On this page
Scroll back through a month of chats with an AI and you may see a pattern. The same few chapters keep coming up. You ask about the sign of work done by friction, then ask again in a different form ten days later. Those repeats are a map of your weak spots, written in your own words. Here is how to read it in 30 minutes.
Why are old AI chats useful for revision?
A doubt is a record of a moment when your understanding broke. Research on errors suggests those moments are worth revisiting. In her review Learning from Errors, Janet Metcalfe of Columbia University reports that making errors and then getting corrective feedback helps learning. She also notes that analysing the reasoning that led to the mistake is crucial. Mistakes made with high confidence turn out to be corrected more readily than low-confidence ones.
Your chat history already holds the error and the feedback. What it lacks is you checking that the fix held.
The trap is rereading old answers. A large review of ten study techniques by Dunlosky and colleagues (2013) rated rereading and highlighting as low utility. It rated practice testing and distributed practice (spreading study out over time) as high utility. So don't reread your chats. Turn them into questions.
How do you scan your chat history for weak spots?
Set a timer for 30 minutes, pick one subject and work through these steps.
Open your chat history and scroll back three to four weeks. If your chat app can search past chats, try chapter names too.
For each chat that was a real doubt, write one line in a notebook: the chapter and the doubt in your own words.
Tag each line with one letter. C means concept (you did not understand the idea). M means method (you understood it but could not set up the problem). R means recall (you forgot a formula or fact). S means slip (a sign, unit or arithmetic error).
Put a tick next to any doubt you asked more than once, even if you worded it differently.
Count the tags per chapter.
The ticks matter most. A doubt you asked twice is one the first answer did not fix.
What does a revision list from old chats look like?
Here is a made-up example from a Class 11 physics student's three weeks of chats.
Work done by friction on a block, negative or positive? (C, ticked)
Setting up equations for two blocks on a pulley (M, ticked)
Formula for range of a projectile (R)
Why the normal force is not always mg (C)
Lost a minus sign in relative velocity (S)
Two blocks connected by a string on an incline (M)
Four of the six lines are concept or method doubts, including both ticked ones. The problem is setting up forces, not memory. Recall and slip doubts go on a short formula sheet. Concept and method doubts become the revision list.
Turn each one into a question you could be asked cold, for example: "A block slides to rest on a rough floor. Is the work done by friction positive or negative, and why?" Then write the answer with the chat closed, and only then open the old chat to check.
How do you make the fixes stick?
Test yourself, and space it out. In a study by Roediger and Karpicke (2006), students who took recall tests on a passage remembered more on tests two days and one week later than students who restudied it. On a test after five minutes, restudying looked better, and it made students more confident. That is why a quick reread before an exam can mislead you.
A simple plan:
Day 1: answer every question on your list from memory and mark the ones you got wrong.
Day 3: redo only the wrong ones, plus two that you got right.
Day 7: redo the whole list, mixed up, with a timer.
Then fold what is left into a four-week spaced revision schedule.
A doubt log that keeps AI answers solved saves you the scan next month. Don't revise only this list, though. Revising only your mistakes leaves gaps in topics you never thought to ask about.
Where does an AI tutor fit in?
Match the help to the tag. For concept doubts, read the idea again in a lesson, not a one-line answer. The Fermi Concept Library covers every concept in physics, chemistry, biology, mathematics and computing, each with its own lesson. For method doubts, the Fermi Problem Library has a step-by-step solver that explains the reasoning rather than printing an answer. If you have no idea where to start, Fermi's AI tutor can work out what you already know and build a plan for the session.
Tonight, set the timer, open last month's chats and write the first six lines.
