How do you end a conversation politely?

Almost nobody ends a conversation when they wanted to. Not you — nobody. It has been counted.

What was counted

Researchers took 932 conversations: some between friends and family, some between strangers brought into a laboratory and left to talk for as long as they liked. Afterwards each person was asked, separately, when they had first been ready to stop.

Both people got the ending they wanted in 1.59% of conversations. Fewer than two in a hundred.

That number gets quoted a lot, usually as a piece of gloomy trivia. The half that gets dropped is the direction, and the direction is the whole point of this page. In 46.83% of those conversations, both people had wanted to stop earlier than they did. In 9.52%, both had wanted to keep going. The typical gap between the length people wanted and the length they got was about a third of the conversation.

Nearly half of all conversations are two people separately waiting for the other one to leave.

Why it happens

Not because anybody is rude or oblivious. Because the information that would fix it is the one thing everyone is careful not to share.

Saying "I'd like to stop now" reads as a judgement on the other person, so people don't say it. They also can't infer it, because the other person is hiding it just as carefully, for the same reason. The researchers point out that people were bad at guessing when their partner wanted to stop, and specifically underestimated how far apart the two of them were. Both parties end up negotiating blind, with the only relevant fact deliberately withheld on both sides.

So what leaving actually means

If you have been carrying the idea that ending a conversation is something you do to someone, you can put it down. On the numbers, the person you are talking to has more likely than not been ready to go for a while and is waiting for you to move first.

This matters more than it sounds, and not mainly for the endings. A conversation you cannot get out of is an expensive thing to start, and people who find this hard are usually running that calculation before the first message — not "will this go well" but "and then what, how long am I committed for". Leaving being cheap is what makes starting cheap.

Ending one, in practice

Ours, not the research

The counting above tells you the situation, not what to do about it — nobody has tested ways of leaving. These are our suggestions. The last two at least lean on the numbers; the first two are just what seems to us to follow from them.

Say the true thing, briefly. "I should go — this was good." No manufactured excuse. The excuse is what makes it feel like a lie, and nobody needed it.
Don't apologise for leaving. An apology frames it as a wrong being done. It isn't one, and framing it that way invites the other person to insist you stay.
Don't wait for a natural gap. Roughly half the time you are both already past it. The gap you are waiting for is a coordination problem that neither of you can solve by waiting harder.
Leave while it's still good. The most common regret in the data is that it ran too long, not that it ended too soon.

In text, and here

Everything above was measured out loud and face to face, where walking away is visible and costs something. In text it costs less, and that cuts both ways.

It is easier to leave, which is the good half. The other half is that silence is ambiguous: somebody who stops replying may be bored, or may be on a train, or may have put their phone down and meant to come back. You will not find out which. That ambiguity is the price of the exit being cheap, and it is worth knowing that it applies to the other person reading your silence exactly as much as it applies to you reading theirs.

Which is a reason to send the one line rather than just going quiet — not out of politeness, but because it is the only way either of you learns anything about how the conversation actually went.

Where this comes from

Conversations almost never end when both people wanted them to, and the common failure is by a long way that they run too long — not that somebody left too early.

What the study actually did

932 conversations across two studies. In the first, 806 people described a recent conversation with a friend or family member and said when they had wanted it to end. In the second, 126 pairs of strangers talked in a lab for as long as they liked, then reported separately when they had each been ready to stop. Both people got the ending they wanted in 1.59% of conversations. In 46.83% both had wanted to stop earlier; in 9.52% both had wanted to keep going. The typical gap between the length people wanted and the length they got was about a third of the conversation.

What it doesn't show. Why any of it happens, or that a specific person wanted to leave. The first study relied on memory and on people reporting what they thought their partner wanted, which the authors flag themselves; the second fixed that but was a paid laboratory task where nobody had much reason to keep talking. Both were spoken, face to face. No one has measured this in text, where leaving costs less and says less.

Mastroianni, A. M., Gilbert, D. T., Cooney, G., & Wilson, T. D. (2021). Do conversations end when people want them to? PNAS, 118(10), e2011809118 · 10.1073/pnas.2011809118