How much of a conversation will you remember? What the studies measured

In a 2023 laboratory study, pairs of students talked for ten minutes and were asked a week later, without warning, to recall what had been said. Across the pairs, the conversations contained 7,856 distinct ideas. Only 369 of them, 4.7%, were recalled by both partners. [1]

That figure is a better place to start than the percentages that usually answer "how much of a conversation do people remember?" It comes from a paper you can read, with a stated method. Several of the more familiar numbers cannot be traced that far.

The short answer, with its conditions

When researchers record ordinary conversations and later spring a recall test, people reproduce well under half of what was said. Recent studies that give credit for the gist of an idea report roughly 22–30% after a 20-minute delay. Two studies from the 1980s report about 10%. What people do recall is mostly accurate at first and noticeably less accurate a week later. The one remedy tested in this group of studies, writing an account soon afterwards, protected the ideas that went into the account and nothing else.

These are laboratory results, mostly from US undergraduates at one university talking with strangers or classmates about everyday topics. In the three recent studies nobody was trying to remember, because the test was a surprise; the 1980s studies varied whether participants expected a test. Whether a long friendship or a high-stakes meeting changes the numbers was not tested.

What "remembered" means here

Researchers transcribe the conversation and divide it into idea units, which this literature defines as "the smallest unit of meaning that has informational or affective value." They divide each recall the same way and count matches. [2] An idea counts as recalled when its gist is there, even if the wording differs. [1] In one recent study, a five-minute conversation held about 220 idea units on average. [2]

So "28%" means the share of expressed ideas a person reproduced in substance, after a stated delay, when asked unexpectedly. It does not measure exact quotes, and on its own it does not say which details survived.

[3] [4] [1] [5] [2]

Only the abstracts of the two 1980s papers could be read for this article, so their sample sizes and scoring rules are left out; the seven-minute length comes from the 2023 paper's summary of the 1987 study. [1] Averaging 10% with 28% would blend different decades, procedures and scoring. The 2024 paper summarizes earlier estimates as "between 5% and 20%" of ideas after delays of minutes to days. [5]

What fades first

Accuracy holds, then slips. After 20 minutes, 95.8% of what active participants recalled was accurate. [2] In the note-taking study, written accounts at five minutes were 91% accurate; oral recall a week later was 78% and 76%. [1] The 1987 abstract describes the same drift over a month: less content, more inferences, lower accuracy. [4]

The end of a conversation goes before the beginning. In both recent studies, ideas from earlier in the conversation were more likely to be recalled. [2] [5]

Who said it gets vague. A week out, about 15% of relevant recollections carried no attribution or an ambiguous one, such as "we started talking about taco places." [1]

Overheard talk is thin. A third student sitting about four feet away, busy with a computer chess game, recalled 12.05% of the conversation. [2] A 1989 study found observers who watched a video, heard audio or read a transcript recalled as much as the participants, so divided attention may be what matters. [6]

Your words or theirs: older and newer studies differ. The 1984 and 1987 abstracts say people remembered more of their partner's contributions than their own. [3] [4] Both recent studies found the reverse: 29.80% of their own ideas against 26.53% of their partner's among strangers, [5] and 29.83% for speakers against 22.86% for the people they addressed in the three-person study. [2] Expect to keep more of what you said, but not by a wide margin.

Two people leave with different conversations

Twenty minutes after a five-minute talk, only 11.5% of its ideas had been recalled by both people who took part, and 2.9% by all three people in the room. Individual performance varied widely: the best participant recalled 44.12% of a conversation, the worst 7.37%. [2] A week later, in the other study, the shared portion was the 4.7% above. [1]

The studies did not examine disagreements. Still, the pattern suggests a plainer reading of "that's not what we agreed": each of you may have kept a different small slice of the same talk.

Confidence is a poor guide, too. A week on, the students who had written an account rated their own memory lower than their partners did (3.30 against 4.25 on a 1–7 scale), although they recalled more. [1]

The tested remedy covers only what you write

In that 2023 study, one partner in each pair spent 15 minutes typing a surprise account of the conversation, starting after a five-minute walk. The account held 24% of the conversation's ideas. A week later, the writer recalled 46% of the ideas in the account and 9% of those left out. The partner who wrote nothing recalled 11% of the same left-out ideas. [1]

Writing did not sharpen memory for the conversation in general. It preserved what reached the page, and accuracy was similar for people who wrote and people who did not. [1] The account was long and unplanned, and the sample was 21 pairs.

Figures to stop repeating

• "We forget about half of what we hear straight away." This is usually credited to Ralph Nichols and Leonard Stevens, "Listening to People," Harvard Business Review, September 1957. HBR's public page shows the byline and the opening paragraph, and the figures could not be found in any text we could open. What was measured, and whether it was conversation, is unknown. [7]
• "35% after eight hours." This and similar hour-by-hour figures appeared in search summaries and Q&A answers during research for this article, without a source. We found none.
• "10% of what we read, 20% of what we hear." A historical study traced these "learning pyramid" lists back more than 160 years and concluded they "did not originate from research." [8]
• "About 10% after five minutes." This one is traceable to the 1984 abstract. [3] Quote it with its conditions: a 1984 laboratory study whose abstract says memory goals and recall mode changed how much people recalled.

A three-line note, written within minutes

The studies suggest a note's design. Write it soon, because unwritten detail fades and accuracy slips with time. Name who said each thing, because attribution blurs. Mark any exact words as a quote, and keep your own interpretation out of the record, because inferences grow with delay. Stop at what you will need, because the benefit covers only what you write.

A fictional example with invented people:

Fri 9 Oct, coffee with Ines Moreau

1. Ines offered to introduce me to the editor at Larchmont Press; I said I'd send two writing samples first.
2. Her words: "the list closes at the end of the month."
3. Me: email samples to ines@example.com by Tue 13 Oct.

Treat this format as an untested adaptation. The study's account was a 15-minute typed narrative, and no study here tested whether three lines give the same protection. If the detail you lose most often is the person's name, a separate piece deals with that.

Kluro's Remember takes a note like this, including "something someone said in person," and keeps it beside the connected conversations with the same person. Added context retains its origin: a note you wrote is not treated as the same thing as an imported passage.