Skip to content

Cookies

Preppi stores a few things in your browser so that it works: your login and your choices. We would also like to count visits to each page, anonymously: no name, no account, no network address. That stays off unless you accept. Change it any time with Cookie settings in the footer. What we store

Preppi
Log inSign up

The science behind Preppi

Preppi is built on a few well-studied ways of learning. Each is explained here in plain words, with where the evidence comes from and where it is thin. None of it is a promise about your result.

How to read the numbers

An effect size (written g or d) is a number that shows how much better one group did than another. About 0.2 is small, 0.5 is medium and 0.8 is large. These are what the studies found for their own groups, not what Preppi will do for you.

Answer from memory first

Pulling an answer out of your own memory makes the memory stronger than reading it again. The effort is the point.

What the research found. Across 222 classroom studies with about 48,000 students, quizzing raised achievement by about half a standard deviation (g = 0.50).

In Preppi. Every round asks you to answer first, then shows you why.

Limits. It helps most when you get feedback. With no feedback and fewer than about half of first attempts right, there was almost no benefit (Rowland, 2014). Recalling an answer from scratch helped more than picking from choices. But the NMAT is multiple choice, so choosing is the right method for it.

Based on research: Yang et al. (2021); Rowland (2014). Strength of evidence: Strong.

Come back after a gap

Revisiting something after a gap beats doing it all at once. The best gap gets longer the longer you need to remember.

What the research found. Spaced retrieval beat massed retrieval by g = 0.74 across 39 comparisons (Latimier et al., 2021). The best gap was roughly 20 to 40% of a one-week delay and 5 to 10% of a one-year delay (Cepeda et al., 2008).

In Preppi. A question you miss comes back soon; one you get right comes back later, each time a little further apart, and never after your test day.

Limits. In nine introductory university STEM courses, the benefit was clear in only two. Once calculus was left out, the benefit was not significant overall (Bego et al., 2024). Expanding gaps were no better than steady gaps (Latimier et al., 2021).

Based on research: Latimier et al. (2021); Cepeda et al. (2008); Bego et al. (2024). Strength of evidence: Strong in experiments; mixed in real university courses.

Mix the kinds of question

Mixing different kinds of questions in one session feels harder than doing one kind for an hour, and helps you tell similar things apart.

What the research found. Mixing beat doing one kind at a time by g = 0.42 across 59 studies (Brunmair & Richter, 2019). A classroom trial in mathematics found a large gain (d = 0.83, 54 classes; Rohrer et al., 2020).

In Preppi. A quick round mixes kinds of question, and no two in a row are the same kind in review sets.

Limits. The gain for math tasks was smaller (g = 0.34) and there was no significant effect for plain text (Brunmair & Richter, 2019). It helped most for kinds that look alike and are easy to confuse.

Based on research: Brunmair and Richter (2019); Rohrer et al. (2020). Strength of evidence: Strong for math and pictures.

See it worked out, then fade the help

When a topic is new, a problem with every step shown teaches more than struggling alone. Then the steps are taken away one at a time until you do it yourself.

What the research found. Studying worked examples in mathematics helped by g = 0.48 across 55 studies (Barbieri et al., 2023).

In Preppi. A new math or science topic starts with a fully worked example, then fades step by step to independent problems.

Limits. In these studies, examples that contained mistakes taught less than correct ones. Adding prompts to explain the example lowered the benefit (Barbieri et al., 2023).

Based on research: Barbieri et al. (2023). Strength of evidence: Good.

Know why, straight away

Being told whether you were right helps. Being told why you were right or wrong helps more.

What the research found. Across 435 studies, feedback improved learning by d = 0.48 on average, with very wide differences between studies (Wisniewski et al., 2020).

In Preppi. Every answer shows why the right choice is right and why the choice you made was not.

Limits. What the feedback says shapes its effect strongly (Wisniewski et al., 2020). The best timing is not settled. Giving feedback right away in short drills, and at the end of a section in a full test, is our choice, not a finding.

Based on research: Wisniewski et al. (2020). Strength of evidence: Strong that useful feedback helps; the best timing is not settled.

Knowing how sure you are

Saying how sure you are does not make you learn more. It does show you where you were sure and wrong, which is worth a second look.

What the research found. We did not find a study showing that rating how sure you are makes you learn more, so we do not claim it.

In Preppi. Optional confidence ratings feed your map and the review schedule. They are information for you, not a score.

Limits. The evidence for training this skill is thin. Using the ratings to show you where to look again is our own design choice, not a research finding.

Our own design choice, not a research finding. Strength of evidence: Thin.

Why the tutor works the way it does

Why the tutor will not just give you the answer

In a controlled study of nearly 1,000 high school math students, a chatbot that gave answers raised practice scores. But on a later exam taken without it, those students scored 17% lower than students who never had it. A version that guided with hints and was given teacher-written solutions raised practice scores even more. It largely avoided the harm to the exam (Bastani et al., 2025).

Based on research: Bastani et al. (2025).

Why it names its notes

In an introductory physics course at Harvard (194 students), a tutor taught more in less time than an in-class active-learning session (Kestin et al., 2025). The tutor was built on research-based teaching. It was given correct step-by-step solutions and led students through each problem in order. Preppi's tutor works from stored solutions and study notes, and shows which one it used.

Based on research: Kestin et al. (2025).

What we chose not to do

Sorting you into a learning style

Teaching to a preferred "style" has no reliable support. In a review of 21 studies (Clinton-Lisell & Litzinger, 2024), matching teaching to style gave a small pooled benefit (g = 0.31). But only 26% of the outcome measures showed the pattern the idea requires: each style doing best with its own kind of teaching. So Preppi does not sort you into a style.

Based on research: Clinton-Lisell and Litzinger (2024).

Points and badges as the main idea

Game elements gave a medium gain in knowledge (g = 0.49) but smaller, less stable effects on motivation (g = 0.36) and behavior (g = 0.25) (Sailer & Homner, 2020). So streaks here are forgiving and optional, and a quiet day is just a day.

Based on research: Sailer and Homner (2020).

References

  • Yang, C., Luo, L., Vadillo, M. A., Yu, R., & Shanks, D. R. (2021). Testing (quizzing) boosts classroom learning: A systematic and meta-analytic review. Psychological Bulletin, 147(4), 399–435. https://doi.org/10.1037/bul0000309
  • Rowland, C. A. (2014). The effect of testing versus restudy on retention: A meta-analytic review of the testing effect. Psychological Bulletin, 140(6), 1432–1463. https://doi.org/10.1037/a0037559
  • Latimier, A., Peyre, H., & Ramus, F. (2021). A meta-analytic review of the benefit of spacing out retrieval practice episodes on retention. Educational Psychology Review, 33(3), 959–987. https://doi.org/10.1007/s10648-020-09572-8
  • Cepeda, N. J., Vul, E., Rohrer, D., Wixted, J. T., & Pashler, H. (2008). Spacing effects in learning: A temporal ridgeline of optimal retention. Psychological Science, 19(11), 1095–1102. https://doi.org/10.1111/j.1467-9280.2008.02209.x
  • Bego, C. R., Lyle, K. B., Ralston, P. A. S., Immekus, J. C., Chastain, R. J., Haynes, L. D., Hoyt, L. K., Pigg, R. M., Rabin, S. D., Scobee, M. W., & Starr, T. L. (2024). Single-paper meta-analyses of the effects of spaced retrieval practice in nine introductory STEM courses: Is the glass half full or half empty?. International Journal of STEM Education, 11, Article 9. https://doi.org/10.1186/s40594-024-00468-5
  • Brunmair, M., & Richter, T. (2019). Similarity matters: A meta-analysis of interleaved learning and its moderators. Psychological Bulletin, 145(11), 1029–1052. https://doi.org/10.1037/bul0000209
  • Rohrer, D., Dedrick, R. F., Hartwig, M. K., & Cheung, C.-N. (2020). A randomized controlled trial of interleaved mathematics practice. Journal of Educational Psychology, 112(1), 40–52. https://doi.org/10.1037/edu0000367
  • Barbieri, C. A., Miller-Cotto, D., Clerjuste, S. N., & Chawla, K. (2023). A meta-analysis of the worked examples effect on mathematics performance. Educational Psychology Review, 35(1), Article 11. https://doi.org/10.1007/s10648-023-09745-1
  • Wisniewski, B., Zierer, K., & Hattie, J. (2020). The power of feedback revisited: A meta-analysis of educational feedback research. Frontiers in Psychology, 10, Article 3087. https://doi.org/10.3389/fpsyg.2019.03087
  • Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), Article e2422633122. https://doi.org/10.1073/pnas.2422633122
  • Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). AI tutoring outperforms in-class active learning: An RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, Article 17458. https://doi.org/10.1038/s41598-025-97652-6
  • Clinton-Lisell, V., & Litzinger, C. (2024). Is it really a neuromyth? A meta-analysis of the learning styles matching hypothesis. Frontiers in Psychology, 15, Article 1428732. https://doi.org/10.3389/fpsyg.2024.1428732
  • Sailer, M., & Homner, L. (2020). The gamification of learning: A meta-analysis. Educational Psychology Review, 32(1), 77–112. https://doi.org/10.1007/s10648-019-09498-w

The honest limits

  • Almost none of this research was done on NMAT takers or on adult Filipino learners. We found no study of an AI tutor with that group. Treat it as the best general evidence, not a promise about you.
  • We found no peer-reviewed study that speed training raises scores on multiple-choice admission tests. The pace in Preppi is CEM's own recommended time for each part, so you can feel the clock. We describe it as a way to get used to the clock, not as a proven method.
  • Preppi does not predict your NMAT score or percentile and does not promise any result.

Preppi is independent of CEM: see the Disclaimer. How the AI tutor works and what it can get wrong: AI and voice.

Send feedback

We will see the page you are on and your screen size. Nothing else about you unless you tick the box below.

What is it about?