Gossip is a bad habit — talking about people behind their backs is simply wrong
Actually
In research, 'gossip' just means talking about someone who isn't there, and most of it is neutral everyday talk. Whether it helps or harms depends on whether it is true and whether it is meant to hurt, not on whether you do it.
▶ 11-minute animationPaper craft
The evidence
In psychology research, gossip is defined broadly: one person talking to another about an absent third party. Kind words, unkind words and everyday news all count. This entry also uses two everyday words to draw distinctions (our usage, not a research finding): slander, meaning saying bad things about someone on purpose, sometimes made up; and a rumour, meaning a story passed around before anyone has checked it.
How much, and about what: Robbins and Karan (2020) pooled five naturalistic observation studies with 467 people in total. Participants wore a device that sampled roughly 5% to 12% of the sound around them at random over 2 to 5 days, and researchers then coded the clips. The abstract says gossip tended to be neutral and about social information. The specific numbers come from the authors' university news release and interviews (we could not open the full paper): about 52 minutes a day per person on average, which is an estimate from the sample, since about 14% of conversations were about other people and, over 16 waking hours, that comes to nearly an hour; about three quarters neutral, about 15% negative and about 9% positive; and talk about acquaintances about nine times as common as talk about celebrities. The abstract also reports that outgoing people gossiped more, that women gossiped more than men only in the neutral kind, and that younger people gossiped more negatively. The news release says people with lower incomes or less education did not gossip more than anyone else.
The helpful side (laboratory findings): in the Feinberg et al. (2012) experiments, volunteers learned that someone had taken advantage of others in a game. The group with no way to warn anyone saw their heart rates rise a further 1.4 beats a minute or so; the group that could write a note to warn the next player showed no significant change. Note that the difference between the groups was only marginally significant (p = .06), with just 52 people; the clearer result in the paper was self-reported negative feeling, which dropped markedly in the group that could warn and barely moved in the control group. In another experiment in the same paper, 34 of 45 people (76%) were willing to pay out of their own pocket to send the warning. Feinberg et al. (2014) had 216 people play a cooperation game in groups: when they could pass messages about each other, average contributions (out of 60) rose from 17.5 to 29.8; add the option to vote selfish players out of the group and it rose to 42.9; those voted out cooperated about as much as everyone else when they returned. Jolly and Chang (2021), a live-interaction experiment (we read only the abstract), report that gossip helped people learn indirectly who could be trusted, and that people who talked formed more similar impressions and grew closer. All of this is from games and lab settings; how large the effect is in real life cannot be read off directly.
The other side: gossip is also powerful, and it can be untrue. In the experiment by Sommerfeld et al. (2007), 126 German university students could write one comment about another player each round, passed to the next person who dealt with them. Most comments fairly reflected how the person had played. But when people had already seen for themselves how someone behaved, and a pre-written comment was added, 44% changed their decision, most of them in the direction of the comment; the authors say gossip has a strong manipulative potential. In the Dutch experiment by Dores Cruz et al. (2024), when the person passing on gossip had interests tied to the person being discussed, the gossip was more likely to be false, describing uncooperative targets as cooperative; this was not found when the sender's interests were tied to the listener's. Listeners responded the same way to information that was likely true and information that was likely false. A computer simulation by Testori et al. (2022), not an experiment with people, shows that only well-meant gossip promotes cooperation; when lying pays, gossip drags cooperation down. Online, Vosoughi et al. (2018) analysed about 126,000 fact-checked stories on Twitter from 2006 to 2017: after controlling for account characteristics, false news was about 70% more likely to be retweeted than true news; bot accounts spread true and false news at the same rate, so the difference comes from people being more likely to share false news.
This is correlation, not causation: the systematic review by John et al. (2018) found that children and young people who had been bullied online had an odds ratio of about 2.35 for self-harm compared with others. These studies are observational, mostly one-off questionnaires, and the authors state that no conclusions about causality can be drawn; it may run the other way, with young people who are already struggling more likely to be bullied. 'Cyberbullying' is also far broader than 'passing on gossip', so the two should not simply be equated.
'Gossip is like social grooming' is a hypothesis: Dunbar (2004) proposed that primates keep their relationships going by grooming one another, that grooming can take up at most about a fifth of waking time, and that as groups grew larger there was not enough time, so language may have taken over; he also reports that about two thirds of everyday conversation time is spent on social topics. This is speculation about the distant past that cannot be tested directly, and it is not settled. A 2023 study by Rudnicki et al. in Scientific Reports had pairs of university friends talk about gossip or a control topic after a stressor; the gossip and control conversations did not differ in cortisol or beta-endorphins, and the authors say the evidence was not sufficient to support the social-grooming analogy.
As for 'gossip releases oxytocin and relieves stress', there is no settled answer. The only directly measured study we found is small: Brondino et al. (2017) had just 22 female university students, split into two groups for comparison (we read only the abstract, so we do not know the details of the assay). Oxytocin rose more during gossip than in the emotional non-gossip conversation, but the stress hormone cortisol fell in all three kinds of conversation, by similar amounts. We found no study that replicated it, and measuring oxytocin in saliva is itself debated (McCullough et al., 2013). Rudnicki et al. (2023) did not measure oxytocin, so they say nothing about that point.
Sources as of November 2026. If newer evidence turns up, tell us and we'll check and update.
How it spread
In everyday speech, 'gossip' often gets lumped together with slander and rumour, so any talk about other people takes on a bad flavour; old sayings keep pointing at this too: Xunzi's 'Da Lue' quotes the proverb 'rumours stop at the wise', and Zengguang Xishi Xianwen has 'the one who brings you tales is the one making the trouble'. Meanwhile the opposite claim circulates online: 'gossip releases oxytocin and relieves stress'. Both go too far: one treats all talk about others as bad, the other treats a 22-person study as settled.
What to do instead
The dividing line is really just two questions: is it true, and is it meant to harm? The three rules below are our suggestions, not research conclusions:
1. If it isn't checked, don't pass it on.
2. Other people's private matters, such as their health, relationships or family difficulties, aren't material for small talk.
3. If you wouldn't say it to their face, don't say it behind their back: ask yourself, if they were standing right here, would I still say this?
If you hear something hurtful, you can decline to join in, you can decline to pass it on, and you can say a fair word for the person who isn't there. If you are the one being talked about, clear it up face to face where you can; for abuse or threats online, keep the evidence, tell someone you trust, and you can report it to the platform. And kind words can be said behind someone's back too: in that recording study, positive gossip was less than one in ten.
HarmlessWrong, but believing it costs you nothingLimited research
Something wrong here, a dead source link, or a more accurate framing?Email a correction — this site's credibility comes from being correctable, not from claiming rigor.