The Pretesting Effect: Quiz Yourself Before the Lecture

The pretesting effect is the finding that guessing at questions about material you have not been taught yet, then being shown the right answers, leaves you remembering more of it than reading the same material would. It is real and it replicates. It is also smaller and stranger than most write-ups admit.

Two papers published in 2026 tested it harder than usual. Between them they support the technique, extend it to adults over sixty and to a language the learner does not speak, and put a question mark over the most repeated claim attached to it, which is that the guessing has to come first.

What is the pretesting effect?

It is the memory benefit of attempting questions before the teaching, and being corrected, instead of simply meeting the material. The name for the wider family is errorful learning, because nearly every answer is wrong on purpose.

Yeray Mera, Nataliya Dianova and Eugenia Marin-Garcia reported three experiments in a 2026 paper in Memory and Cognition. Ninety-three students at the University of the Basque Country learned 108 weakly related Spanish word pairs, of the shape seal and ball. One group guessed the second word before ever seeing the pair, and got three seconds of corrective feedback after each attempt. A second group studied first and was tested afterwards. A control group simply copied the pairs out twice.

The guessing was as hopeless as it sounds. Only 5.08 percent of pretested trials were answered correctly in the first experiment. Those students still ended up recalling about 63 percent of the pairs, against about 53 percent for the group that copied.

One design choice here matters more than the headline, and it is the reason this is a separate idea from simply producing answers. The control group did not read passively. In each study session they were shown the pair and then typed the target word themselves. The authors say why in the methods: typed responses were used to control for the generation effect, so the comparison would isolate testing itself. Producing an answer is covered in the generation effect, and that post is the place for the general principle of attempting before checking. It also warns against generating on material you have never met, which is worth reconciling with this page rather than leaving as two answers to one question. The warning holds for generation on its own, where a failed attempt with nothing to correct it is close to wasted. The case here is different because the correction is part of the design: the payoff Chua and Pan locate is in how a guess sharpens attention to the answer at the moment it arrives, so it is the feedback that rescues the attempt. Guess before the teaching only when the answer follows immediately.

The third experiment took the same procedure to 47 healthy older adults recruited from a university programme for senior citizens, mean age about 67. Pretesting beat copying there too, roughly 60 percent against 49 percent. Comparing that group against the students showed no effect of age and no interaction, which is the closest thing in either paper to a genuinely robust result.

Does it matter that the test comes first?

This is where the popular version of the story falls down. The whole appeal of pretesting is supposed to be the order: guess first, and the reading afterwards lands differently. The second 2026 experiment set out to confirm that and did not.

Adding a copy-test-copy condition to the same design, the authors report that the pretest, post-test, and copy-test-copy groups all improved memory compared to errorless copying, with no difference between the three. Testing was what mattered. When you tested made no measurable difference.

Three things stop that hardening into a rule, and the authors raise the first one themselves. Their final test came about five minutes after learning, and their discussion says that the difference between the pretest and post-test may emerge when delaying the time between the initial and final test, because the benefits of retrieval practice are known to grow with delay. They point to an earlier study that used a one-week gap and found the post-test advantage without the pretest one. A student revising for an exam next Thursday is in exactly the condition nobody here tested.

Second, the groups are small. A power analysis set the floor at 29 people per group. The first two experiments ran about 31, and the third ran 24 and 23, below the floor the authors set themselves. A null from thirty people per cell is weak evidence of absence, which is a different claim from evidence of no difference.

Third, the abstract is slightly kinder to pretesting than the results table is. In the second experiment the gap between the pretest group and the copy group came out at p equal to .04, against a corrected threshold of .0125, so that particular comparison did not survive correction. Pretesting beating copying held up cleanly in the first and third experiments and went wobbly in the middle one.

The defensible reading is narrow: at a five-minute delay, one 2026 study found no advantage for going first, and its authors expect the picture may differ at longer delays. Testing beat copying in every arrangement they tried. That is a much better-supported claim than anything about sequence, and it is the one worth acting on. If you want the version of this argument that is about spacing the tests out, active recall versus spaced repetition covers the scheduling side.

How big is the benefit?

Modest. Worth having, not worth rearranging your week around.

Tabitha Chua and Steven Pan ran four experiments at the National University of Singapore, published in Cognitive Research: Principles and Implications in 2026. Adults recruited online learned 36 Spanish nouns through the guess-with-feedback exercise that Duolingo and Rosetta Stone are built on: see a word, pick the matching image from four, get the answer. The last two experiments ran it the other way round, showing an image and asking for the word. The read-only condition just studied the pairing.

Relative to reading, the paper reports that pretesting yielded statistically significant performance improvements on subsequent cued recall, with a Cohen d between 0.18 and 0.40, and, in most cases, multiple-choice tests, with a Cohen d between 0.25 and 0.67. Carry the hedge in that sentence, because it is doing real work. Multiple-choice results were significant in all cases except Experiment 3, where the authors suspect a ceiling effect.

Put in plain numbers, the first experiment had pretested words recalled 41 percent of the time against 31 percent for read words. That is ten percentage points on a test taken five minutes later, on a list of 36 words, roughly four in ten remembered instead of three in ten. A real gain from a technique that costs almost nothing. Nothing like the transformation the technique gets sold as.

Does it work on a subject you know nothing about?

This was the live question, and it is the one that makes the vocabulary paper interesting.

The older literature suggested no. Benefits on recall tests showed up reliably when the cue and the answer were already associated in the learner head, and were typically absent when learners lack prior semantic knowledge or familiarity with their constituent terms. Which would make pretesting useless in precisely the situation you would most want it: the night before a lecture on something you have never studied.

Chua and Pan aimed straight at that boundary. Their participants were screened for having no Spanish, and the first experiment excluded twenty people for prior Spanish knowledge, failed comprehension checks or not finishing. Guessing with feedback still beat reading on cued recall in all four experiments. Their conclusion is that the mechanisms contributing to pretesting effects can operate even when semantic knowledge is minimal, and the explanation they favour is that a guess sharpens attention to the correct answer at the moment it appears.

Two honest qualifications. Their learners guessed correctly 35 to 38 percent of the time on a four-option question, which the authors flag as a sign that some words may have been more familiar than the screening caught. They also note that learning gains appeared for incorrectly guessed items too, which is what keeps the finding standing. And the two conditions did not cost the same: a pretesting trial ran 8 seconds of guessing plus 5 seconds of feedback against 5 seconds of reading. What the paper equated was the time the word and its picture were on screen together, and it says so plainly, adding that giving readers extra time with both can "attenuate" the advantages of pretesting without reversing them.

For a student, the boundary matters less than it sounds. A chapter of cell biology in your own language is an easier case than Spanish nouns you have never met. You know every word in the questions, even when you do not yet know a single answer.

Why you do not already do this

Because it feels like failing. This is the part of the research with the clearest practical payoff, and the two papers disagree in a way that turns out to be useful.

The Memory and Cognition study found that participants across the experiments consistently underestimated the efficacy of testing, revealing a gap in metacognitive awareness. One pretested group predicted they had scored about 54 percent when they had actually scored about 72 percent. Asked directly whether correct answers or errors were better for their learning, they said correct answers. The people who had just benefited from making mistakes rated mistakes as unhelpful. That pattern is the classic evidence for desirable difficulties, which is the broader idea that the study conditions which feel smoothest are often the ones teaching you least.

The vocabulary study found the opposite. Its participants viewed pretesting as more effective for learning than reading and preferred to use it in the future, and did so particularly for the word and image pairs, with the same trend in the reverse format falling short of significance. The authors offer two explanations and neither is settled. The first is that learners may read the benefit accurately when pictures are involved. The second is that their learners guessed from four options and got roughly a third right, and prior work shows that learners enjoy being tested when they manage to answer correctly. Note which one the numbers favour: guessing accuracy sat between 35 and 38 percent in all four experiments, including the two where the preference did not reach significance, so the format split tracks the first explanation better than the hit rate does.

That gives a usable rule that neither paper states outright. Whether you keep pretesting depends on your hit rate. Open-ended guessing where you miss 95 percent is the version you will abandon after one session. A multiple-choice pretest where you get a third right is the version you will still be running in March. Both produce the effect. Only one survives contact with a real student in a bad mood.

How do you run a pretest before a lecture?

Ten minutes, four steps, and the third one is the step everybody skips.

Some practical notes on each. Make the questions multiple choice if you want the habit to last, for the hit-rate reason above. Keep the set small, since both papers used short sessions and neither tested a marathon. Do not treat the pretest as a replacement for studying: in every condition of both papers, participants still met the material properly afterwards, and the pretest was an addition to that rather than a substitute for it.

If this sounds like a study method you have already been taught, it partly is. The Question step in the SQ3R method asks you to turn headings into questions before reading, which is a pretest without the corrective feedback. Adding the feedback is what brings it in line with the evidence. The same trick works on a timetabled lecture: skim the slide titles, write questions from them, guess, check, then go and take your notes in the lecture with the gaps already identified.

What the research does not show

Four limits, and they are large enough that overselling this would be easy.

Both 2026 papers tested memory about five minutes after learning. Neither tells you anything about next week, and the Memory and Cognition authors are explicit that a longer gap could change the comparison between pretesting and post-testing. Almost every claim you will read about pretesting and exam performance is an extrapolation across that gap.

The materials are word pairs and vocabulary lists, not chapters. The authors call for future work on the influence of material complexity by moving beyond word pairs to more ecologically valid materials, which is a fair description of what is missing. Pretesting has been demonstrated on factual passages and video lectures elsewhere, so the direction is not in doubt. The size of the benefit on your actual syllabus is not something anyone has measured.

The comparison is always against copying or reading. That is a low bar, and it is not the bar you care about. No experiment here compared pretesting against a well-organised revision schedule, so nothing in either paper says you should swap one for the other.

And the samples are modest: around thirty per group in the first study, with a post-hoc power of 0.74 for the older-adult experiment, which the authors themselves list as a limitation. Treat the older-adult result as promising instead of settled, particularly since those participants were recruited from a university programme and may be more cognitively active than the general population of that age.

Where GeniusPal fits

The practical obstacle to pretesting is never the theory. It is that you need ten questions about a chapter nobody has taught you yet, and writing questions about material you have not read is a genuinely hard thing to do at nine in the evening.

That is the specific step GeniusPal shortens. Upload the chapter and it generates questions from it, so the questions exist before you have read a word. Quiz mode then holds the answer back until you commit to one, which is the shape both papers used, and it marks you and shows the correct answer immediately afterwards. The feedback half is not a detail here. Every experiment behind this effect supplied the right answer within seconds, and a guess you never check is the one version of this that the research does not support. Written recall works the same way on the paid plans, withholding the answer until you have typed yours.

On the free plan you get two study-set generations for the account lifetime, not two a month, which stretches to one chapter and a second attempt at it. A free set runs to ten questions and paid sets go up to thirty, and Free allows two full sittings of each set you make. Quiz mode costs nothing on any plan, as does the daily review, which is assembled from whatever you have missed across all your sets. Student is 14.99 dollars a month. Genius is 59.99 dollars a year, displayed as 5.00 dollars a month.

What it cannot do is the part this post has been careful about. It has no idea what your lecturer intends to cover and it cannot tell you what will be on your exam. It cannot read an image-only file, so a photographed page will be refused rather than turned into questions. And it cannot settle the question the research has not settled either: whether the quiz works better before the reading or after it. On the current evidence, at a short delay, that choice did not matter and taking the quiz at all did.