The AI summary of your set text: why it catches you out in the exam

It is half past eleven, the Of Mice and Men test is tomorrow, and you stalled somewhere around chapter three. The temptation is shaped like a text box: summarise the book, list the characters, give me three quotes. Forty seconds later you have something that looks like revision notes and gives you a calm you have not earned. The problem is not the ethics. It is two very specific traps, and both show up the next morning.

Trap 1: the AI invents the book it does not know well

With the classics on every syllabus, the AI copes: Macbeth, Of Mice and Men, An Inspector Calls. With the young adult novel your teacher chose this year, far less, and then something happens that you cannot see from the outside: when it does not know, it does not lower its voice. It produces a chapter that does not exist, a character with a believable name and a quote that sounds like the author and appears on no page. In May 2025 a Chicago newspaper printed a summer reading list a freelancer had produced with AI: of fifteen recommended books, ten did not exist. Real authors, invented titles, flawless blurbs. If it happened to a newsroom, it can happen to your notes at half past eleven. The OpenAI researchers who studied the phenomenon in 2025 compare it to students facing a hard exam: when unsure, they guess, because guessing scores and I don't know does not.

The two minute test

• Ask for the chapter list of the book your class is reading right now and compare it with the real contents page. If the book has chapter titles, you will see at once whether they are invented.
• Ask for a word for word quote from a specific chapter and look for it in the book. A quote that appears on no page is the clearest signal.
• Ask what a minor character does in the final chapter. Summaries live off the main characters, and that is where the seams show.
• Ask the same question in two new chats. If the answers contradict each other, at least one is invented, and you do not know which.

Trap 2: the exam asks about what is only in the text

A summary gives you the plot, and the plot is what your teacher takes for granted. The marks are for something else: what Candy says after his dog is shot, what Macbeth sees before he goes to Duncan's chamber, how the narrator's tone shifts from one chapter to the next. That level of detail is in no summary, human or machine, because a summary exists precisely to strip it out. Whoever has read the book has it, whoever summarised it does not, and a well set question separates the two in thirty seconds.

Read first, then get tested. AI is no substitute for reading, but it is very good at what comes after: asking you questions, picking apart a thesis, forcing you back into the text. The rule is simple: you only ask it about what you have already read, and when it gives you a detail, you check it in the book, not the other way round.

I have read [TITLE] by [AUTHOR] up to chapter [NUMBER]. Act as a demanding examiner: eight questions, from easiest to hardest, that only someone who has read the text could answer, not someone who read a summary. Include two about a minor character and one about a specific detail in a scene. Ask them one at a time, wait for my answer and correct me without giving the solution until I have tried twice. If you are not sure about a detail of the text, say so instead of inventing it.

My thesis about [TITLE] is: [YOUR THESIS IN ONE SENTENCE]. Argue it with me the way a teacher who disagrees would: demand specific quotes, point me to scenes that contradict my idea and do not give in until I argue with the text in front of me. At the end, tell me which point I defended worst and what I should reread.

I want to understand why [CHARACTER] does what they do in [SCENE] of [TITLE]. Give me three different explanations of their motives, each backed by a specific moment in the book that I can find. Then ask me which one convinces me most and why, and push back on my choice.