The Generation Effect: Why Making Answers Beats Reading

The generation effect is the finding that you remember information better when you produce it yourself than when you simply read it. In the classic experiments, people who completed a word from a clue (rapid – f___) remembered "fast" better than people who just read "rapid – fast."

Psychologists Norman Slamecka and Peter Graf named the effect in 1978 after five experiments in which generated words beat read words on every memory test they tried. A 2007 meta-analysis of 86 studies put the average benefit at about 0.4 standard deviations, a solid effect for something as simple as filling in a blank.

For students, the lesson is to make your study materials make you work: write your own questions, give your own examples, predict, and solve before you look. This guide covers the research, why it works, where it stops working, and six concrete ways to use the generation effect.

What is the generation effect? Examples from the research

In Slamecka and Graf's experiments, everyone studied the same word pairs. The only difference was whether the second word was printed in full or had to be produced from a rule and a first letter. With illustrative pairs, the two conditions looked like this:

They used five rules (associate, category, opposite, synonym, and rhyme) and tested memory in several ways: recognizing the words later, recalling them freely, and recalling them from cues. Generated words won every time.

Outside the lab, you use the generation effect whenever you produce information instead of receiving it:

• Answering "What does the loop of Henle do?" before reading the paragraph that explains it.
• Writing your own example of opportunity cost instead of copying the textbook's.
• Filling in a blank in your notes: "The Treaty of _____ ended World War I with Germany."
• Working a calculus problem before looking at the worked solution.

How strong is the generation effect?

Reliable, and moderate in size.

• Bertsch and colleagues (2007) pooled 445 effect sizes from 86 studies. Generating beat reading by about 0.40 standard deviations. As a rough picture, that moves an average student from the 50th percentile to around the 65th.
• McCurdy and colleagues (2020) analyzed 310 experiments and over 1,600 estimates. A key finding was that how constrained the generation task is changes the size of the effect.

That second point matters for studying. A separate set of experiments by the same team found that fewer constraints produced a bigger generation effect. Producing an answer when several are possible makes you connect it to more of what you know than filling in the one obvious letter does.

So "explain why this drug causes a dry mouth" will usually do more than a blank where only one word fits.

Why does generating help memory?

The leading account, called the two-factor theory, says generating does two things at once.

1. You process the item more carefully. To produce "fast" you have to attend to its meaning, not just its shape on the page. Researchers call this item-specific processing.
2. You connect it to other knowledge. Generating forces you to link the answer to the cue and to related ideas. The more links, the more routes back to the memory later. This is relational processing, and it's the part that grows when the task is less constrained.

Effort is part of the story too. Producing an answer is harder than reading one, and that difficulty is a desirable difficulty as long as you can succeed. As Robert and Elizabeth Bjork put it, looking up an answer you could have generated yourself robs you of a powerful learning opportunity.

Generation effect vs testing effect

The two look similar. Both involve producing an answer instead of reading it. But they aren't the same, and the difference changes how you should study.

• Generation in the lab means producing a word from a clue, with no need to remember anything.
• Retrieval practice (the testing effect) means deliberately pulling out something you studied earlier.

In 2010, Jeffrey Karpicke and Franklin Zaromb gave people the same word fragments with different instructions: "write the first word that fits" (generate) or "use this fragment to recall a word you studied earlier" (recall). Both groups completed about the same number of fragments. Recall still produced better memory.

Completing a fragment beat reading in both groups: that's the generation effect. But completing it while trying to remember the studied word helped more: that's retrieval practice. In their first experiment, with a free-recall test, generating didn't beat reading at all (28% vs 27% of words recalled), while recalling did (38%).

The practical rule: when you fill in a blank or answer a question, try to recall what you learned, not just produce something plausible. That turns generation into active recall, which has the stronger track record.

When doesn't the generation effect work?

It has clear limits.

Meaningless or brand-new material. In four experiments with made-up words, Lutz and colleagues found no generation effect at all. With unfamiliar sentences from textbooks, the effect was much smaller than with familiar phrases. You can only generate from knowledge you have. Read or watch the material once first, then generate.

When you generate wrong answers and never check. A wrong answer you produce and never correct can stick. Always compare with the source and fix errors right away.

Guessing on a pre-test is the exception. Trying to answer questions before you study, then reading the answers, does help, even when the guesses are wrong. In a 2009 study, students who tried to answer questions before reading a passage remembered those concepts better than students who spent the time reading, even for the questions they got wrong. The key is that the correct answer comes right after. See pre-testing for how to use this.

Time. Writing your own questions takes longer than reading someone else's. It's usually worth it for core ideas, less so for trivia you could put straight on a flashcard.

6 ways to use the generation effect when you study

1. Turn your notes into questions

After a lecture, rewrite each heading or key point as a question in the margin: "What are the three causes of inflation in this model?" Cover the notes and answer later. You generate twice: once when you write the question, once when you answer it.

2. Make fill-in-the-blank cards on the key idea

Blank the word that carries the meaning, not a trivial one. "Insulin _____ blood glucose by increasing uptake into muscle and fat cells" tests the core fact. "Insulin lowers blood _____" tests almost nothing. Add a "why?" card for anything you'll need to explain. The flashcard guide covers card wording.

3. Predict before you read

Before starting a chapter, look at the headings and write a one-line guess for each: what it will say, or why. Then read and correct your guesses. Wrong predictions are fine; checking them is the learning.

4. Generate your own examples

For every concept, write one example the book didn't give. Opportunity cost: "Choosing a 6 a.m. shift means giving up the sleep and the gym session I'd have had." If you can't produce an example, you've found a gap in understanding, not just memory. Pairing this with elaborative interrogation, asking why each fact is true, deepens it further.

5. Solve before you look at the solution

Attempt each worked example in the textbook before reading its solution. Even getting halfway gives you a reason to care about the next step. When you do read the solution, explain each step to yourself; the self-explanation strategy shows how.

6. Write your own flashcards, or rewrite generated ones

Writing cards is itself a generation task. If you use pre-made or AI-generated cards, you skip that step. Make some of it back: rewrite card fronts in your own words, delete cards you don't understand, and always say your answer before flipping. If you're turning notes into cards, the notes-to-flashcards workflow keeps the generating step in.

A 20-minute generation routine for one chapter

Use this right after your first read of a chapter or lecture.

1. Questions (5 min). Write one question for each heading or slide group. Make at least half of them "why" or "how" questions.
2. Answers from memory (7 min). Close the material and answer every question in writing. Leave blanks rather than looking.
3. Check and correct (5 min). Compare with the source. Fix errors in a different color. Mark any question you couldn't answer.
4. One example each (3 min). For the two hardest ideas, write an example of your own.

Tomorrow, answer the same questions again, cold, starting with the ones you marked. You've now generated the material, retrieved it twice, and spaced the second attempt: three effects stacked in under half an hour.