Close the source and produce the answer: say it, write it, draw it or code it, then check. It feels worse than rereading and works better after a delay. Match the method to the round: flashcards for facts and trade-offs, blank pages for designs, a blank editor for code, spoken answers for stories.
What is active recall?
Active recall has three parts: a cue, an attempt to answer without the source in front of you, and a check against the source. Rereading your notes on database indexes is passive review. Closing the notes and answering “why can an extra index make writes slower?” in two sentences, then checking, is active recall. The material is the same; the difference is the direction of travel. Review moves information from the page to you, recall makes you move it from memory to the page.
The research literature uses other names for the same thing. Roediger and Karpicke's 2006 paper describes the testing effect: taking a memory test does not only measure what you know, it improves later retention. Dunlosky and colleagues call the technique practice testing. Active recall is also not the same thing as spaced repetition, although the two are usually combined. Recall is what you do inside a study session; spacing decides when the next session happens.
The evidence: what the experiments actually found
The cleanest demonstration is Roediger and Karpicke's second experiment. They gave 180 undergraduates a short prose passage under one of three conditions: four study periods; three study periods and one free-recall test; or one study period and three recall tests, with no feedback on the tests. Half of each group took the final test five minutes later and half a week later. After five minutes, more studying meant more recall. After a week the order reversed, as the table shows. The study-only group had read the passage about 14 times in total against 3.4 times for the repeated-test group, and it was the group most confident that it would remember.
Later work extended the result. Karpicke and Roediger (2008) had students learn foreign vocabulary: once a word had been recalled, studying it again did nothing for delayed recall, while testing it again had a large positive effect, and students' predictions of their performance were uncorrelated with the outcome. Karpicke and Blunt (2011) found that retrieval practice beat elaborative studying with concept maps on science texts, including on inference questions and even when the final test was drawing a concept map. In their 2013 review of ten techniques, Dunlosky and colleagues rated practice testing and distributed practice as high utility, and rated summarization, highlighting, the keyword mnemonic, imagery for text and rereading as low.
Know the limits before you quote this in a study group. The participants in these experiments were students, the materials were prose, vocabulary and science texts, and the outcome was a memory or comprehension test. No experiment here measured coding skill or interview results, and nothing in it says flashcards replace solving problems. What plausibly transfers is the mechanism: producing an answer strengthens your ability to produce it again.
| Condition | After 5 minutes | After 1 week |
|---|---|---|
| Four study periods (SSSS) | 83% | 40% |
| Three study periods, one recall test (SSST) | 78% | 56% |
| One study period, three recall tests (STTT) | 71% | 61% |
- Roediger & Karpicke (2006), full text (Purdue Learning Lab copy)
- Karpicke & Roediger (2008): The Critical Importance of Retrieval for Learning, Science 319
- Karpicke & Blunt (2011): Retrieval Practice Produces More Learning than Elaborative Studying with Concept Mapping, Science 331
- Dunlosky et al. (2013): Improving Students' Learning With Effective Learning Techniques, PSPI 14(1)
How to do active recall: five methods compared
For technical interviews, five methods cover almost everything. Flashcards are the cheapest per item and suit anything you can answer in under a minute. A closed-book explanation means reading a section, closing it and explaining the mechanism aloud. Blank-page recall means reproducing a structure, such as a diagram or a list of steps, on an empty page. Teach-back means explaining a topic to a person, or a recorder, who can ask why. Coding from memory means rebuilding a solution or a query in an empty editor and running it.
All five share the same rules, and breaking any of them turns recall back into review. One more finding is worth knowing when a session goes badly: Kornell, Hays and Bjork found that trying to answer before seeing the answer improved later learning even when the attempt was wrong (their materials were made-up general-knowledge questions and word pairs). A failed attempt followed by a check is not wasted time.
- Attempt first: the source stays closed until you have produced an answer.
- Commit to the answer by saying or writing it; “I basically know this” is not an attempt.
- Check against the source and mark exactly what was missing or wrong.
- Turn each miss into the next cue, and come back to it after a gap of days.
| Method | Effort | What it trains | Use it when |
|---|---|---|---|
| Flashcards | Low per item | Facts, definitions, trade-offs, complexity | Daily, for answers under a minute |
| Closed-book explanation | Medium | Mechanisms and “why” answers, spoken | Right after reading a chapter or doc |
| Blank-page recall | Medium to high | Structure: diagrams, flows, lists | System design and architecture topics |
| Teach-back | High | Order of explanation; finding gaps | Before mock interviews, with a peer |
| Coding from memory | High | Execution, edge cases, speed | Algorithms, SQL and API usage |
Active recall examples: SQL, system design and behavioral stories
SQL. The cue is a five-row employees table in which two rows have a NULL manager_id. Before running anything, write down what COUNT(*), COUNT(manager_id) and COUNT(DISTINCT manager_id) return, and how many rows survive WHERE manager_id <> 1. Then run it. With SQLite 3.51.0 through Python's sqlite3 module the script prints (5, 3, 2) and then (1,): COUNT(column) skips NULLs, and the filter returns one row, not three, because a comparison with NULL is not true. The learning happens in the prediction; running the query is only the check.
System design. Take a blank page and eight minutes: draw the write path and the read path of a URL shortener, and list three numbers you would estimate before choosing storage. Then compare your page with a reference walkthrough and mark every missing component in another color. Those marks are tomorrow's cues. Rereading the walkthrough would have felt smoother and shown you none of the gaps.
Behavioral. Write the question and three keywords on a card, for example “a time you failed: index migration, seven minutes, linter rule”. Tell the two-minute story aloud from those cues, record it, and check the recording against your written version for the numbers and the result. Recall the facts and the structure, not a script; a memorized script breaks at the first follow-up question.
import sqlite3
db = sqlite3.connect(":memory:")
db.executescript("""
CREATE TABLE employees (id INTEGER PRIMARY KEY, name TEXT, manager_id INTEGER);
INSERT INTO employees VALUES
(1, 'Ada', NULL), (2, 'Ben', 1), (3, 'Cy', 1), (4, 'Dee', 2), (5, 'Eli', NULL);
""")
row = db.execute("""
SELECT COUNT(*), COUNT(manager_id), COUNT(DISTINCT manager_id)
FROM employees
""").fetchone()
print(row)
print(db.execute("SELECT COUNT(*) FROM employees WHERE manager_id <> 1").fetchone())Active recall with AI: using an assistant as a quizzer, and its limits
An AI assistant is useful for the two chores that make self-testing tedious: writing questions and waiting for your answer. Vendors now build for this. OpenAI's announcement of study mode for ChatGPT, dated July 29, 2025, describes it as guiding with questions instead of giving answers outright, and its feature list includes knowledge checks with quizzes and open-ended questions. The announcement said it was rolling out that day to logged-in users on the Free, Plus, Pro and Team plans; plans and availability change, so check ChatGPT's current documentation. A plain chat works too if you set the rules yourself, as in the prompt below.
The limit is correctness. OpenAI's own help center, as of November 14, 2026, says ChatGPT can produce incorrect or misleading outputs, can sound confident when it is wrong, and advises users to verify technical information. A grader that is confidently wrong is worse than no grader, so quiz from your own notes or the official documentation instead of the model's memory, and make it quote the line that supports each correction. Two more rules: never paste confidential code or documents from your employer, and do not let the assistant answer for you. If it produces the explanation and you nod along, you are rereading again.
You are my examiner for a backend interview. Use ONLY the notes I paste below.
Rules:
1. Ask one question at a time. Do not show the answer.
2. Wait for my answer. Then grade it: correct / partly correct / wrong.
3. Quote the line of my notes that supports your grade. If the notes do not
cover the question, say so instead of answering from general knowledge.
4. After each answer, ask one follow-up that changes a constraint.
5. After 8 questions, list the ones I missed as short question prompts.
My notes:
<paste your own notes or the official documentation section here>A 30-minute daily active recall routine
Half an hour a day is enough to keep a preparation plan honest if the time is spent producing answers. Split it into three blocks and rotate topics across the week so each one returns after a few days. Spacing is the other half of the method: the schedule guide explains which gaps to use and how to compress them before an interview date. A spaced-repetition app such as RecallDeck can handle the scheduling for the flashcard block, so the daily question is only which cards are due.
Keep the longer formats on a weekly rhythm. One teach-back or mock interview a week shows whether recall holds while someone interrupts you, and one timed coding set shows whether it holds under a clock. Flashcards, blank pages and recordings prepare those sessions; they do not replace them.
- 10 minutes: due flashcards, answered aloud before flipping.
- 10 minutes: one blank-page or closed-book item, such as a design or a mechanism.
- 10 minutes: one query or function rebuilt in an empty editor and run.
- Last minute: write down today's misses as tomorrow's cues.
Common mistakes that turn recall back into rereading
Most failed attempts at active recall are review in disguise. The usual signs: you flip the card the moment the answer feels close, you choose from multiple-choice options and call it recall, or you “go over” a design by reading it with the page open. Recognition is easier than recall and feels the same from the inside; the study-only group in the 2006 experiment was also the most confident one. The fix is mechanical: produce the full answer in words, on paper or in code before any check.
The second group of mistakes concerns what you test. Cards that ask for a list of eight unrelated items fail for reasons that have nothing to do with understanding; split them. Decks that contain only definitions prepare you for a vocabulary quiz, not an interview, so add cards that ask why, when not to, and what breaks. Finally, recall without feedback hardens errors: always check against a source you trust, and correct the card when the source disagrees with it.
Quick answers
Frequently asked questions
What is active recall?
Active recall is a study method in which you retrieve an answer from memory before checking it, instead of rereading the material. Researchers call it retrieval practice or the testing effect: the act of producing the answer improves how well you can produce it later.
How do you do active recall?
Take a cue, close the source, and produce the answer in full: aloud, on paper or in code. Then check it and note what was missing. Flashcards, closed-book explanations, blank-page recall, teach-back and coding from memory are all versions of that loop for different kinds of material.
Is active recall better than rereading notes?
For retention after a delay, yes in the classic experiments. In Roediger and Karpicke's 2006 study, students who only restudied a prose passage did better five minutes later, but a week later the group that practiced recall remembered 61 percent against 40 percent for the restudy group.
What is the difference between active recall and spaced repetition?
Active recall is how you study inside a session: you retrieve instead of reread. Spaced repetition is when you study: reviews are spread over growing gaps. They are separate techniques that work well together, and flashcard apps combine them by scheduling recall attempts.
Can AI be used for active recall?
Yes, as a question writer and examiner, as long as you answer first and it grades second. Quiz from your own notes or official documentation, and verify corrections, because OpenAI's help center states that ChatGPT can give incorrect answers while sounding confident.
Source notes
References and review policy
Information checked on November 14, 2026. Section links identify sources for factual claims and technical explanations. Interpretations, practice scenarios and preparation recommendations are RecallDeck’s editorial work.
- Roediger & Karpicke (2006): Test-Enhanced Learning, Psychological Science 17(3)
- Roediger & Karpicke (2006), full text (Purdue Learning Lab copy)
- Karpicke & Roediger (2008): The Critical Importance of Retrieval for Learning, Science 319
- Karpicke & Blunt (2011): Retrieval Practice Produces More Learning than Elaborative Studying with Concept Mapping, Science 331
- Dunlosky et al. (2013): Improving Students' Learning With Effective Learning Techniques, PSPI 14(1)
- Kornell, Hays & Bjork (2009): Unsuccessful Retrieval Attempts Enhance Subsequent Learning, JEP:LMC 35(4)
- OpenAI: Introducing study mode (July 29, 2025)
- OpenAI Help Center: Does ChatGPT tell the truth?
From reading to recall
Practice the full interview loop.
RecallDeck schedules the concepts you miss and keeps coding, design, and behavioral fundamentals available when the interviewer changes direction.