How AI Finds the Gaps in What You Actually Understand
Some links below are affiliate links — I may earn a small commission, at no extra cost to you, if you sign up for a tool I recommend. More on how I choose what to recommend.
You read the chapter three times. You highlighted it on the second pass and made notes on the third. Walking into the hall you felt fine. Then question two asked you to explain the mechanism, and what came out of your pen was four sentences of confident-sounding mush that you knew, as you wrote it, wasn’t going to score.
Afterwards you couldn’t work out what went wrong. You did the work. You could have told anyone in the corridor what that topic was.
Here’s the uncomfortable part: you could recognise it. You couldn’t produce it. Two different abilities, and only one gets tested.
Why re-reading lies to you
The second time you read a page it’s easier than the first. The third time, easier still. Your brain reads rising fluency as rising understanding — from the inside, the two feel identical.
But what improved is recognition. You’ve got better at going “yes, I know that” when the material is in front of you. In an exam nothing is in front of you. You generate it from a blank page, and generating is a skill you didn’t practise once.
The fix has been known for decades: stop reviewing, start retrieving. Explain it out loud, from memory, book shut. The hard part was always the next step — someone has to tell you where your explanation was weak — and that someone used to need expertise and spare time.
The move: explain it badly, then ask for a diagnosis
Open a chat. Pick one topic. Close your notes. Explain it from memory, in a couple of hundred words — including the parts you’re shaky on. Especially those. Don’t polish it: a tidy explanation assembled by peeking is worthless here.
Then ask it to diagnose rather than grade. A grade tells you 6/10 and you learn nothing. A diagnosis sorts your explanation into three categories:
What you got factually wrong
The straightforward one. You mixed up two terms, had the causation backwards, attributed something to the wrong process. Uncomfortable to read, easy to fix — you now know which page to reopen.
What you left out entirely
The parts that never appeared at all. You can't notice these yourself: absences are invisible from the inside, and your account felt complete precisely because everything in it was something you knew.
What you stated confidently but vaguely
The important one. Sentences that sound like understanding, carry no content, and earn no marks. "It's because of the pressure difference." "The two systems interact." You wrote them fluently and believed them — and an examiner reads them as someone who doesn't know the mechanism.
Category three is the whole reason to do this.
You cannot catch your own confidently-vague statements, because from the inside they feel identical to real knowledge. “I know this” is a sensation produced by fluency, and a vague phrase you’ve read fifteen times is extremely fluent. A highlighter can’t see it. A re-read can’t see it. Your study group mostly can’t either — they nod at the same phrases you would.
A reader who takes your words literally and asks “what does that actually mean?” sees it instantly. That’s what AI does here, at 11pm, as often as you want.
The exact prompts to use
Copy these into ChatGPT or Claude. The free tier handles all of it.
1. The core diagnosis
I’m studying [subject] and I have an exam in [time]. Below is my explanation of [topic], written from memory without looking at my notes. It isn’t polished and isn’t meant to be.
Don’t grade it and don’t encourage me. Diagnose it, in three sections:
- Factually wrong — anything I stated that is incorrect. Quote my words, then correct them.
- Missing — anything important I didn’t mention at all, ranked by how likely it is to appear on an exam.
- Confidently vague — sentences where I sounded certain but said nothing precise. Quote each one and tell me what a marker would want instead.
Be harsh. If a sentence would earn zero marks, say so plainly. Finish with the weakest part of my understanding and what to re-read first.
My explanation: [paste your explanation]
2. Make it interrogate you
You are examining me on [topic] in [subject], at [level — e.g. second-year undergraduate].
Ask me one question at a time and wait. Do not reveal the answer, a model answer, or any hint until I have committed to a response — even if I ask you to.
After each answer, tell me what was right, what was missing, and whether I was being precise or waving my hands. Then go one level deeper on anything I was vague about.
Start broad and get harder. Ten questions, then summarise my three weakest areas.
3. Find the gap you don’t know you have
Based on the explanation I just gave you, what am I likely to get wrong on an exam about [topic] that I haven’t mentioned at all?
Work out what my explanation implies I believe, and where that belief would break down. Give me three exam-style questions I’d probably answer badly right now, and name the underlying misunderstanding behind each.
That third prompt is the underrated one. It isn’t asking about your explanation — it’s asking what your explanation reveals. Describe a process only in the direction it usually runs, and a question that runs it backwards will destroy you. Nothing in your explanation looked wrong.
A worked example, so you can judge the output
An example from first-year economics. The student shut the textbook and typed this:
When inflation is high the central bank raises interest rates. Borrowing gets more expensive, so people spend less, and because there's less money going around, prices come down. It works through the money supply, basically. It takes a while because of the lags. Higher rates also make the currency stronger, which helps. The main risk is that if they raise too much you get a recession.
Reads fine. Sounds like someone who did the reading. The diagnosis, condensed:
Factually wrong. “because there’s less money going around, prices come down.” Modern central banks set the price of credit, not the quantity of money. The mechanism runs through demand: borrowing costs rise, saving gets more attractive, spending and investment fall, firms lose the ability to raise prices. “Money supply” here signals a model your course isn’t using.
Missing. Three things never appeared. Inflation expectations — arguably the main channel, since a credible central bank changes behaviour before rates bite. The labour market, and how weaker demand feeds through wages. And the difference between demand-pull and cost-push inflation: you treated one mechanism as though all inflation were the same kind.
Confidently vague. “It works through the money supply, basically.” The word “basically” is carrying the whole sentence; there’s no mechanism, only a gesture at one. “It takes a while because of the lags.” Which lags, how long? “Makes the currency stronger, which helps.” Helps how — you skipped the step where a stronger currency lowers import prices. “The main risk is a recession.” Via what path?
The hidden-gap question then returned this: you’d likely face a question about an oil price shock, and you’d apply the demand story to a supply problem. Nothing here suggests you know there’s a difference.
Nothing in that paragraph was obviously broken. That’s the point. Three re-reads would have produced it again, with more confidence.
It works just as well for essay subjects
There’s an assumption that this only suits science and maths, where answers are crisp. The opposite is closer to true: confidently vague is endemic to essay subjects, because the vocabulary sounds substantial on its own. Explain your argument instead of a mechanism, and ask which claim you didn’t support:
Below is my argument for the essay question: [question]. It’s from memory and rough.
For each claim, tell me: (a) did I give evidence for it, (b) is that evidence actually relevant, (c) what’s the strongest counter-argument. Flag any sentence that sounds analytical but is really just describing. Then name the claim my argument most depends on, and say whether I’ve earned it.
[paste your argument]
“Sounds analytical but is really describing” is the essay version of confidently vague — and the most common reason a competent-feeling essay comes back in the middle band.
Two things that make this work better
A realistic 20-minute round
Pick the topic that makes you uneasy (1 min)
Not the one you enjoy — the one you keep meaning to come back to. Paste in the module's learning objectives too, if you have them.
Close everything and explain it (5 min)
Notes shut, tab closed, phone face down. Two hundred words, or two minutes of talking. Include the shaky parts — a sample that hides them diagnoses nothing.
Run prompt 1, then prompt 3 (6 min)
Read category three first — the part you had no way of finding alone. Then ask the hidden-gap question and write down the three exam questions it predicts you'd fumble. Those are your revision targets, in priority order, chosen by evidence rather than mood.
Go back to the actual source (5 min)
Your notes, your textbook, your slides. Confirm each correction from material you trust. Not optional — see the limits below.
Re-explain the same topic (3 min)
Same prompt, fresh chat, from memory again. Watching the list shrink is the only reliable evidence that something moved.
What AI can’t do here (the limits to know)
Common mistakes the prompts will catch
Studying something where this falls apart — problem sheets, a language, a viva, a clinical placement? Email help@aiforyourday.com and tell me what happened. Real reader cases make the next version of this article better.