need the tutor. Once a topic clicks, drop it and let them fly solo
Last tested: pending. This walkthrough was prepared from the products’ documented capabilities and has not yet been verified step-by-step by a real person. If you try it (whether it works or not), tell us on GitHub.