Ground it in good sources
Where AI slips
Basics of well-established subjects are usually explained well. Errors cluster in specialist detail (exact formulas, dates, legal or medical specifics), recent developments, local rules, and niche topics with little written about them. Be most careful exactly where you know least.
Learn from your own materials
Upload your course notes, textbook chapters or official guides and ask the AI to teach and quiz you from those only. Tools such as Gemini Notebook (formerly NotebookLM) are built for this and show which passage each answer came from. Grounded answers are far less likely to wander.
Ask for sources, then open them
For facts you'll rely on, ask where they come from and check the source exists and says what the AI claims. A good textbook, an official site or a recognised course still beats a chatbot's summary. If the AI and your course disagree, ask the AI to explain the difference, then go with the course unless you find strong evidence otherwise.
Test yourself without AI
The final check is whether you can do it unaided: solve the problem, explain the idea, hold the conversation. Every week or two, do one session with the AI switched off. It tells you what you've actually learned, and it is the best preparation for exams, interviews and real use.
If you learned one wrong 'fact' from AI this year, how would you ever find out?