How to use active recall without staring at a blank page

Why "What did I just read?" fails

You read a page, close the book, and ask: "What did I just read?" Nothing comes back.

That question is too broad. It is hard to check an answer when you have not decided what you need to remember.

Start with material you understand

Piotr Wozniak's first rule for writing memory prompts is to understand the material before you memorize it. A memorized answer is less useful if you cannot explain what it means. Read Wozniak's rules.

Read one short section first. Explain its main point while the source is open. Then decide what you will need to remember.

Facts still matter. If you have to look up "regression" in every paragraph, you lose track of the explanation. Memorize the terms you need to use the subject.

Ask for one thing at a time

In 2022, I made a note while reading Andreas C. Müller and Sarah Guido's Introduction to Machine Learning with Python. The note separates two kinds of supervised-learning task by the output they predict. Scikit-learn uses the same distinction: classification identifies a category, while regression predicts a continuous-valued attribute.

Here is a condensed paraphrase of the note: "A classification model predicts a label from a predefined set. A regression model predicts a continuous number. Recognizing a language is a classification task. Predicting annual income is a regression task."

"What did I learn about supervised learning?" is too broad. Each of these prompts asks for an answer you can check:

1. What kind of output does a classification model predict?
2. What kind of output does a regression model predict?
3. A model recognizes which language appears in a document. Is that classification or regression? Why?
4. A model predicts annual income. Is that classification or regression? Why?

One small unit per attempt

The last two prompts ask you to explain why each example fits.

Wozniak calls this the minimum-information principle. Retrieve one small unit in each attempt. An explanation can still fit that rule when it asks for one relationship. Split a prompt when an accurate answer requires several unrelated points.

Attempt, check, and repair

Hide the source and answer before looking. Suppose I answer the first two prompts like this: "Classification predicts a category. Regression predicts a number."

The first answer matches the note. The second is incomplete. Regression predicts a continuous number, not every kind of number. I would correct that missing word, then answer the two application prompts.

Correct the part you missed. Do not copy the whole passage as punishment. If you cannot identify the missing part, rewrite the prompt with a clearer expected answer.

Why retrieval is different from rereading

Researchers usually call active recall retrieval practice. You study something, remove the source, and try to produce part of it from memory. A short-answer question, flashcard, blank diagram, or explanation can create that attempt. Rereading keeps the answer visible.

I have a note in my knowledge base called "We tend to confuse familiarity with mastery." A sentence can look obvious on the fourth reading and still be unavailable when an exam asks for it without the page in front of you.

Henry Roediger and Jeffrey Karpicke demonstrated that difference in a 2006 experiment with prose passages. Some students restudied the passages. Others practiced recalling them without feedback. Restudy performed better after five minutes. Retrieval performed better after two days and one week. The students who restudied were also more confident that they would remember. Read the study.

Retrieval can test more than isolated terms. In a 2011 experiment, students used retrieval practice or concept mapping to study science texts. Retrieval practice produced better results on questions that assessed comprehension and inference in that study. Read the study abstract.

Read first to understand the material. Then try to explain it without looking.

Return later

One successful answer does not establish durable memory. Spacing research shows that the useful gap depends on how long the material must be retained. In a study of more than 1,350 people, the best gap changed with the delay before the final test. There was no single interval that was best for every goal. Read the study abstract.

You do not need to calculate that interval by hand. Try again later, without looking at the source first. Active recall describes what you do during the attempt. Spaced repetition describes when the next attempt happens.

A short starting routine

Choose one page or one short section:

1. Read until you can explain the main point with the source open.
2. Write a few questions that each ask for one fact, relationship, explanation, or application.
3. Hide the source and attempt each answer.
4. Check the answer and repair the exact gap.
5. Return on another day and attempt the answer again before looking.

Keep what you need to produce

Do not convert every sentence into a card. Keep the definitions, relationships, steps, and distinctions you will need to produce without help.

If writing prompts is the slow part, you can paste notes into Top of Class, the app I build. With Scholar, Atlas drafts recall cards; you decide which ones are accurate and worth keeping. Top of Class brings saved cards back when they are due.

Related articles

• 8 October 2026

  Anki vs Quizlet: pick by how long you need to remember it

  Anki is built for remembering material for months, Quizlet for quick study of ready-made sets. How each schedules reviews, what each costs, and which to pick.

  Read article