How do you know what you don’t know?

In 1966, Roger Brown and David McNeill gave people definitions of uncommon words. Sometimes participants could not retrieve a word but had the distinctive sense that they nearly had it. In some cases, they correctly reported parts of the unavailable word—features of its sound, the number of syllables, or its stress pattern.

Failure to say a word does not necessarily mean having no information about it. The feeling of approaching an answer may reflect accessible fragments rather than an inspection of a fully formed memory waiting out of view.

Joseph Hart’s earlier research examined a related judgment: when people cannot recall an answer, can they predict whether they would recognize it among alternatives? Such feeling-of-knowing judgments can be informative without being infallible.

Thomas Nelson and Louis Narens proposed an influential 1990 framework separating the work of memory from its evaluation. Monitoring supplies information about performance: an answer seems familiar, a detail feels uncertain. Control changes behavior in response: reviewing again, trying a different cue, checking a source, or deciding not to answer.

Feeling unprepared for a test is a monitoring judgment; choosing to revisit the material is a control decision. Useful remembering involves choosing how to handle incomplete access. Detecting a gap only helps if it leads to a suitable next step.

Asher Koriat’s 1993 accessibility model proposes that feelings of knowing partly depend on the relevant information that becomes accessible during a retrieval attempt. If those fragments belong to the right event, the feeling can guide a productive search.

But the same process can mislead. A remembered venue may belong to another meeting, or a familiar face may come from a neighboring experience. An answer that arrives easily can be wrong, while an accurate memory may require effort. Metamemory is not a truth detector; it estimates access and reliability from the clues available at the time.

Robert Hampton’s 2001 experiment showed two rhesus monkeys an image and, after a delay, tested them with that image and three alternatives. A correct choice earned peanuts; an incorrect one brought a timeout. On many trials, the monkeys could decline the test for a smaller guaranteed reward—before seeing the alternatives.

As delays increased, the monkeys declined more optional tests and performed better on tests they chose to take. Under variable-delay conditions, one monkey was approximately 88% accurate on chosen tests versus 67% on compulsory tests; the other’s figures were 78% and 69%.

Controls strengthened the interpretation that memory availability mattered. Still, this was a study of two animals. It supports a functional parallel with human metamemory, not human-like introspective awareness. In this task, having an option to decline helped the monkeys use memory more selectively.

Asher Koriat and Morris Goldsmith showed that allowing people to withhold uncertain answers can improve the accuracy of what they report, usually at the cost of saying less. Accuracy and quantity are different measures.

Reporting where a gathering took place while withholding who arrived first gives a less complete but potentially more trustworthy account. The goal is not maximum silence but a balance between what can be reported and what the evidence supports. What comes to mind? What supports it? What remains unresolved? Those questions lead to different actions: answer, qualify, check, or wait.

The research suggests that the feeling of knowing is useful but imperfect. A tip-of-the-tongue sensation often means the answer is in memory somewhere, so giving it a few minutes or a related cue can be worthwhile. Confidence alone, though, is not proof that a detail is right.

When accuracy matters, such as recalling what a doctor said or who agreed to a plan, it can help to separate what you are sure of from what you are guessing. Saying "I remember the main point but not the exact number" is often more honest and more useful than filling in a confident but shaky detail.

This is the same choice participants faced in the studies on saying less: reporting fewer details can make what you do report more accurate.