Studying with AI goes wrong in one predictable way: you ask it to *explain* a topic and it explains a slightly different topic than the one in your course, from memory, with one wrong date. The fix is to give it your chapter and ask for a format, not an explanation. Both tasks below do that.
Task 1: study notes from a chapter
| Prompt | What you get | Problem |
|---|---|---|
| "Explain the causes of World War I" | A generic essay from the model's memory | Not your syllabus; may include claims your textbook does not make |
| "Here is my chapter: [text]. Make revision notes as 5 headings, 3 bullets each, only from this text" | Notes that match what you will be examined on | None you cannot see: every bullet traces back to your chapter |
Your turn
Your turn: the notes prompt
Here is a chapter from my [SUBJECT] textbook. Make revision notes with [NUMBER] headings and 3 bullets under each. Use only facts that appear in the text. End with a one-line summary I could say out loud. CHAPTER: [CHAPTER TEXT]
Hint: "Use only facts that appear in the text" is the line that stops the model padding your notes with things your teacher never said.
Task 2: practice questions you can mark
The second skill is the same move with a different format: ask for questions and answers, so you can test yourself without the model in the room. Run this one as written; the passage is included.
Try it
Practice questions from a short passage
The wrong fact test
A model writes confident, well-punctuated sentences whether or not they are true. Students get caught because a wrong date *looks* the same as a right one. Try it: which of these two answers to "When was the Treaty of Versailles signed?" should you trust?
Which would you trust?
Which answer should you trust more?