The pretesting effect: why guessing before you learn works

You open a language app on the train. It shows you llave, four pictures, and a prompt to pick the right one. You have never seen this word. You tap the picture of a door because doors seem key-adjacent, and the app turns it red.

It was a key. You have learned nothing and wasted 4 seconds, or so it feels.

That reaction is close to universal, and it's the reason most textbooks put the quiz at the end of the chapter. Psychologists call the exercise pretesting, and the research keeps finding it beats the sensible-looking alternative.

01 / 06

The guessing group keeps beating the reading group

Pretesting is attempting an answer you can't possibly know, then immediately seeing the correct one. The guess is close to worthless on its own. Almost all of them are wrong, by construction.

The obvious objection writes itself. If you don't know the word, guessing is theater. Skip it, read llave = key, and spend the 4 seconds on a second look.

That's the comparison researchers actually run, and the guessing group keeps winning.

It helps to be precise about what pretesting is not. It isn't a diagnostic quiz to find your level, and it isn't the pop quiz a teacher springs to check whether you did the reading. Both of those measure. Pretesting teaches, and the measurement is a side effect nobody looks at.

The name is also slightly unfortunate. Researchers sometimes call the same thing prequestioning, and the two labels drift depending on whether the material is vocabulary or prose. Same idea underneath: answer first, learn second.

02 / 06

A small effect, but it pointed the same way every time

Tabitha Chua and Steven Pan at the National University of Singapore published the clearest test of this for language learners in March 2026, in Cognitive Research: Principles and Implications. They took the Duolingo exercise itself as the object of study.

341 adults with no Spanish learned word translations across 4 experiments. One group did multiple-choice guessing with feedback. The other read the word-picture pairs. Same material, same exposure.

Two directions got tested separately: Spanish word to picture in experiments 1 and 2, picture to Spanish word in 3 and 4. Then cued recall and multiple-choice tests afterward.

The effect is small to moderate, and it never pointed the other way Chua and Pan ran a pretesting group against a group that simply read the same word–picture pairs. The two effect sizes are ranges across the 4 experiments; across all of them, and both directions, guessing first came out ahead.

341

adults with no prior Spanish, across 4 experiments

40

upper bound of the cued-recall effect, Cohen's d 0.18 to 0.40

67

upper bound on multiple-choice tests, d 0.25 to 0.67

Chua & Pan (2026), Cognitive Research: Principles and Implications. Effect sizes are Cohen's d, reported as ranges across the 4 experiments; the two large numbers are hundredths, 0.40 and 0.67. Small to moderate, and consistent in direction.

Those are honest numbers rather than dramatic ones. A d of 0.18 is small. The interesting part is the sign: across 4 experiments and both directions, guessing first came out ahead, and it never came out behind.

Learners also said they preferred it, especially the word-to-picture version. That's rare. The techniques that win in this literature usually feel worse, which is most of what makes desirable difficulty such a hard sell.

03 / 06

The gain shows up on the words they got wrong

Here's the finding that makes people argue with me at dinner. The benefit shows up on items the learner guessed incorrectly.

Not just the ones they happened to know already. The words they got wrong, saw corrected, and later remembered better than words they had simply read.

Pan put it plainly in the NUS release: attempting an answer first, even when you haven't learned the word, and then seeing the correct response boosts memory. The wrong answer isn't a cost you pay for the occasional lucky hit. It's doing work.

This lines up with older work. Kornell, Hays and Bjork found in 2009 that unsuccessful retrieval attempts improved later memory for the answer, provided feedback followed. Chua and Pan extend it to pictures and to the specific format your language app already uses.

04 / 06

The result is solid; the explanation is still being argued over

Three explanations compete, and they aren't mutually exclusive.

Curiosity. A guess creates a gap you now want closed. You committed to door, and when the answer arrives you're leaning toward it rather than reading past it. Attention gets allocated where it was already pointed.

Error generation. Producing something wrong, then seeing the correction, encodes the correct answer more strongly than passive study does. The mismatch between what you produced and what's true is itself a signal, and your brain treats surprising information as more worth keeping.

Extra routes. Guessing activates a spread of related concepts, including irrelevant ones. When llave arrives it can attach to that whole activated network, which gives you more paths back to it later.

Worth saying plainly: the mechanism isn't settled. Seabrooke and colleagues ran a set of experiments with impossible tests, where the correct answer couldn't be guessed at all, specifically to isolate whether error correction is doing the work. The picture that came back was messier than the clean story above. A related effect in the same family has failed to replicate twice.

So the result is solid and the explanation is still being argued over. Pan and Carpenter's 2023 review runs to book length partly because the field keeps finding the benefit and keeps disagreeing about its source.

That gap matters less than it sounds for a learner. You don't need to know which mechanism is running to press the button.

05 / 06

Pretesting is first contact, not retrieval

These two get conflated constantly, including by people who should know better.

The testing effect is about retrieving something you have already learned. You studied bosque on Tuesday, you pull it back on Friday, and the act of pulling strengthens it. Karpicke and Roediger's Swahili study is the canonical demonstration.

Pretesting happens before any of that. You have never met the word. You guess, you're wrong, you see it, and that sequence prepares the ground better than reading would have.

Same shape, opposite ends of a word’s life in your head Both ask you to commit to an answer and then show you the right one. What separates them is whether you have ever met the word before.

Pretesting llave ?  Before you have learned the word. Guess, fail, see the answer. The failure prepares the encoding.

Testing effect la      After you have learned the word. Retrieve it from memory. The retrieval strengthens what’s there.

Schematic, not measured. It names the two moments rather than plotting them. Pretesting: Chua & Pan 2026. Testing effect: Karpicke & Roediger 2008.

They stack. Pretesting handles first contact, retrieval practice handles everything after, and the thing they share is that a wrong answer followed by the right one beats never having tried.

06 / 06

Guess before the reveal, and read the correction

Most of this is already happening to you if you use an app. What changes is whether you cooperate with it.

• Guess before the reveal, always. Tapping through to see the answer faster removes the only part that was working.
• Commit to the guess out loud or in your head. A vague half-thought isn't a retrieval attempt.
• Stop treating red as failure. On a word you have never seen, red is the mechanism running correctly.
• Look at the correction properly. The benefit lives in the 2 seconds after the mistake, and skipping ahead throws it away.

Before a tutor session, this gives you something better to do than rereading your notes. Cover the translations in last week's vocabulary and guess your way down the list. You'll be wrong a lot, which is the point, and it beats the 20 minutes of cramming you were going to do instead.

It works on the lesson ahead of you too. If your tutor has told you Thursday is about the kitchen, spend 2 minutes on Wednesday guessing the Spanish for pan, sink and drawer. You will get most of them wrong. Thursday's lesson then lands on a set of questions you have already asked yourself, which is a better state to arrive in than a blank one.

One caveat from the research worth carrying: the feedback has to actually come. Guessing with no correction, or with a correction you skim past, is the one version of this that reliably does nothing. Chua and Pan's design gave feedback every time, and the whole literature treats it as load-bearing rather than optional.

The words worth pretesting are the ones about to show up anyway. Guessing at vocabulary you'll never meet again is a way to feel studious for 10 minutes.

Verbamor sits on the other side of this line, and I'd rather be straight about it. Every card is a fill-in-the-blank retrieval from a sentence in your own lesson, which is the testing effect, not pretesting. By the time a word reaches a card you have already met it with your tutor. What the two share is the shape: commit to an answer, grade yourself honestly, see the correction immediately.

If your app makes you guess first, that's not a design flaw to endure. It's the most defensible thing it does.

Which is a good reason to be careful about treating this as a head-to-head. Duolingo and Verbamor solve different problems, and for a lot of people the honest answer is Duolingo. I wrote out where that line falls.

---

Sources

Every load-bearing claim Verbamor makes is traced to its paper on the research page.