There's a page on this site about what retrieval practice is and what the research does and doesn't support. This isn't that page. This one assumes you're already doing it, or about to, and deals with the part that decides whether it works: the task itself.
The three studies it rests on, Roediger and Karpicke (2006), Karpicke and Blunt (2011), and Dunlosky and colleagues (2013), are set out with their limitations on the retrieval practice page. We haven't added further studies here, because we'd rather keep a short list than cite something we haven't checked ourselves.
Twelve minutes, from the bell
A worked example rather than a principle. Adjust the clock to your own lesson.
Minute 0. The questions are already on the board, or the link is already shared. Nothing about the topic gets said out loud, since a reminder is a free answer.
Minutes 1 to 8. Students answer, books closed. You don't circulate helping. You circulate reading, because eight minutes of watching what people get stuck on is the best diagnostic in the lesson.
Minutes 8 to 10. Answers go up. Students mark their own and write down the ones they got wrong rather than a total.
Minutes 10 to 12. You reteach one thing: the one the most people missed.
Those last two minutes are the ones that get dropped when time runs short, and they're the ones that turn a score into a change.
The seven ways it stops working
1. It's recognition, not retrieval. The answer is visible somewhere, on a display, in an open book or in the question itself. Students feel fluent and remember nothing. Change: close the books, cover the display, and check your own questions for the answer.
2. It became an assessment. Once a score is recorded, a student who's unsure stops attempting, and the attempt is the part that works. Change: stop recording it. If you need a grade, assess separately and be clear which is which.
3. The feedback arrived too late. Answers given the following lesson land on a student who's forgotten what they were thinking. Change: answers in the same lesson, ideally in the same ten minutes.
4. You only ever retrieve last lesson. It becomes a check on whether they were listening rather than on whether they still know it. Change: make half your questions about content from three weeks ago or more.
5. Everyone retrieves what they already know. Given a choice, a class tends to pick the comfortable topic, and the practice reinforces the strong bits. Change: you choose the topic, and choose it from what went wrong last time.
6. It replaced teaching. Retrieval practice strengthens what's already there. It doesn't build something that was never taught properly in the first place. Change: if the class can't retrieve it because they never really had it, that's a reteaching problem rather than a retrieval one.
7. It stopped by November. This is the most common one by some distance. A routine that needs preparing every time gets dropped the first week you're drowning. Change: get the preparation cost close to zero. A pre-made activity you share as a link survives a bad week in a way a hand-written quiz doesn't.
When it is not working, start here
| What you are seeing | Likely cause | What to change |
|---|---|---|
| High scores, poor exam performance | Recognition, not recall | Close the books, remove visible prompts |
| Students not attempting | It is being recorded | Stop recording it |
| Same errors repeating for weeks | Feedback too late, or no next step | Answers in the same lesson, one thing retaught |
| Strong on new work, weak on old | Only ever retrieving recent content | Half the questions from three weeks back |
| Confident class, disappointing mocks | Retrieving only comfortable topics | You pick the topic, from last time's errors |
| Nobody can recall anything | It was never securely taught | Reteach, then retrieve |
| It has quietly stopped | Preparation cost too high | Use something pre-made and shareable |
Running it all term without marking
The workload question decides whether this survives past half term, so it's worth being concrete.
Self-marking in the room. Students mark their own against answers on the board. You read the room rather than the books. This costs you nothing, and it's the version most teachers end up running.
A shared link that checks itself. The activities below are mapped to a specification, and each answer is checked as the student gives it, so the feedback arrives when it's most useful. Nothing comes back to you, and there's nothing to set up beyond sharing the link.
A record the student keeps. Since nothing is reported to you, the record lives with the student. The printable revision record is one week to a page, and reading four of them at the start of a lesson tells you more than a set of marked books.
What still needs you: choosing the topic, deciding what to reteach, and noticing the student whose sheet is empty. Those are the parts worth your time, and nothing automates them.
What this page is not claiming
It doesn't claim a size of effect. No percentages, and no "students who do this improve by", because the honest version of that number depends on which study you pick.
It doesn't claim these seven failure modes are exhaustive or drawn from research. They're the ones that come up, described plainly so you can recognise them.
It doesn't claim retrieval practice is enough on its own. It strengthens what teaching has already built.
Where to go next
For the evidence and its limits, read retrieval practice for GCSE. For specific tasks you can run tomorrow, on screen and off, see ready-to-run revision activities. To put a run of these in front of one class, build a sprint, or browse by subject on the revision pages.