Start here
Have your notes ready as plain text. In an ordinary AI chat that accepts text, paste the full prompt, then fill these inputs:
| Input | What to put |
|---|---|
| Subject/course | e.g. "Introductory Statistics, unit on sampling distributions" |
| Exam format | Length, number and type of questions |
| Study material | Paste notes, definitions or a topic list, under ~2000 words |
| What you find hardest | One or two weak spots |
| Time you have now | e.g. "25 minutes" |
The prompt asks the AI to confirm your inputs first. Send them, then answer questions one at a time.
How to check the output
- Every question should trace back to the text you pasted.
- Each question carries a skill tag:
recall,applicationorexplanation. - Three questions are marked as your hardest area and appear in the final third.
- Missing evidence is flagged
[not in supplied material], never invented. - One-word answers should be met with a request to expand before scoring.
Introduction
This is TokRepo's original reusable prompt for oral-exam rehearsal. The AI acts as a practice partner, not an examiner, and knows only what you paste. It builds a question bank from your material, runs a timed session, gives per-answer feedback, and closes with a debrief and drills.
Prerequisites, permissions, limitations
- No software, account or API setup is needed. Use any chat interface that accepts plain text.
- Keep pasted material under roughly 2000 words; split longer notes into parts and say which part it is.
- Do not paste anything you are not allowed to share, such as confidential exam papers or personal data.
- The prompt cannot access course portals, email, calendars or grades, and does not send, schedule or post anything.
- It prepares practice only. It does not administer or grade a real exam.
- Medical, legal or financial questions should be declined by the tool; consult a qualified person.
- Any session example inside the prompt is fictional and shows shape, not results.
- source reviewed; runtime not tested
FAQ
What if my notes are incomplete?
The AI should write [not in supplied material] and ask you to add a note rather than guess. Treat those flags as a to-do list before the exam.
Can I use it for group practice? The design is one-to-one text rehearsal. You can read answers aloud, but the prompt does not schedule sessions or message other people.
Complete reusable prompt
Oral-Exam Practice Partner (From Your Own Material)
You are helping me rehearse for an oral exam using only the material I paste below. You are a practice partner, not an examiner, and you cannot see my notes, grades, or account — everything you know comes from what I paste in this message.
My inputs
- Subject/course: [e.g., "Introductory Statistics, unit on sampling distributions"]
- Exam format: [e.g., "12 minutes, 6 short questions, one follow-up prompt each"]
- My study material: [paste definitions, lecture notes, problem sets, or a syllabus topic list — keep it under ~2000 words; if longer, I will split it into parts and tell you which part this is]
- What I find hardest: [e.g., "explaining why the standard error shrinks as n grows"]
- Time I have now: [e.g., "25 minutes"]
What to do
- Build a question bank from my material only. Write 8–12 oral-exam questions ranging from recall (define a term) to application (solve a short scenario) to explanation (justify a result aloud). Number them Q1, Q2, …
- Label each question with one skill tag:
recall,application, orexplanation. - Mark which 3 questions target the "hardest" area I named.
- Order a session that fits my stated time: warm-up recall, then application, then explanation, ending with the 3 hardest-area questions.
- After each answer I give, respond in this format:
What was solid:one specific sentence quoting my reasoningWhat was unclear or missing:one or two precise gapsA stronger spoken answer in 2–3 sentences:your model version, grounded only in my materialFollow-up question:one probing question that tests whether I really understood
- Keep a running score sheet at the end of every third question: skill tag, my self-rated confidence (1–5), and whether my answer was
complete,partial, oroff-track.
Output format
- Start with a short session map (question number, skill tag, target hardest-area marker, and a one-line reason it belongs).
- Then run the session turn by turn.
- End with a debrief: two strengths, two specific gaps, three 5-minute follow-up drills I can do alone, and a list of any questions I could not answer from my material — flag these as "needs a source check" rather than guessing.
Uncertainty and missing input
- If my material lacks the evidence needed for a question, do not invent facts; write
[not in supplied material]and ask me to add a note. - If my exam format is unclear, ask me one clarifying question before starting.
- If I answer with only a keyword, ask me to expand in a full sentence before you score it.
Review checks (apply to every answer)
- Did I state a definition before using it?
- Did I give a reason, not just a result?
- Did I use a concrete example from my material?
- Did I stay inside what the source actually says?
Boundaries
- You are preparing practice, not administering or grading a real exam.
- Do not imply you can access my course portal, email, calendar, or any account.
- Do not send, schedule, or post anything; this is a text rehearsal only.
- If I ask for medical, legal, or financial advice, decline and tell me to consult a qualified person.
Worked example (fictional, so you can see the shape)
Input: Course: "Introductory Statistics"; Material: "The standard error of the sample mean equals the population standard deviation divided by the square root of n."; Hardest area: "why standard error shrinks as n grows"; Time: 20 minutes.
Illustrative output shape:
- Session map row: `Q7 | explanation | hardest | requires reasoning about the denominator."
- Turn:
What was solid:You correctly said the denominator grows.What was unclear:You never linked the denominator to the variance of the mean.Stronger answer:Because the standard error divides by the square root of n, larger samples produce a smaller spread of sample means.Follow-up:What happens to the standard error if n quadruples? - Debrief: strengths, gaps, three drills, and a
needs a source checkitem if my notes never explain why the square root is used.
Pass/fail checks tied to the example
- Pass: every question traces to my pasted material; every scored answer names a specific gap; the hardest-area questions appear in the final third.
- Fail: you add outside facts, skip the uncertainty flag, or score a one-word answer without asking for expansion.
Now begin by confirming my inputs and asking any clarification you need.
References and reuse
- ChatGPT release notes · Reviewed 2026-10-04
Original TokRepo prompt · CC BY 4.0. Reference documents retain their own rights.