A Study Plan for Exam Season: What the Research Says

Every exam period starts the same way: you write a plan that looks beautiful, and three days later it is scrap paper. This is rarely a lack of discipline. It is because most study plans are built from wishful thinking rather than from what is actually known about memory.

This guide builds your plan the other way round: first the findings, then the plan. And where the evidence is thin, that is said too.

First truth: you underestimate how long things take. Everyone does.

In 1994, Buehler, Griffin and Ross asked psychology students when they expected to finish their thesis. The best estimate averaged 33.9 days, while they actually needed 55.5 days. Only 29.7% of those surveyed met their own deadline. This is called the planning fallacy.

The countermeasure from the same work is interesting: the optimistic bias disappeared when people explicitly linked their forecast to earlier experiences. For you, this means: do not estimate how long a chapter should take. Check how long the last comparable chapter actually took. The study does not, by the way, support a blanket buffer such as "plus 50%". Your own numbers are the better buffer.

Second truth: spreading out beats cramming

The spacing effect is one of the best-studied findings in learning research. The meta-analysis by Cepeda and colleagues (2006) evaluated 839 comparisons from 317 experiments. Gap between sessions and time until the test work together: the later the test, the larger the gap between repetitions may be.

A small study by Kornell (2009) with 25 students and word-pair flashcards shows what this looks like in practice. Those who studied spread over four days scored 54% correct on the test. Those who packed everything into the last session scored 34%. Same number of passes per card, just distributed differently.

The tricky part comes next: in Kornell's experiments, the vast majority of participants benefited from spreading out, yet many of those who perceived a difference thought cramming was better. Cramming feels more effective. That is exactly why it is so popular.

Put honestly: a new meta-analysis from school classrooms (Ruitenburg et al., 2026) found a smaller effect for distributed learning, g = 0.20. Compared with massed practice it was g = 0.39, and compared with learning that was already somewhat distributed, even more distribution brought no significant additional benefit. The message is therefore: spread it out at all, rather than putting everything on the last evening. "The bigger the gap, the better" does not hold up that simply.

How large should the gaps be?

In a 2008 online experiment with 1,354 people, Cepeda and colleagues systematically varied how far apart two study sessions were and when the test took place. Result: there is an optimal gap, and it depends on when the test occurs. For a test one week later, the optimum was roughly 20 to 40% of that time span; for a test one year later, 5 to 10%. The optimal gap compared with no gap at all produced on average 64% more correctly recalled answers.

What this is not: a ready-made calendar. The study tested two learning occasions with facts, not a repetition sequence like "after 1, 3, 7 days". Take the logic with you, not the percentages as a timetable. This is exactly what spaced repetition software is for: it calculates the gaps per card. How this works is explained in Spaced repetition explained and FSRS explained.

Third truth: practice tests are learning, not just checking

Many people save old exam papers for the last few days as a final test. This wastes their greatest benefit. Retrieval practice is itself one of the most effective study methods.

In Roediger and Karpicke (2006), people who tried to recall a text from memory after reading it retained 68% after two days instead of 54%. And this was without feedback. The large classroom meta-analysis by Yang et al. (2021) across 222 studies found an average advantage for quizzes of g = 0.499, roughly half a standard deviation.

The catch: how strongly tests worked depended, among other things, on corrective feedback, repetition, and how similar practice and final tests were. A practice test whose mistakes you never look up brings far less. Research does not provide an optimal timing for the last practice test.

These cards are, by the way, a good example of exam-relevant questions: they ask for the result, not just the name of the study. More on this in Writing good flashcards.

Fourth truth: interleaving helps, but not everywhere

Interleaving means practicing related task types mixed together rather than in blocks. Brunmair and Richter (2019) found an advantage of g = 0.42 in their meta-analysis of 59 studies. But it depended heavily on the material. For math problems g = 0.34, for explanatory texts no significant effect, and for words block learning was even better.

What many confuse with interleaving: switching between entire subjects. This was typically not studied at all. "Switch from biology to law every 25 minutes" is therefore not a proven interleaving trick. What makes sense is this: within one subject, mix the different exam-relevant task types the way they appear mixed in the exam.

Fifth truth: sleep is part of the plan

Newbury and colleagues (2021) summarized research on sleep deprivation and memory. A completely sleepless night before learning clearly worsened later memory performance (g = 0.621). Sleep deprivation after learning also harmed memory, though less strongly (g = 0.277). Caveat: both analyses found signs of publication bias, and what was studied was complete sleep deprivation, not "one hour less".

An observational study of students in a chemistry course (Okano et al., 2019) fits the picture: sleep duration, quality, and regularity over weeks were associated with quiz and exam results and explained about 24% of the differences. The single night directly before the test, by contrast, was not significantly associated with the result. This is not causal proof, and certainly not a license to pull an all-nighter. It rather suggests: the sleep of the whole exam period matters, not just the last night.

Sixth truth: if-then beats "I should do this"

"Tomorrow I'll study more" is not a plan. "When I sit at my desk after breakfast, I will answer the first practice question without my notes" is one. Psychology has studied such implementation intentions extensively. Gollwitzer and Sheeran (2006) found an effect of d = 0.65 on goal attainment across 94 tests.

To be honest about it: a larger, more recent synthesis of 642 tests found d = 0.36, and after correcting for publication bias only d = 0.15. The effect is therefore probably smaller than previously thought, and it concerns goals from all kinds of life areas, not directly exam grades. But if-then plans cost you nothing except one sentence.

And Pomodoro?

Breaks do help, that is fairly certain. A meta-analysis by Albulescu and colleagues (2022) found more vitality and less fatigue after short breaks. Performance did not rise significantly, however. A randomized study with Pomodoro-like intervals (Biwer et al., 2023) found a more pleasant learning experience, but no differences in effort or completed tasks. No optimal work-break rhythm was identified. Take breaks when it suits you. The timer is a matter of taste, not a learning hack.

The plan in five steps

1. Inventory. List all exams with dates and material. Estimate the effort using your real numbers from earlier semesters.
2. Turn material into cards and practice questions early. What you do not translate into questions, you cannot practice retrieving.
3. Retrieve daily, spread out. Better to do a bit from every subject each day than one subject per week in one go. The app calculates the gaps for you.
4. Build in old exam papers early, look up your mistakes, and turn them into new cards.
5. Schedule sleep like an appointment, not like leftover time.

Exam date in Anki and checkst

Spaced repetition apps normally plan for long-term retention, not for a fixed date. In Anki, Custom Study or a filtered deck helps with this: you can select by tags, forgotten cards, or material not yet due, and the manual explicitly names exam preparation as a use case. If you turn off "Reschedule cards based on my answers", it is pure cramming, and the cards return to their old schedule afterwards. A higher desired retention under FSRS shortens the intervals but increases the workload. The manual recommends values below 97%. And importantly: do not rate forgotten cards as Hard to push intervals down. That distorts your data.

In checkst there is the exam mode for this: you set the exam date, and checkst schedules repetitions up to that date at short intervals of hours, without permanently changing your cards' normal intervals. After the exam, your deck simply continues on the normal FSRS rhythm. There is also Custom Study, a widget with your due cards, and a streak that catches you on the days the plan would otherwise slip.

There is, by the way, no controlled study showing that a particular setting or filter strategy is optimal for exam dates. The principles underneath, spreading out, retrieving, and sleeping, are well supported, however. Start with those.