Nobody has run a study on retrieval practice and job interviews. We looked. There's a large literature on how people learn and remember things, and a much thinner one on interview coaching, and no one has connected the two directly. So everything below is the learning research plus our own inference about how it applies to interviews, and we'll be clear about which is which.

That caveat matters because the first finding is uncomfortable. The thing that makes you feel prepared and the thing that makes you prepared are not the same thing, and they can move in opposite directions.

Recognition feels like knowing, and it isn't

Read an interview question. "Tell me about a time you disagreed with a manager." You think: yes, I'd talk about the migration project. Move on to the next one.

That feeling of readiness is doing something specific, and it isn't what you think. When you judge how well you know something while the answer is in front of you, you're partly reading the answer's presence as evidence of your own recall. Koriat and Bjork called this foresight bias: what a cue plus its answer brings to mind is not what the cue alone will bring to mind later, when the answer is gone and someone is waiting (Koriat and Bjork, Journal of Experimental Psychology, 2005).

Roediger and Karpicke found the sharp version of this. Students studied prose passages and either took repeated recall tests or restudied the material the same number of times. On a final test five minutes later, restudying won. On tests two days and a week later, testing won by a wide margin, and the authors note that this held "even though repeated studying increased students' confidence in their ability to remember the material" (Roediger and Karpicke, Psychological Science, 2006).

Read the timing there, because it's the whole point. Rereading wins if the test is in five minutes. It loses badly if the test is in a week. Your interview is in a week.

We are not going to tell you this has been proven for interviews, because it hasn't. But the shape matches what people describe after a bad interview more often than any other explanation: they had read the question before, they had a story in mind, and it came out as mush anyway.

Produce the whole answer before you look at a better one

The format of your practice appears to matter more than the volume of it.

Greving and Richter ran a study with 92 participants and found no testing effect at all for multiple-choice questions. It showed up only for short-answer items, where you had to generate the response yourself rather than pick it (Greving and Richter, Frontiers in Psychology, 2018). Picking the right answer from a list and producing it from nothing are different mental acts, and only one of them is the act an interview asks for.

The practical version: say or write the entire answer, start to finish, before you look at any sample answer. Not the bullet points you'd hit. The actual sentences, including the boring connective parts where most answers fall apart. If you can't get through it without stopping, that's the finding, and you've just learned something a list of "top 50 questions" would never have told you.

This is the single easiest thing to get wrong with any practice tool, ours included. Reading a good sample answer is pleasant and feels productive. It's the part of the process that does the least.

Rehearsed answers travel less far than you'd hope

Here's the finding that should change how you practice, and it cuts against the obvious sales pitch for this whole category.

Retrieval practice reliably helps you recall the material you practiced. Whether it helps on new questions is much shakier. Pan and Rickard's meta-analysis of transfer covered 192 effect sizes across 122 experiments and 10,382 participants, and found a moderate overall benefit. But they also reported that once you correct for publication bias, the predicted benefit drops substantially, "often indicating no positive transfer" when certain conditions aren't met (Pan and Rickard, Psychological Bulletin, 2018). What it hinges on most is response congruency: how closely the thing you practiced saying matches the thing you eventually have to say.

Read plainly, that means rehearsing twenty polished answers does not reliably make you better at the twenty-first question. The interviewer almost certainly asks something you didn't rehearse.

So what does carry? Our read, and this is inference rather than citation: practice the shape rather than the script. Structured interviews score answers against a rubric, and they do that because it works. In the largest recent reassessment of selection methods, structured interviews came out at .42, the highest-validity predictor of job performance examined, above cognitive ability tests (Sackett, Zhang, Berry and Lievens, Journal of Applied Psychology, 2022). That paper also revised decades of older validity estimates downward by roughly .10 to .20, so if you've seen the widely quoted 1998 numbers, they're inflated.

A rubric is the useful thing here. What a structured interview scores is roughly the same across questions: did you set up the situation, say what you actually did, and land what came of it. That pattern is general in a way a rehearsed story about the migration project is not. Practicing until the pattern is automatic is a much better bet than practicing until twenty specific answers are memorized.

Space the rounds out, and don't overthink the schedule

Two findings, one useful and one that saves you effort.

Spacing works, and the effect is large. A meta-analysis of spaced retrieval practice reported "a strong benefit of spaced retrieval practice in comparison with massed retrieval practice" (Latimier, Peyre and Ramus, Educational Psychology Review, 2021). Two practice rounds a week apart beat two rounds back to back on a Sunday.

The effort-saving one: in the same analysis, expanding intervals (spacing sessions progressively further apart) showed "no significant difference" from uniform ones, across 54 effect sizes. The elaborate schedules people build are not the active ingredient. Gaps are. One broad qualification from the wider literature is that the ideal gap grows with how far away the test is (Cepeda, Pashler, Vul, Wixted and Rohrer, Psychological Bulletin, 2006), so a month out you can leave a week between rounds, and the week of, tighten up.

The feedback is the part everyone skips

Here's the pattern that wastes most of the value: someone answers five questions, reads the scores, reads the sample answers, thinks "right, that's better than what I said," and closes the tab. That's rereading with extra steps.

Do the slower thing instead. Read the feedback, close it, wait at least a day, and answer the same question again from scratch without looking. The second attempt is the one that counts, and it's also a clean test of the transfer problem above. If your second answer is the sample answer half-remembered, you've memorized a script. If it's your own story told tighter, that's the part that survives contact with a question you didn't expect.

Per-answer feedback matters more than a round score for the same reason. One number tells you how it went. Five specific notes tell you what to change, and only the second kind is actionable tomorrow morning.

Practice the types you're avoiding

People practice what they're already comfortable with. Engineers grind technical questions and skip behavioral ones because "tell me about a time you disagreed with someone" feels like it has no right answer. Managers do the reverse. Both get caught by the half of the interview they avoided.

Simple test: whichever question type you'd rather not do right now is the one to pick. If system design questions make you want to close the laptop, that's the signal, not the excuse.

Same with level. If you're going for a role a step up from what you've done, practicing at your current level mostly confirms what you already know. Set it a notch higher than is comfortable and let the round go badly while it's free to.

A three-week plan that fits around a job

Most people preparing for interviews already have a job, so practice competes with evenings. This is the shape that fits, built on the spacing finding above rather than on anything clever.

Five to seven rounds across three weeks, roughly twenty minutes each. Under three hours total, spread out, which is less than most people spend rereading the same question list and feeling ready.

For what it's worth on ours: interview practice is five questions a round, any job title, five levels from intern to lead, and four question types you can mix. You get one skip per round, a score and written feedback on every answer, and a sample answer to compare against after you've written your own. First round is free with no card. After that a round costs five credits out of the same balance the rest of the tools use, which works out to somewhere between about thirty and sixty cents depending on which package you bought. We wrote up the criteria we hold it to, including where it falls short, in what actually makes an AI mock interview tool useful.

What practice can't fix

Three honest limits.

You need raw material before drilling helps. The testing effect depends on retrieval actually succeeding; practicing recall of something you never learned mostly produces failed attempts. If you can't talk through a system design trade-off at all, reading about system design comes first and practice comes second. Practice sharpens what's there.

The coaching research is thin, and we won't pretend otherwise. The studies on whether interview coaching improves performance are small, mostly paywalled, and more focused on whether coaching damages an interview's validity than on how much it helps. We couldn't verify an effect size we'd be willing to print. Anyone quoting you a precise number for how much interview prep improves outcomes is going beyond what the literature supports.

Practice doesn't fix blanking. If you know the material cold on Tuesday and lose it Wednesday morning, more rounds address the wrong problem. That's a retrieval failure under pressure, which is the gap a live interview copilot is built for. Which kind of help you need depends on which of these is actually going wrong, and we mapped that out in which type of AI interview helper you need.

One more caution about measuring yourself by confidence. An Apna survey of over 10,000 job seekers in India found that people using mock interviews or AI prep tools reported higher confidence than people who didn't, 38% against 24% (Careers360, May 2025). Treat that carefully: it's a single-market survey from a company selling interview preparation, and what it measured was confidence, not performance. Which is exactly the variable Roediger and Karpicke showed can move the wrong way.

Questions people ask about interview practice

How many practice rounds do I actually need? Fewer than people assume, spread wider than people assume. Five to seven over three weeks beats fifteen in a weekend, because the gaps are doing part of the work.

Should I memorize answers to common questions? No, and the transfer research above is the reason. Memorized answers help on the exact question and travel poorly to the one you get. Practice the structure until it's automatic instead.

Is it better to practice out loud or in writing? Either works as long as you produce the complete answer without looking. Out loud is closer to the real thing and catches the sentences that fall apart when spoken. Writing is easier to review afterwards.

Does practicing with AI help more than with a friend? Different strengths. A friend gives you a real human reacting in real time, which AI doesn't. A tool gives you role-matched questions on demand at 11pm and consistent per-answer scoring you can compare across rounds. If you have a willing friend in your field, use both.

When should I stop practicing? When your second attempt at a question you've already done comes out clean without looking. That's the signal the material has moved from recognition to recall, and it's a far better stopping rule than a number of rounds.

The uncomfortable summary is that good practice feels worse than bad practice. Rereading questions is pleasant and produces confidence. Producing full answers cold, getting scored, and doing it again a week later is tedious and produces the thing you actually need. If you want to try a round on that basis, the first one is free.