Active recall vs rereading: the study evidence comparison from WeSolve+

Here is the evidence and the honest place rereading still holds.

The head to head numbers

The direct comparison has been run many times, and one version is famous. In Roediger and Karpicke's 2006 experiments, students who read a passage once and then recalled it outperformed students who read it repeatedly when tested days later, despite the rereaders predicting the opposite about themselves. The follow up 2008 Science study put delayed recall close to 80 percent for practiced retrieval versus 36 percent for rereading. The pattern holds beyond word lists: the 2006 experiments ran on prose passages, the kind of material courses actually assign, and the advantage survived the switch. Same time on task; very different memories a week out.

Why the worse feeling method wins

Rereading feels productive because recognition is easy: the page looks familiar, so the brain reports progress. That report is the fluency illusion, and it is the reason study judgments and study outcomes part ways. Retrieval feels rougher precisely because it exercises the skill the exam grades, producing the answer unaided. The friction is not a cost of the method; the friction is the method.

What rereading is still for

First contact is the honest exception. A chapter you have never met has to be read before it can be recalled, and a source you will quote deserves a careful second read for accuracy. The failure mode is the third pass and beyond, where familiarity keeps rising while delayed recall barely moves. A useful private rule: after the second full read, every further visit to the material should happen through a question.

Converting notes into recall practice

The practical barrier is that retrieval needs prompts and most notes are prose. Two converters fix that. Handwritten or typed notes can become question and answer pairs automatically: upload the PDF or a photo and the quiz pipeline writes gradeable questions from it, while the card maker turns definition shaped lines into decks. On numbers, a plain statement: no usage metric of ours appears in this piece because we have not published any; the percentages are the cited authors' measurements.

Not all retrieval is the same, and the format changes the gain

Retrieval is a category rather than a single act, and its versions differ in what they demand. Producing an answer from a blank page is the most demanding form. Being handed a cue and completing the answer is easier. Choosing the right answer from four options is easier still, because the correct answer is sitting there on the page and the task quietly becomes recognition wearing a test's clothing. The measured benefit tracks that ordering, which matters as soon as you start building your own practice. A multiple choice question you can answer by elimination has exercised something, but not the thing an essay will ask for. This is the main reason a deck where you say the answer aloud before flipping does more than a quiz you click through at speed. The rule that follows is short. If the exam asks you to produce, practise producing, and keep the recognition formats for your first pass through unfamiliar material.

Retrieval without feedback leaves part of the gain unclaimed

An attempt that fails and is never corrected is a half finished repetition. The attempt itself does real work, because searching memory for something strengthens the path whether or not the search succeeds. But the item you could not produce is precisely the item that needs the correct answer supplied, and supplied while the failed search is still fresh. Checking immediately is what turns a missed card into a learning event rather than a rehearsal of the gap. This changes how a good session feels from the inside. Getting things wrong is not a sign the session is going badly, provided every wrong answer is followed by the right one within a few seconds. The pattern to avoid is the familiar one where you skip the cards you cannot answer, promise to look them up later, and never do, because that removes from the deck exactly the items with the most to gain.

Where the effect gets smaller, and where it is not established

Three boundaries belong here, because the numbers above were measured under specific conditions and do not stretch everywhere. The experiments used prose passages and word lists tested after set delays, which resembles a vocabulary list or a history chapter more than it resembles a proof or a design brief. For material that has to be applied to a problem never seen before, retrieval practice appears to help, but the size of that benefit is not settled by the studies quoted here, and a precise figure for transfer would go past the evidence. The second boundary is comprehension. Retrieval strengthens what is already understood and does not manufacture understanding, so a card you can recite without knowing what it means is a card that will not travel. The third is the delay itself. The advantage over rereading is measured on delayed tests, and on an immediate test taken minutes after study rereading can look equal or even stronger, which is one reason the illusion survives so comfortably.

The prediction gap is the most reproducible part

The finding that repeats most reliably in this literature is not the recall gap but the gap in what students expected. In the original comparisons the rereaders rated their own future performance above the retrievers, and then scored below them. Both groups were wrong about themselves, in the same direction. The reason is that we judge our own learning from how the material feels while we are in front of it, and rereading makes it feel smooth. Fluency is a genuine signal, but it is a signal about recognition, and recognition is freely available while the page is open. It says very little about what will still be there on Friday with the page closed. This is why the case for retrieval cannot be settled by trying it once and consulting your impression afterwards. Your impression is the specific thing these experiments showed to be unreliable. The only check that works is a delayed test, which happens to be the only check the exam itself performs.

What this means for the questions you build

Put the findings together and the design of a practice set falls out of them. Favour prompts that make you produce a sentence rather than pick an option. Check the answer at once, particularly on the items you missed. Then return to the missed items sooner than the ones you got right, which is the point where retrieval and spacing stop being two separate habits and become a single schedule. The interval half of that question has its own page: how long to wait before reviewing works through what the gap should be for a test next week and for a subject you need next year.

A week of practice, laid side by side

Run the comparison on yourself. Split a chapter in half; reread one half on three days, recall the other half through generated questions on the same days, then test both halves cold at the week's end. Most students find they can answer the recalled half in full sentences, while the reread half comes back only as vague recognition, which is the lab result reproduced at a desk. The follow ups worth reading next are the memorization guide and the forgetting explainer.

By Mert Erarslan (Exam prep and subject routes)

Last updated: 2026-08-15 · Written with AI assistance and reviewed before publishing.