Linguists and psychologists have a term for the small machinery people use to get out of a conversation. Closing rituals: the stock line (“It’s been great talking to you”), the invented errand (“I’ve got a thing at noon”), the segue that works like a door (“So anyway”). Those phrases have their own literature.
What no one had studied, Adam Mastroianni and three colleagues argued in a 2021 PNAS paper, was how and when people decide to reach for them.
Their design had a laboratory half, where both people were asked separately, minutes after they stopped talking, and a survey half, where one person recalled a conversation and guessed at the other. Across 932 conversations, the endings landed where nobody had wanted them.
Asked separately, minutes after
Three hundred and sixty-six people came to a Harvard laboratory in pairs, sat at a small table in a quiet conference room, and were told to talk about whatever they liked for more than one minute and less than 45. They were also told they would be staying the full hour either way, so cutting the conversation short bought them nothing. When they finished, they went to separate cubicles and were asked when they had first felt ready to stop, and when they thought the other person had.
Fifty-seven of the 183 pairs were still going when the 45 minutes ran out and the experimenter had to interrupt them. Those pairs, roughly one in three, were dropped from the analysis, which left 252 people. The authors are explicit that with the experimenter ending it there was no way to tell whether those conversations had run long, short, or exactly as long as the pair wanted — so the pairs who talked longest are precisely the ones the numbers below cannot describe. The authors say they re-ran every analysis with those pairs included and the results did not meaningfully change; that comparison sits in a supplement this piece did not open.
In fewer than 2 percent of the remaining conversations, both people had wanted to stop at the same moment and had managed to. Fewer than a third ended when exactly one of them wanted it to. In close to half, both had wanted out earlier than they got out. In nearly one in ten, both had wanted to keep going.
Nor did the pairs split the difference. Conversations did not end at the midpoint between two incompatible wishes, and most did not end anywhere between them. The authors’ own summary: “Although we cannot say when conversations ended, we can say when they didn’t end: They didn’t end when the conversants wanted them to.”
The gap each side misjudged
Each participant in those laboratory pairs had also guessed when their partner wanted to stop. The guesses missed by about 29 percent of the conversation’s length in the typical case, and the average miss was more than twice that — a spread wide enough, judging by the reported variability, to suggest a long tail of very bad guesses. This is the one thing the laboratory arm could do that the survey arm could not: check each person’s guess against what the other person actually said.
People also believed they were standing closer to their partner than they were. On average they thought their own preferred ending and their partner’s were about a third of the conversation apart. The real distance, on average, was more than twice that, and the paper’s phrasing for the mismatch is that the gap was “more than twice as large as the participants and their partners thought it was.”
The numbers do not describe a world of people desperate to escape. About two-thirds said there was a point at which they had felt ready to stop, in the laboratory and in the survey arm alike; the rest wished the conversation had gone on longer, and in the signed averages that wish is the one that wins. What the studies describe is an ending that lands somewhere nobody chose.
The authors did check whether “felt ready to stop” meant anything, in two extra samples run for the purpose. Of those who said they had felt it, 85 percent said they would have preferred the conversation to end at that point, though only 73 percent said it would have been better if it had; in the second sample, 91 percent said the feeling stayed with them for the rest of the conversation.
No other group has run it again
The larger arm of the evidence is memory. Eight hundred and six people recalled their most recent in-person conversation, which for 78 percent of them had happened earlier that day or the day before. The two-thirds who reported a point at which they felt ready to stop placed it to the nearest minute. Four in ten said the conversation had been ended by something external, which the authors note could as easily have been a partner’s polite excuse. The laboratory arm has its own seam: it is three smaller studies pooled and described as one, and the answer formats were not identical across them.
Juliana Schroeder, in a commentary published the same year, put the sharpest question to the idea itself. What does it mean to want a conversation to end? A moment of boredom can switch to interest a sentence later. Half of that objection the paper anticipated: the 91 percent above is its own check for transience, run in a separate sample, and it holds. Her second point is not answered anywhere in the paper — that the person who leaves might sometimes be the one making the mistake, having gained more by staying.
We went looking for an independent replication and did not find one. Five years on, the paper has been cited a few dozen times, Schroeder ran her own count on the shared data and got 15.6 percent — the share of individual participants, across both studies and before the exclusions, whose conversation ended when they wanted it to, and an independent researcher published a conversation-analytic study of how nineteen pairs handled being asked outright whether they wanted to continue. Nobody has run the experiment again. The authors say as much themselves: future research, they write, “will determine whether these results generalize to a wider range of human beings and social situations.” A single unrepeated paper is a finding to hold loosely, and the limits above are the paper’s own, not a reader’s suspicion.
Sustained distress about conversations — feeling consistently unwelcome in them, or consistently unable to leave — is worth taking to a therapist or counsellor.
Why nobody says so
The explanation the authors offer is not that people are bad at conversation. It is that they are busy protecting each other. Saying plainly that you would like to stop talking carries, in their words, a significant risk of offending the other person, so people decline to say it and wait for an ambiguous offramp instead: the seats filling up, the coffee finished, the pause that can be read as an exit.
The conventions behind that, they write, “may be referred to as politeness when they govern behavior toward strangers and as kindness when they govern behavior toward intimates, but in both cases, they lead people to withhold information about when they want to stop talking.”
Withholding is what makes the second finding bite. If you knew the person across from you wanted to keep going, you might well want to keep going too, and the reverse holds for them.
What the study establishes is bounded, and the boundary is worth stating plainly: it documents a coordination failure in two samples, one remembered and one staged. Nearly one conversation in ten is not a general licence to assume the person opposite is hoping you will stay.
Still, that is the figure to carry out of the room. Nearly one conversation in ten there ended with both people wanting more of it, and the authors’ explanation for that is that neither said so. The guess you are making about the person across the table is being made about you at the same moment, and the exit you both take is assembled out of two of them. People assemble it several times a day, in kitchens and doorways and lifts, and mostly leave believing the other one wanted to go.
