Guide
Using ChatGPT to build a revision timetable — and what it gets wrong
Asking an AI to plan your revision is a genuinely good idea, and it fails in a specific, predictable way. It is excellent at deciding what you should revise and unreliable at the arithmetic of when. Knowing which half to trust is most of the skill.
The prompt that works
Vague questions get vague plans. Give it the constraints it cannot guess:
I'm sitting A-levels in Maths, Biology and Geography. My papers are: Maths 14 and 19 May, Biology 22 May and 5 June, Geography 8 and 15 June. I'm confident in Maths, shaky on Biology paper 2, and Geography is somewhere in between. I can do about 2 hours on a school night and 4 at weekends, and I want Sundays off. Build me a revision plan that starts in January, covers every topic at least twice, and puts the final pass for each subject just before that subject's paper.
Four things make that prompt work: every paper date, honest confidence per subject, real available hours rather than aspirational ones, and an explicit instruction about the shape — cover twice, final pass before each paper.
Three things it reliably gets wrong
It blocks subjects instead of interleaving them. Ask for a revision plan and you will usually get "Week 1: Maths. Week 2: Biology." That is the tidiest possible answer and it throws away the spacing effect, which is the entire reason a plan beats no plan. Ask explicitly for two to three subjects a day, rotating.
It loses the thread over long documents. A plan running from January to June is twenty-odd weeks of dated entries. Somewhere in the middle the structure drifts — proportions stop adding up, a subject quietly disappears for a fortnight, the final pass lands after the exam. This is not a knowledge failure; it is what happens when you ask a language model to hold arithmetic across a long generation. We wrote up how badly this can go after one put 92% of a 45-week plan into a single phase.
It agrees with your worst ideas. Tell it you will do eight hours a day through February and it will build you eight-hour days. It has no stake in whether you can sustain that, and a plan you abandon in week three is worse than a modest one you finish.
Check these three things before you trust it
- Does every subject's last session fall before that subject's paper? Scan the end of the plan first. This is the most common single error and the most damaging.
- Does every topic appear at least twice, with a gap between? Once is coverage, not revision.
- Add up one ordinary week. If it exceeds what you actually said you had, the rest of the plan is fiction.
The bigger problem: it stays in the chat
Even a perfect plan is a wall of text in a conversation you will scroll past tomorrow. There is nothing to tick, no way to see whether you are on track in March, and no way to adjust it when a fortnight disappears — which it will.
That gap is the entire reason Consistency Calendar exists. Your assistant does the part it is good at — reading your syllabus, judging which topics need three passes, writing sessions that say something more useful than "revise Biology" — and the dates, phases and spacing are computed rather than improvised. You get a private calendar you tick off, and when the plan needs rebuilding in March, the sessions you have already done survive it.