The Past-Paper Method: How to Study 5 Themes Instead of 50

You commented PAPERS on the carousel. Here's the full playbook — both prompts, the workflow, and how to use them without falling into the obvious traps.

What you actually need

• The last 5 past papers for your course (more is better — go to 8 if your department has them)
• A free Claude.ai account (the topic-map prompt works best here — Claude is better at structured extraction)
• A free ChatGPT account with Study Mode enabled (the marking prompt works best here — Study Mode coaches instead of just giving answers)
• The official syllabus / course outline for your module
• About 3 hours for one full cycle. You'll repeat the cycle per major topic.

The 2 prompts (copy-paste ready)

Prompt 1 — The Topic Map (paste into Claude)

I have a final exam coming up in [COURSE NAME].
Below are the last 5 past papers for this course.

Do four things:
1. List every distinct question theme that appears across these papers.
2. Rank them by frequency (how often each theme is tested).
3. For each theme, tell me which sub-topic of the syllabus it tests.
4. For each theme, describe what a full-mark answer would include — the structure, the depth, the type of evidence the examiner is looking for.

At the end, give me the 5-8 themes that carry roughly 80% of the marks across these papers.
That's my study list.

Here are the papers:
[PASTE PAPER 1]
---
[PASTE PAPER 2]
---
[PASTE PAPER 3]
---
[PASTE PAPER 4]
---
[PASTE PAPER 5]

Prompt 2 — The Marking Loop (paste into ChatGPT Study Mode)

You are an experienced examiner for [COURSE NAME] at a university level.

Below is my answer to the following question:
[PASTE THE EXAM QUESTION]

My answer:
[PASTE YOUR 10-MINUTE COLD ANSWER]

Do three things:
1. Mark this answer against typical examiner expectations for a final exam in [COURSE NAME]. Give me a score out of the marks available.
2. Show me exactly where I lost marks — quote the line and tell me what's missing or weak.
3. Reverse-engineer the marking scheme: what would a full-mark answer include that mine didn't?

Be honest. Don't soften it. I'd rather find out now than in the exam.

The Calibration Trick (do this once, before you start marking)

The first time you mark anything in a course, the AI is guessing what "examiner expectations" actually means. Give it an anchor and the marking gets sharp instead of soft.

Before running Prompt 2 for the first time in a course, find ONE known full-mark answer — from your textbook, your lecturer's model answer, a published marking memo, or a senior student's high-scoring paper. Then run this version of Prompt 2 first:

Below is a known full-mark answer for [COURSE NAME]:

[PASTE THE MODEL ANSWER]

Calibrate your marking against the standard of the answer above. From now on, when I paste my own answers, mark them against this benchmark — be honest about the gap.

My first answer to mark:

[PASTE THE EXAM QUESTION]

[PASTE YOUR 10-MINUTE COLD ANSWER]

You only do this once per course. The AI carries the calibration through the rest of your marking session in that chat.

Without calibration: AI marking is soft and vague — "good effort, just add more detail." With calibration: AI marking is sharp and specific — "your paragraph 2 misses the framework. The model uses three named cases here. You used none. Marks lost: 5."

The 4-step workflow

Step 1 — Build the topic map (30 min) Run Prompt 1 in Claude. You'll get back 5-8 ranked themes. That's your study list. Print it. Stick it on your wall.

Step 2 — Cold-write one answer per theme (10 min each, 50-80 min total) For each theme, find one representative question from the past papers. Write your best 10-minute answer cold — no notes, no Googling, no tools. This is your baseline. It will be worse than you think. That's the point.

Step 3 — Calibrate once, then mark each answer (5 min per answer) The first time you mark in this course, use the calibration prompt (above) to anchor the AI against a known full-mark answer. After that one calibration step, paste each cold answer into the same chat using Prompt 2. Note the gaps. Note what a full-mark answer would add.

Step 4 — Re-write, re-mark, repeat until it scores (20-40 min per theme) Re-write the answer with the gaps closed. Re-mark. Keep going until the AI gives you a full-mark or near-full-mark score. Three rounds is usually enough.

By the end, you don't have notes on 50 topics. You have 5-8 polished answer shapes, already exam-ready, that map to 80% of the marks.

How to not screw this up

• Don't skip the cold-write. It's tempting to look up the answer first. Then you're studying recognition, not retrieval. The cold-write is what makes this work.
• The AI is not the examiner. It's a calibrated proxy. If your university uses very specific marking criteria (e.g. medicine, law, accounting), find one official model answer and ask the AI to mark in that style.
• Don't trust round 1. AI marking is roughest on the first pass. Always do 2-3 rounds before assuming you've nailed it.
• Past papers older than 5 years drift. If your course content was updated recently (new syllabus, new lecturer), don't go back further than 5 years. The patterns change.
• Some courses have one banger question that always shows up. When the topic map ranks one theme way above the rest, take it seriously. Lecturers have favourites.

What to do next

1. Run Prompt 1 today. Get your topic map.
2. Cold-write one answer per theme tomorrow.
3. Mark + re-write through the weekend.
4. Hit the exam.

If this lands well, the same method works for:

• Oral exams / vivas (replace Prompt 2 with a roleplay grilling prompt)
• Practical/applied exams (give the AI the case study and ask it to grade your approach)
• Open-book exams (the topic map matters MORE, not less — you're searching, not memorising)

Follow @africantechbro for more. Tag me when you write your final answer — I want to see the round 3 vs round 1 difference.