The task
The lecture ended four minutes ago. You followed about 60% of it, your notes have three sentences that trail off mid-derivation, and office hours are tomorrow at 2:30. You want to walk in with five specific questions — each tied to a slide, a line in your notes, or a moment when the professor said “obviously” and you wrote ”???” in the margin — not a vague “could you explain everything from the second half again?”
Where AI helps — and where it does not
AI is excellent at turning fuzzy notes into precise questions, ranking them by which one would unlock the most downstream material, and prompting you to attempt an answer yourself before asking. It is also good at catching the questions you should have but don’t realize yet — the cross-concept ones that connect today’s lecture to last week’s. What AI cannot do: ask about content your notes did not capture. If you missed slide 23 entirely, the model has no signal that slide 23 exists. Either go back to the slides, or feed AI the lecture recording / textbook chapter so it has the missing context.
The named failure mode: the explain-it-again question. AI defaults to “Can you re-explain the proof on slide 17?” which makes the professor rerun the lecture for one student. Force every question to reference a specific line, equation, or step and to name the exact gap.
What to feed the AI
- Your lecture notes — typed, photo-OCR’d, or even an audio transcript dumped raw
- Slide numbers if you have them, even partial (“got lost around slide 22-25”)
- 2-3 moments when you got lost, even if you cannot articulate what — “around the derivation transition”, “when she switched indices”
- Whether you have the textbook chapter or course reading available as backup context
- Last week’s lecture notes if today’s lecture builds on them
- Your goal for office hours — pass the next problem set, prep for the midterm, ask one deep question
- Whether the professor allows / encourages emailed questions in advance
- The course level — intro, upper division, grad — which sets how precise the question can assume
Copy-ready prompt
Generate 5 follow-up questions from my lecture notes.
Notes (raw or cleaned):
{paste notes, OCR text, or transcript}
Moments I got lost: {list with slide / timestamp / line reference if possible}
Backup context available: {textbook chapter y/n, last week's notes y/n}
Course level: {intro / upper / grad}
Office hours goal: {problem-set / midterm prep / one deep question}
For each of the 5 questions return:
1) The question itself — precise, referencing a specific slide / line / equation / step. Not "can you re-explain X."
2) Why this question — name the exact gap it exposes ("the index switched without comment", "the assumption stopped being stated", "the example is a special case but it wasn't labeled").
3) Priority rank from 1-5 — which question, if answered, would unlock the most downstream understanding for the next problem set?
4) What I should try to answer myself first before asking — a 30-second self-check (re-read X, look up Y, sketch Z).
Rules:
- No question may start with "can you explain X again."
- At least 1 question must connect today's lecture to a prior concept.
- At least 1 question must ask "is this a general result or a special case?" — that's where most confusion hides.Shorter variant — 1 sharp question only
From these notes: {paste 200-300 words}
Generate the single sharpest question I should ask in office hours. Tie it to a specific slide or line. Include a 1-sentence self-check I should try first.Sample output
A useful office hours question: “On slide 23, in the derivation of the gradient update, the summation index changes from i to k between lines 2 and 3 without comment. My notes say ‘because of summation’ but I cannot see what is being summed across k versus i. Is k indexing the batch and i the feature, or vice versa?” The professor answers in 30 seconds; you walk out unstuck.
A useful cross-concept question: “In last week’s lecture we introduced regularization as an L2 penalty. Today’s loss function uses what looks like the same penalty term but the lambda is multiplied by the number of samples. Is that a different regularization (mean vs sum), or the same one with a different convention?”
A useful self-check note: “Self-check first: re-read the textbook’s footnote on convention before asking — it likely covers the lambda-vs-N-lambda distinction. If still unclear, ask.”
How to refine
- Reference a specific anchor: “Every question must point to a slide number, an equation label, or a line in my notes. Generic questions get generic answers.”
- Name the exact gap: “For each question, the ‘why’ line must say the specific operation that confused me — ‘index changed’, ‘assumption was dropped’, ‘example wasn’t labeled as special case’.”
- Demand a self-check: “Add a 30-second self-check I should try before asking. If the check would resolve the question, don’t ask it — ask the next one.”
- Find the connector question: “At least one question must link today’s lecture to last week’s. Cross-concept questions reveal whether the course is building on something I missed.”
- Prioritize for the problem set: “Re-rank: the top question should be the one most likely to appear on next week’s homework.”
Common mistakes
- Asking “can you re-explain X” — the professor either reruns the lecture (wasting both of you) or summarizes (you learn nothing new)
- Showing up with 10 questions — pick 2-3 highest-priority, save the rest for next office hours or email
- Not trying to answer yourself first — embarrassing in front of the prof and discourages you from the next harder attempt
- Asking only narrow notation questions and missing the conceptual one — fluency at the symbol level without understanding the why
- Phrasing questions to hide that you didn’t follow — clarity costs you 5 seconds of pride and saves the prof 5 minutes
- Letting AI hallucinate slides you don’t have — if it references slide 31 and you only have slides 1-25, drop the question
- Going to office hours without re-reading the relevant slides first — most of your questions resolve in the re-read
- Treating office hours as the place to learn the content — it’s the place to fix specific gaps, not to absorb the lecture from scratch
FAQ
- What if my notes are messy?: OCR or transcribe first, then ask AI to clean and restructure before generating questions. Cleaner inputs produce sharper questions. The cleanup also surfaces gaps you didn’t realize you had.
- Should I email the questions in advance?: Yes if the professor or TA encourages it. They will arrive prepared and you both save the awkward “let me find the slide” minute. If unsure, ask once and observe how they react.
- What if I missed an entire 10-minute stretch?: Tell AI explicitly: “I missed roughly 10 minutes around slide 20-23. Generate 2 catch-up questions whose answers would let me reconstruct the missed stretch.” Pair that with the textbook chapter as backup context.
- Should questions assume I read the textbook?: Yes for grad and upper-division; the prof will phrase the answer at that level. For intro, signal explicitly: “I haven’t read the chapter yet — pitch the question for the lecture material only.”
- The questions still feel generic — what gives?: Add this to the prompt: “Replace any question that could be asked about a different course’s lecture. Every question should be impossible to ask without the specific slides I gave you.”
Related
- AI lecture note cleanup
- AI study buddy
- Lecture Notes Cleanup Prompts for Study-Ready Material
- Textbook Chapter Summary AI
- AI Exam Study Plan
- Explain a Difficult Concept
Tags: #AI writing #Learning #Workflow #Study