Direct Answer: A judgment of learning is a learner’s prediction about how likely they are to remember or use something later. It is not memory itself. Learners build these predictions from cues: how familiar the material feels, how easily it was processed, how recently it was seen, whether the answer came to mind quickly, whether the topic seems easy, and how confident they feel. Some of these cues are useful; others can be misleading. Strong judgments of learning therefore have to be calibrated against later performance. The learner predicts, closes the material, retrieves after a delay, compares prediction with outcome, identifies which cues were misleading, and adjusts future study. The educational goal is not perfect confidence. It is confidence that becomes increasingly answerable to evidence.
HOW LEARNING WORKS · JUDGMENTS OF LEARNING
Feeling ready is useful information. It is not a score.
A learner becomes better calibrated when predictions of future performance are repeatedly tested against what actually happens later.
The simplest definition
A judgment of learning, often abbreviated JOL, is a metacognitive prediction about future memory or learning performance.
A student might study a vocabulary item and estimate, “I am 80% sure I will remember this tomorrow.” They might finish a Mathematics topic and say, “I think I could solve this without the worked example.” They might read a Science explanation and predict, “I will probably remember the mechanism but not the exceptions.”
The prediction matters because it can influence what the learner studies next. If the judgement is inaccurate, study time can be allocated to the wrong material.
The judgment-of-learning mechanism
STUDY → CUES ARE EXPERIENCED → LEARNER INFERS FUTURE SUCCESS → JOL IS FORMED → STUDY PRIORITY CHANGES → DELAY → RETRIEVAL / PERFORMANCE → PREDICTION IS COMPARED WITH OUTCOME → CUE RELIABILITY IS UPDATED → NEXT JOL IMPROVES
The important word is infers. Learners cannot directly inspect tomorrow’s memory. They estimate it from what the present learning experience feels like.
1. A judgment is an inference, not an internal measurement
There is no mental gauge that directly displays future recall.
Recent reviews describe JOLs as inferential judgements constructed from multiple cues that may or may not diagnose future performance accurately. A 2026 review in Psychonomic Bulletin & Review emphasises this cue-based nature of JOLs and distinguishes predictions made about oneself from predictions made about others. Wei, Soderstrom and Meade: Making judgments of learning for oneself versus others.
This matters educationally because the learner can improve the cues they use. If “this looks familiar” repeatedly predicts failure, familiarity should lose authority.
2. Fluency is one of the most persuasive cues
Material that is easy to read or process often feels well learned.
But processing fluency can come from the page rather than the learner. A model answer looks obvious while it is visible. A definition seems easy after the fifth rereading. A familiar graph feels understood because its shape has been seen many times.
Fluency should trigger a test, not a declaration of mastery.
3. Familiarity and retrievability are different cues
Recognition answers: “Have I seen this before?” Retrieval answers: “Can I produce it when the answer is not visible?”
A learner can have high familiarity and low retrievability. This is common during rereading. The page feels known because nothing looks surprising, yet the learner cannot reconstruct the argument after closing the book.
JOLs become more useful when they draw on actual retrieval rather than visibility alone.
4. Calibration asks whether confidence matches level of performance
Suppose a student predicts 80% recall and later recalls 50%. They were overconfident. If they predicted 40% and recalled 70%, they were underconfident.
This absolute match between prediction and performance is one form of calibration.
Perfect calibration is not required for useful learning. The practical goal is to reduce systematic error enough that confidence helps allocate study sensibly.
5. Resolution asks whether confidence ranks stronger and weaker items correctly
A learner may predict all items too optimistically while still knowing which ones are relatively weaker.
If the learner assigns higher JOLs to the items they later remember and lower JOLs to the items they later forget, their relative discrimination is useful even if the absolute percentages are inaccurate.
This distinction is valuable for study planning: relative ranking can tell the learner where to spend time even before confidence becomes perfectly calibrated.
6. Immediate judgments can borrow too much from the present study state
Immediately after reading an answer, the answer is still active.
A learner who predicts future memory at that moment may partly be judging current accessibility. A later test asks a different question: can the knowledge be recovered after the study context has faded?
Where practical, separate the prediction from the richest support. Close the page, wait briefly, attempt retrieval, then judge how stable the knowledge appears.
7. Delayed retrieval gives a better cue than repeated exposure
A simple study move is to retrieve before rating confidence.
For vocabulary, cover the definition and produce it. For Science, reconstruct the causal chain. For Mathematics, perform the first two steps without the worked example. For English, state the main claim and evidence without looking back.
The learner now has evidence about accessibility rather than only a feeling about familiarity.
8. JOLs can themselves change memory
Making a judgment is not always psychologically neutral.
A 2025 meta-analysis in Psychological Bulletin examined 344 effect sizes from 175 independent experiments reported across 49 records, involving 15,079 adults. On average, making immediate JOLs was associated with a small positive change in later memory, but the size and even direction of the effect varied across conditions; related and unrelated word pairs showed different patterns, and the authors also found indications of publication bias. Ingendahl, Halamish and Undorf (2025).
The educational boundary is important: asking students for confidence ratings can alter the learning episode, but this does not justify treating JOLs as a universal memory intervention. Their primary classroom job here is monitoring and calibration.
9. Recent research strengthens the reactivity warning
A 2026 open-access study reported that JOLs could modify memory even when made covertly or with a non-numerical scale, adding to the evidence that the act of judging can influence later performance. Halamish, Meer and Undorf (2026).
This does not make JOL data unusable. It means the measurement is part of the learning event and should be interpreted accordingly.
10. A percentage can look precise without being accurate
“72% sure” is not automatically more scientific than “high confidence.”
Numbers can help track calibration over time, but learners should not confuse numerical precision with predictive validity. A three-level scale—low, medium, high—can be sufficient for classroom use if it leads to later comparison with performance.
The important operation is prediction → outcome → recalibration.
11. JOLs should control study allocation—but not mechanically
Low-confidence material often deserves more attention, but priority also depends on importance, prerequisite status, examination weight, time cost and how quickly the item can be repaired.
A foundational low-confidence concept may deserve immediate repair. A low-confidence rare detail may be less urgent. A high-confidence concept should still be sampled occasionally because confidence can be wrong.
JOL is one input to study planning, not the entire scheduler.
12. Overconfidence wastes study time in one direction
Overconfident learners stop too early.
They reread once, recognise the material and move on. The later test reveals that accessibility was much weaker than expected.
The repair is not “study everything longer.” It is “replace confidence based on exposure with confidence based on retrieval and transfer.”
13. Underconfidence wastes study time in the other direction
Underconfident learners repeatedly study material they can already retrieve.
This can happen when difficult learning feels effortful even though it produces strong later performance. The learner interprets effort as evidence of weakness.
Repeated successful retrieval can recalibrate confidence upward and release study time for genuine gaps.
14. Difficulty is an ambiguous cue
A difficult item may be poorly learned. It may also be difficult because the learner is doing something educationally valuable, such as retrieving after a delay or solving a varied problem without hints.
“That was hard” should therefore be followed by “What did the later performance show?”
The learner needs to distinguish harmful confusion from productive retrieval effort.
15. Mathematics JOLs should be tied to changed problems
A student can feel confident after following one worked example.
Ask for a prediction before a changed problem: “How confident are you that you can choose the method without the chapter label?” Then solve. Compare judgement with method selection, working and checking separately.
This prevents a correct final answer from hiding uncertainty about the route.
16. Science JOLs should separate vocabulary from mechanism
A learner may feel confident because the keywords are familiar.
Ask separate JOLs: Can you define the term? Can you explain the mechanism? Can you predict a changed condition? Can you state what evidence would support the claim?
Confidence should match the capability that matters.
17. English JOLs should be tied to evidence and production
A passage can feel understood because the learner can follow it sentence by sentence.
Ask: “How confident are you that you could state the main idea, justify one inference and explain one language choice with the passage closed?” Then test.
For writing, predict whether a paragraph satisfies purpose, evidence and cohesion criteria before comparing with feedback.
18. AI can inflate judgments of learning
A fluent explanation generated instantly can create a strong feeling of understanding.
Separate product fluency from learner capability. Close the AI output. Reconstruct the explanation. Solve a fresh task. Then judge what remains.
The strongest JOL after AI use is informed by what the learner can now do without the response in view.
19. Confidence labels should remain local, not become identities
“Low confidence on this item” is useful. “I am bad at Mathematics” is not a JOL; it is a broad identity conclusion.
Keep judgements tied to specific knowledge, task conditions and time. Calibration improves through repeated local evidence.
20. The best calibration loop compares prediction with delayed performance
Before the test, predict. After the test, compare.
Do not ask only whether the answer was right. Ask whether confidence was appropriate. A high-confidence error deserves special attention because the monitoring system failed to warn the learner. A low-confidence success deserves attention too because the learner may be wasting future study time.
21. The final receipt is better study allocation
Metacognitive monitoring matters because it changes action.
As JOLs become better calibrated, the learner should spend less time rereading material that only feels familiar, more time on genuine gaps, and less time overstudying material already retrievable.
The prediction system becomes valuable when it improves the next learning decision.
What judgments of learning are not
- A feeling of fluency is not future memory.
- A percentage confidence rating is not an objective measurement.
- High confidence can accompany error.
- Low confidence can accompany successful learning.
- Immediate familiarity is not the same as delayed retrievability.
- Making JOLs can itself affect memory, so ratings are not always neutral observations.
- Confidence should remain tied to a specific task rather than a fixed identity.
A judgment-of-learning diagnostic map
| What adults see | Possible JOL problem | Useful next move |
|---|---|---|
| Student says “I know it” but cannot retrieve | Familiarity driving confidence | Close materials before judging |
| Stops studying too early | Overconfidence | Use delayed retrieval before stopping |
| Repeatedly studies mastered material | Underconfidence | Compare prediction with repeated successful performance |
| Confidence is high only with worked example visible | Support-dependent JOL | Judge again after support is removed |
| Rates whole chapter “fine” but has uneven gaps | Judgement too global | Rate sections or concepts separately |
| High-confidence errors repeat | Monitoring system not learning from outcomes | Log prediction, outcome and misleading cue |
A practical JOL calibration cycle
- Study a bounded item or concept.
- Remove the richest support.
- Predict future performance.
- Record the judgement simply.
- Wait long enough that immediate visibility no longer carries the answer.
- Retrieve, explain or solve.
- Compare judgement with outcome.
- Identify the cue that drove confidence.
- Adjust study allocation.
- Repeat until confidence becomes better calibrated.
For parents
- “How confident are you that you could do this tomorrow without the notes?”
- “What evidence is your confidence based on?”
- “Can we close the page and test one part?”
- “Did your confidence match what happened?”
- “Which topic deserves more study now?”
For students
- Predict before checking.
- Judge after support is reduced.
- Use retrieval as a cue, not familiarity alone.
- Track high-confidence errors.
- Track low-confidence successes too.
- Spend study time according to evidence, importance and prerequisites.
- Let performance update your confidence.
How do we know JOLs are improving?
- High confidence corresponds more often to successful later performance.
- Low confidence identifies genuine weak areas more accurately.
- High-confidence errors become less frequent.
- Underconfident successes are recognised and released from unnecessary study.
- Study allocation becomes more targeted.
- Confidence depends less on rereading fluency.
- Judgements increasingly use retrieval and transfer evidence.
The complete JOL chain
STUDY → EXPERIENCE CUES → PREDICT → REMOVE SUPPORT → DELAY → RETRIEVE → COMPARE → CALIBRATE → REALLOCATE STUDY → PREDICT AGAIN
Research and evidence boundary
Judgments of learning are a mature area of metamemory research. Current reviews describe them as cue-based inferential predictions rather than direct measurements of memory. Research also shows that eliciting JOLs can itself alter later memory performance, with effects varying across materials and experimental conditions. The 2025 meta-analysis by Ingendahl, Halamish and Undorf reported a small positive average reactivity effect across a large experimental literature while also documenting meaningful moderator differences and indications of publication bias. This page therefore treats JOLs primarily as a monitoring-and-calibration mechanism, not a guaranteed intervention for improving memory. Wei, Soderstrom and Meade (2026); Ingendahl, Halamish and Undorf (2025); Halamish, Meer and Undorf (2026).
