TL;DR
Feed an AI your exam date, your topic list with honest confidence ratings (or real practice scores), and your true daily hours. Ask for a week-by-week plan that gives weak topics at least twice as much practice as reading, starts every session with 15 minutes of retrieval on an older topic, schedules at least 2 fully timed mocks, and keeps one catch-up afternoon free each week. The single biggest lever is forcing practice-and-retrieval over re-reading: a meta-analysis of study strategies found combining spaced practice with retrieval lifts retention roughly 25% over using either alone. Copy the prompt below, paste your real numbers, then refine until weak topics are front-loaded and the plan has 20% slack.
The task
Final exam is in 3 weeks. The syllabus has 12 topics: you’ve covered 6 confidently, 2 you understand on a good day, and 4 you’ve barely started because homework dragged. You can study 2 hours on weekdays and roughly 4 on Saturday and Sunday — about 22 hours a week minus the time when life happens. You want a week-by-week plan that respects two facts: the topics you genuinely don’t know need practice, not re-reading, and the topics you sort-of know need spaced retrieval, not one end-of-week review. Plus at least 2 full mock tests before exam day.
Where AI helps — and where it does not
AI is good at building the week-by-week schedule, balancing new-material against revision slots, weighting practice over re-reading for low-confidence topics, and spacing mock tests so they give you usable signal. It can build in catch-up afternoons so a missed Tuesday doesn’t cascade through the whole plan.
What AI cannot do is diagnose what you actually don’t understand. “Confidence 2” on a topic might mean “I forgot one formula” or “I never understood the foundation.” Feed it real signal: a recent practice-test score or an honest self-assessment per topic. AI also can’t predict your real attention span. If you say “2 hours weekday” but your actual focused time is 80 minutes plus 40 of phone, the plan will look right and execute badly.
One specific failure mode: AI defaults to a daily mixed-topic schedule that switches every 30 minutes. For exam revision, longer blocks on one topic (60-90 min) plus a daily 15-min retrieval check on prior topics beats constant context-switching. Tell it the block structure you want.
Why the plan looks the way it does
The plan’s two non-obvious rules — practice over reading, and daily retrieval instead of end-of-week review — aren’t preferences; they track decades of cognitive-science findings. Re-reading feels productive but is one of the lowest-yield encoding methods. Retrieval practice (testing yourself) and spaced repetition (revisiting at growing intervals) are consistently rated among the most effective. A widely cited meta-analysis of the two found that combining spaced and retrieval practice improves long-term retention by roughly 25% over using either strategy on its own, with the biggest gains in durability and transfer to new problems. That is exactly the lift you want three weeks out: knowledge that survives to exam day and bends to questions you haven’t seen.
What to feed the AI
- Exam date, subject, and format (multiple choice / problem set / essay / oral)
- Full topic list with self-rated confidence 1-5 per topic (1 = barely started, 5 = could teach it)
- Recent practice-test or mock scores per topic, if you have them — these beat self-rating
- Hours available per day, honestly — weekday and weekend separately
- Days you cannot study at all (other deadlines, social commitments, recovery)
- Your strongest retrieval methods (problem sets, flashcards, redoing past papers, teaching a peer)
- Methods that haven’t worked before (often re-watching videos, highlighting)
- The one topic you’re avoiding (every revision plan has one) — name it so the model schedules it early, not last
Which AI to use (June 2026)
Any current chat model can build this plan. The differences are practical:
- Gemini 3.1 Pro — its 1M-token context (Google AI Pro, $19.99/mo as of June 2026) lets you paste a full syllabus PDF or several past papers in one go and have the plan reference them directly. Strong at pulling in current facts with fewer fabricated citations. Best when your source material is long.
- ChatGPT (GPT-5.5, Free $0 / Plus $20/mo) — best for math-heavy and reasoning-heavy subjects; it explains problem-set logic step by step like a Socratic tutor. Save your plan as a Project so follow-up “I fell behind, replan” chats keep your context.
- Claude (Sonnet 4.6, Free $0 / Pro $20/mo) — the clearest pedagogical explanations when a weak topic needs the concept re-taught, not just scheduled. Persistent memory reached the free tier in early March 2026, so it remembers your subject and confidence ratings across sessions; a Claude Project keeps your syllabus and constraints pinned.
- Dedicated tools — a generic study-planner app gives you calendar sync but rarely the practice-over-reading and spaced-retrieval logic above, so you end up re-prompting anyway. For ingesting your own notes, NotebookLM is worth pairing with whichever chat model builds the schedule.
A practical move: keep the plan in a ChatGPT Project or Claude Project so that mid-week “I missed two days, rebuild from today” requests inherit your full context instead of starting cold.
Copy-ready prompt
Build my revision plan for an exam on [date].
Subject + exam format: [paste]
Topic list with confidence 1-5 (or practice-test scores): [paste]
Days I cannot study: [paste]
Hours per day: weekday [N], weekend [N]
My most effective retrieval methods: [paste]
Methods that haven't worked: [paste]
The one topic I'm avoiding: [paste]
Structure:
1) Week-by-week table for the [weeks] weeks until the exam.
2) Each study day: topic, mode (read / practice / mock test / retrieval review), minutes per slot. Use 60-90 min focus blocks, not 30-min switches.
3) Start each session with 15 min of retrieval review on a prior topic — spaced repetition, not end-of-week-only.
4) Mock-test spacing — at least 2 full mocks, one at the end of week (total weeks - 2), one at the end of (total weeks - 1). Full timed conditions, plus 60 min review of wrong answers.
5) One built-in "catch-up afternoon" per week (suggest Sunday afternoon) — if I hit the week, that's rest; if I missed days, that's recovery.
6) Rest days — schedule 1 fully off per week, but not the day before the exam.
Rules:
- For any topic at confidence 2 or below, practice minutes must be at least 2x the read minutes. Reading without practice does not lift confidence on weak topics.
- The avoided topic must appear in week 1, not week 3.
- Leave 20% slack per day for slippage. Do not pack the plan to the minute.
- Past exam papers count as both practice and mock; tell me which I should treat as which.Shorter variant — one-week sprint
1 week to exam. Topics still weak: [paste with confidence]. Hours: weekday [N], weekend [N]. Give me a 5-day plan that hits each weak topic at least 3 times via practice (not re-read), plus a final mock 48 hours before the exam with 24 hours of wrong-answer review after. Skip rest days this week.Sample output
A useful week-1 row: “Mon: 60 min Ch 3 practice problems (confidence 2, the avoided topic — start here) + 15 min retrieval review on Ch 1 (confidence 4). Tue: 90 min Ch 4 practice + 15 min retrieval review on Ch 3. Wed: rest evening. Thu: 60 min Ch 5 read + 30 min Ch 5 practice + 15 min retrieval on Ch 4. Fri: 60 min Ch 4 deep practice (still confidence 2 after this week) + 15 min retrieval on Ch 5. Sat: 120 min full mock for Ch 1-3 + 60 min wrong-answer review. Sun: catch-up afternoon (2-3 hours if needed) or rest.”
A useful “avoided topic” note: “Ch 3 (Markov chains) is your stated avoidance. Scheduled Monday week 1, not end of week 3. Confidence 2 today — you need at least 3 practice rounds before the exam; placed on Mon week 1, Thu week 2, Sat week 3.”
A useful mock-test setup: “Mock 1: end of week 1, covers Ch 1-3. Time it strictly (120 min). Spend 60 min the same day reviewing wrong answers — not the next day; same-day consolidation matters. Mock 2: end of week 2, covers Ch 1-7, same protocol. The week-3 mock is optional, for confidence calibration.”
How to refine
- Force practice-over-read on weak topics: “For every topic at confidence 2 or below, the schedule must show practice minutes at least 2x read minutes. If a weak topic has 60 min read and 30 min practice, swap the weighting.”
- Place the avoided topic in week 1: “Re-check: the topic I’m avoiding must appear in the first 3 days, not the last week. If you placed it later, move it forward and explain what gets bumped.”
- Use 60-90 min blocks: “Re-read the daily schedule. If any day has 4 topic switches across the 2 hours, consolidate into 1-2 blocks. Switching every 30 min is read-and-forget; sustained blocks build.”
- Add daily 15-min retrieval: “Each session should start with 15 min of retrieval on a topic from 1-3 days ago. Add this if missing. Spaced retrieval is what makes weak topics stick.”
- Build in 20% slack: “Re-check: every day should have about 20% unscheduled buffer. If a day is packed minute-to-minute, cut the lowest-priority slot. Plans without slack don’t survive contact with real life.”
Common mistakes
- Treating “read again” as revision — re-reading is the lowest-yield encoding method; for exam recall, problem-solving and retrieval clearly beat re-reading
- No mocks until the last week — late mocks only tell you what you can’t change; the value is in week 1-2 when you can still fix gaps
- Cramming the weak topic on day 1 and never returning — once is not learning; spaced repetition across at least 3 sessions is what moves confidence
- Redistributing missed days across the whole remaining plan — it overpacks every day, you miss them all, the plan dies; use the catch-up afternoon, then resume normal
- Packing the plan to the minute — life happens; 20% daily slack is the difference between a plan you execute and one that breaks on Wednesday
- All-mixed-topic days — switching every 30 minutes hurts encoding; longer focus blocks plus short retrieval reviews beat it
- No rest day — fatigue compounds and you lose the last week to a fog you can’t push through; one full rest day per week is part of the plan, not a luxury
- Confidence based on hope, not data — “I’m a 4 on this” without a practice score is wishful; use actual scores whenever possible
FAQ
- What if I miss a planned day?: Use the built-in catch-up afternoon (usually Sunday). Don’t redistribute the missed work across every remaining day — that overloads each day and you miss them all. If you miss two days, one catch-up afternoon recovers about half; accept that the lowest-priority topic gets less coverage and continue.
- How do I handle topics I haven’t started?: Give them confidence 1 and the model will schedule them as high-priority new learning with both read and practice slots, not just revision. Don’t wait until week 2 to start them; the foundation needs time to settle.
- Should I do mocks under timed conditions?: Yes. Untimed mocks are a different cognitive task — you can’t learn pacing or pressure management without the timer. Time it strictly, same conditions as the real exam, no looking at notes mid-mock.
- My exam is open-book — does this change the plan?: Yes. Reading slots become “build your reference document” slots; an open-book exam tests how fast you can navigate the material under time pressure, not what you memorized. Mocks become “practice flipping to the right section in 90 seconds.”
- The model keeps under-scheduling mocks — what fixes it?: Add: “Mock tests are non-negotiable. Schedule at least 2 full timed mocks, the first at the end of week (total - 2) and the second at end of week (total - 1). Each mock includes 60 min same-day wrong-answer review. Do not let other study displace mocks.”
- Which AI should I use?: Gemini 3.1 Pro if you want to paste a long syllabus or past papers (1M-token context); ChatGPT for math and reasoning-heavy subjects; Claude for the clearest re-teaching of a weak concept. Keep the plan in a Project so “I fell behind, replan” requests keep your context.