The Research Behind Flashcard Buddy

Flashcard Buddy isn't built on trends. It's built on decades of cognitive science research -- principles tested in classrooms, labs, and large-scale field experiments. Here's the research behind each core feature.

The forgetting curve

In 1885, German psychologist Hermann Ebbinghaus memorized lists of nonsense syllables and tracked how quickly he forgot them. What he found -- the forgetting curve -- changed how we think about memory. It decays exponentially. Without review, you lose roughly 50% of new information within an hour and 70% within 24 hours.

But he also found something hopeful: each time you review material at the right moment, the curve flattens. The memory sticks around longer. This is the foundation of spaced repetition, and it's the core of how Flashcard Buddy schedules your reviews.

Our algorithm times each card review right before you'd forget it. Cards you struggle with come back sooner. Cards you know well get pushed further out -- up to six months between reviews. You spend your time on what actually needs work.

Active recall beats re-reading (by a lot)

You probably studied for exams by reading your notes over and over. So did most of us. Turns out that's one of the least effective study strategies.

In 2008, Jeffrey Karpicke and Henry Roediger published a study in Science comparing repeated reading to retrieval practice -- basically, testing yourself. Students who tested themselves remembered 80% of the material a week later. Students who only re-read? 36%.

This is called the testing effect, and it's been replicated hundreds of times since. Pulling information from memory -- even when you get it wrong -- strengthens the neural pathways that encode that knowledge.

Every study mode in Flashcard Buddy is built around this idea. Flashcard review asks you to produce the answer before flipping the card. Learn Mode gives you multiple choice and typed answers. Practice Tests simulate exam conditions. Write Mode makes you type the answer from memory. You're always retrieving, never just reading.

Why struggling is good for you

Cognitive psychologist Robert Bjork introduced the concept of "desirable difficulty" in 1994. His research showed that conditions making initial performance harder -- spacing out practice, interleaving topics, testing yourself -- actually produce stronger long-term retention.

The key word is "desirable." Too easy and nothing sticks. Too hard and you give up. The sweet spot is material that feels challenging but achievable -- what Vygotsky called the "zone of proximal development."

Spaced repetition creates desirable difficulty by default, showing cards at the edge of your memory. Max Scholar, our AI tutor, takes it a step further. He doesn't give you the answer when you're stuck. He asks guiding questions and drops hints, keeping you in that zone where the material is hard enough to stick.

What a 2025 UPenn study tells us about AI tutoring

One recent study behind our approach comes from 2025 work out of UPenn's Wharton School and Department of Computer Science.

Researchers ran a randomized controlled trial with 770 high school students across 10 schools in Taipei over five months -- much longer than the typical edtech study, which is worth noting because short-term gains often wash out. They wanted to know whether combining an AI chatbot tutor with personalized problem sequencing could actually improve learning.

Here's what they found:

Students who got personalized sequencing (adaptive difficulty based on performance) scored 0.15 standard deviations higher on the final exam -- roughly equivalent to 6 to 9 months of additional schooling
The gains came from increased engagement, not from completing more problems or encountering harder material
Beginners benefited the most, with a 0.215 SD improvement (p = 0.012)
Students in the personalized group had higher-quality interactions with the AI tutor (p < 0.001) -- they asked more "why?" and "how?" questions instead of just requesting answers
The intervention helped close the gap between advantaged and disadvantaged students, with larger gains in lower-tier schools

Two findings here matter a lot for how Flashcard Buddy works.

First, personalizing the sequence of what students study drove the results -- not just giving them access to an AI chatbot. This is exactly what spaced repetition does. It personalizes the order of card reviews based on your individual performance.

Second, how students used the AI tutor mattered. Explanation-seeking conversations produced significantly better outcomes than answer-seeking ones. This is why Max Scholar is designed to guide rather than hand you the answer. His Socratic approach isn't a gimmick. It's what the data says works.

(Chung, A.T., Zhang, B., Kung, L., Bastani, H., & Bastani, O. (2025). "Effective Personalized AI Tutors via LLM-Guided Reinforcement Learning." SSRN 6423358.)

What a 2025 Harvard trial found about AI tutoring

In 2025, Harvard physics lecturers Greg Kestin and Kelly Miller published a randomized controlled trial in Scientific Reports that compared an AI tutor with an active-learning class. In active learning, students work through problems in small groups instead of listening to a lecture. Many studies show it beats lectures, so the AI tutor faced a strong comparison.

Here is how the trial worked:

194 students in an introductory physics course took two lessons, one on surface tension and one on fluid flow.
Each student learned one lesson at home with the AI tutor and the other lesson in class. The two groups swapped in the second week.
Every student took a short test before and after each lesson.

Here is what they found:

Median scores rose from 2.75 out of 5 before the lesson to 3.5 after the class and 4.5 after the AI tutor. The gain with the AI tutor was more than double the gain from class.
After controlling for prior physics knowledge, the effect size was 0.63 standard deviations. Many students hit the top score on the test, so the authors estimate the true effect at 0.73 to 1.3 standard deviations.
Students spent less time. The median student finished the AI lesson in 49 minutes, and the class ran 60 minutes.
Students rated themselves more engaged (4.1 vs. 3.6 out of 5) and more motivated (3.4 vs. 3.1) after the AI lesson. They enjoyed both lessons about the same.
83% said the AI tutor's explanations were as good as or better than their human instructors' explanations.

The researchers credit the tutor's design for the result. They did not hand students a general chatbot. They wrote the tutor's instructions to keep students working actively, present one small step at a time, and encourage effort. They gave it step-by-step solutions written by the instructors so it would not make mistakes, and the platform walked students through each part of a problem in order. The authors point out that earlier studies, where students used ChatGPT without this structure, found weaker results.

The study has limits. It covered two lessons in one college physics course, with a tutor the researchers built for those lessons. The authors recommend using AI tutoring to prepare students for class and to free up class time for harder work. They do not recommend replacing teachers.

Max is built on the same practices. He starts by finding out what you already know, asks one question at a time, gives hints before a full solution, and checks your understanding after each new idea. Max Live also rechecks every calculation before it goes on the whiteboard. Max did not take part in the Harvard study, but he follows the design choices the researchers credit for their result.

(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, 17458. doi.org/10.1038/s41598-025-97652-6)

How these pieces fit together

Most flashcard apps give you one or two of these. Flashcard Buddy combines all four:

Research principle	How we use it
Forgetting curve / Spaced repetition	SRS algorithm schedules reviews right before you'd forget
Active recall / Testing effect	Study modes built on retrieval practice (flashcards, learn, practice test, write)
Desirable difficulty	Adaptive card scheduling + Max Scholar's guided tutoring
AI-guided personalization	Max Scholar adapts to your level and walks you through concepts

These aren't independent features bolted together. They work as a system: spaced repetition personalizes when you study each card, while Max Scholar personalizes how you understand the material behind it.

That's the difference between a flashcard app built on research and one that just looks nice in a screenshot.

Want to put these principles into practice right now? Try the AI flashcard maker to generate study cards from your notes, or the MCQ generator to create practice tests from any material -- both free, no account needed.