attraction.lab

When Friendliness Gets Read as Flirting, and What Can Actually Be Changed

Most people have had it happen, in both directions. In the one experiment that held the behaviour constant, what moved the reading was not the behaviour.

Research-backed writing on attraction, dating and relationships. Drafted by AI against a named bibliography, published only when every claim links to its source.
34 min read
Editorial illustration for: When Friendliness Gets Read as Flirting, and What Can Actually Be Changed

A note before anything else. If you are here because someone treated your friendliness as permission — pushed on after a no, told you afterwards that you had been sending signals, or made it hard to refuse at work — that is not a misunderstanding you caused, and nothing on this page moves any part of it onto you. Find A Helpline (findahelpline.com) lists free, confidential services by country, and in the United States RAINN runs a 24-hour line on 1-800-656-4673. Both are in the sources at the end.

With that said, the question is a reasonable one, it is extremely common, and it has actually been measured.

In the best replication available, 88.3% of women and 70.6% of men said someone had read their friendliness as sexual interest at some point in their lives. Per occasion the rate is far lower: friendly photographs were sorted as friendly about four times in five, and in recorded conversations the absence of flirting was recognised 84% of the time. The error is real, it is not constant, and the odds already run in your favour.

What the research then does is take the dial out of your hands. In the one study that held the described behaviour completely fixed and changed only who was performing it, the rating of sexual intent moved by a full point on the scale. The difference between readers turns out to be sensitivity rather than threshold, which means a quieter version of you gets misread too, in the other direction. And the ability to tell how you come across to one particular person is the weakest form of self-knowledge anybody has measured.

The behaviours you would have to remove — looking, smiling, laughing, standing nearby, talking at length — are the behaviours of ordinary liking in every kind of relationship, and in many settings they are also socially required of you.

What is left is narrower and more useful than a list of things to stop doing: one sentence, said early, in the channel people read accurately. Our editorial policy is why every number here arrives with a link attached.

#1

Most people have been on the receiving end of this, and it has been counted

Start with the thing almost no page on this subject bothers to establish: how often it actually happens.

In 2014 Mons Bendixen ran a direct replication of a 2003 study by Martie Haselton, using 308 heterosexual Norwegian undergraduates — 181 women and 127 men, aged 18 to 30, out of 344 questionnaires returned from 400 handed out. The questions were put in the plainest possible form, and they are worth reading as the question people type rather than as a finding: Have you ever been friendly to someone of the opposite sex only to discover that he had misperceived your friendliness as a sexual come-on? And: Have you ever been in a situation with a member of the opposite sex in which you were just trying to be nice but he assumed you were sexually attracted to him?

88.3% of the women and 70.6% of the men said yes. The mirror question — someone reading your interest as mere friendliness — got a yes from 39.4% of women and 74.6% of men.

Two things follow, and this genre usually keeps only the first. Being misread as interested was more common overall than being misread as uninterested, d = 0.82, and the gap between the two was much wider for women than for men, d = 0.96 for the interaction. But the total amount of being misread did not differ significantly between the sexes, F(1, 306) = 2.92. What differs is the direction, not the frequency.

Bendixen reports the figures from Antonia Abbey's original 1987 survey as 72% of women and 60% of men, with a similar lifetime count of around 4.8 incidents for both.

The limits are real and belong in the same breath. These are retrospective self-reports from students, the questions are framed around opposite-sex encounters only, and nobody in them is describing the same incident from both sides. What they establish is modest and still worth having: if this has happened to you, you are in the majority, repeatedly, in every sample anyone has collected.

#2

Occasion by occasion, friendliness mostly lands as friendliness

A lifetime count makes the problem sound constant. The laboratory measures the per-occasion rate, and it is much lower.

Coreen Farris, Teresa Treat, Richard Viken and Richard McFall showed 280 undergraduates a set of 280 photographs and asked them to sort each one into friendly, sexually interested, sad or rejecting. The photographs had been pre-sorted by an earlier panel, and attractiveness was held constant across the four sets.

The friendly photographs were read as friendly 79.9% of the time by men and 87.0% by women. They were read as sexual interest 12.1% of the time by men and 8.7% by women.

Jeffrey Hall, Chong Xing and Seth Brooks got a compatible number out of live conversation. Fifty-two pairs of strangers talked for ten to twelve minutes and then answered separately whether each of them had flirted. When a partner was not flirting, people recognised that 84% of the time. When a partner was flirting, it was detected in 7 of 25 cases.

Put the two together and the shape of the problem changes. Not flirting is the easy case. It is read correctly about four times in five in photographs and five times in six in conversation, and the famous unreliability of this channel is mostly about missed interest rather than invented interest — the full numbers, and the width of the uncertainty around the small ones, are in what the evidence supports about reading someone's interest.

So the honest answer to how much am I being misread is: less often than it feels, and more often than never. An error rate around one in eight, applied to every warm interaction across years, produces exactly the handful of memorable incidents the surveys count. None of that requires you to be doing anything unusual.

#3

The experiment that held the behaviour constant and changed the reader

Here is the result that settles whose behaviour is being measured, and almost nobody outside the field has read it.

In 2000 Martie Haselton and David Buss ran a study in which 289 participants rated how much sexual intent lay behind described behaviours. At a party, if a woman smiles at a man repeatedly over the course of the evening, what is the likelihood that she wants to have sex with him? Touching an arm on a date. Going out to a bar. The scale ran from extremely unlikely to extremely likely.

Men rated women's sexual intent higher (M = 0.12) than women rated other women (M = −0.34), and far higher than women rated themselves performing the same behaviours (M = −1.69), F(1, 283) = 136.89, p < .001. That much is the standard finding, reported in some form since 1982.

Then the authors did something the standard finding does not require. They gave 131 of the same participants the same items about their own cross-sex sibling. Same described behaviours. Same scale. The only thing that changed was who was performing them.

Men's ratings of their sisters' sexual intent fell to M = −1.03, against M = 0.02 for unrelated women, F(1, 57) = 23.62, p < .001. Women rated their brothers lower than other men too: M = −0.50 against M = 1.37.

Read that slowly, because it is this page in one result. The smile did not change. The arm did not change. The bar did not change. The reading moved by a full point on the scale because of who was doing it, in the mind of the person doing the reading.

The caveats are the usual ones for vignette work: these are ratings of sentences, by students, about hypothetical people, not observations of anyone's evening. What the study establishes is not how often a particular smile gets misread. It is that the quantity being measured was never only in the smile.

#4

What differs between readers is skill, not standards

If much of the variance sits in the reader, the next question is what it consists of, and there is a precise answer that changes the advice.

The folk explanation is a threshold: men are said to need less evidence before calling something sexual interest. Farris and colleagues built their sorting task to separate threshold from sensitivity, and found no evidence that men had the more lenient threshold. What differed was sensitivity — the ability to tell the two categories apart at all.

That distinction has a consequence nobody asking this question expects. A low threshold would mean you could fix the problem by lowering your own signal, because the reader would stop clearing the bar. Low sensitivity means the errors run in both directions: in the same study, men read genuine sexual interest as mere friendliness 37.8% of the time, against 31.9% for women. Turning yourself down moves you from one error towards the other, and a reader who cannot discriminate will misread the quieter version too.

The individual-difference work points the same way. Angela Jacques-Tiura, Antonia Abbey, Michele Parkhill and Tina Zawacki asked 356 male college students how often they had misperceived a woman's sexual intent, and the frequency was predicted by hostile masculinity, an orientation towards impersonal sex, and drinking in dating and sexual situations. It is self-report and cross-sectional, so it cannot show what causes what. It does show that the error is not spread evenly across people, which means the thing you are trying to control is partly a property of who happens to be in the room.

What the research does and does not support about reading interest from behaviour at all is set out in telling flirting from ordinary friendliness.

#5

Turning the cues down does move the reading, and does not close the gap

The honest counterweight is that behaviour is not irrelevant, and one study measured both things at once.

Eric Koukounas and Nicole Letch showed 183 men and 186 women one of three filmed versions of a conversation between a woman and a man, with the nonverbal content turned up from segment to segment: more eye contact, more touch, closer physical proximity, and more revealing clothing on the female actor.

Two results, and they belong together.

Perceived sexual intent rose as the cues were magnified. The reading tracked the behaviour. Anyone telling you that how you act makes no difference is overselling the case.

And at every level of cue, the men perceived more sexual intent in the female actor than the women did. The gap did not close at the quiet end.

One caveat matters more than the others and is rarely stated: the study moved four things at once. Eye contact, touch, proximity and clothing were varied as a package, so the design cannot say which of them did the work, and it should not be converted into advice about any one of them — least of all the clothing, which is the one the genre always reaches for. It is also a filmed actor watched by students, which is a long way from your Tuesday.

The same paper found that the men scored higher on sexual preoccupation and lower on what the authors call sociosexual effectiveness, and that those variables went with inferring more intent. Reader again, not sender.

What all this licenses is a narrow claim. The volume knob exists. It is not wired to the thing you were hoping to switch off.

#6

Who gets misread, and how little of it is something you are doing

The next question people ask themselves is why it keeps happening to them in particular. The measured predictors are an uncomfortable answer.

Carin Perilloux, Judith Easton and David Buss ran a speed-meeting study with 199 participants. Men overperceived women's sexual interest and women underperceived men's, which by now is the expected pattern. The predictors are the part worth having. Men who were more oriented towards short-term mating, or who rated themselves more attractive, overperceived more. And the magnitude of a man's overperception was predicted by the physical attractiveness of the woman he was talking to.

Bendixen's replication adds the same lesson from the other end. In his analysis, a participant's age and their attitudes towards casual sex changed the profile of being misread; self-rated mate value did not. Relationship status shifted the overall amount of misperception without changing its direction, so being partnered protected nobody.

Look at what those predictors are made of. One is how old you are. One is how the person opposite happens to be oriented. One — being judged attractive — is something done to you rather than something you do.

So the premise under the search, that somewhere there is a dial on your side with your name on it, is only partly true, and the variables that predict best are not on your side at all. That is not a counsel of helplessness. It is the reason the useful moves further down this page are about language and framing rather than about demeanour, and it is why the question what am I doing wrong can be asked for years without returning an answer.

#7

You are the worst-placed person to audit your own signal

Suppose you decide to audit yourself anyway, and work out honestly whether you come across as flirtatious. There is a literature on exactly that ability, and it is not encouraging.

David Kenny and Bella DePaulo reviewed eight studies using the social relations model, 569 participants in total, all designed to measure meta-accuracy — how well people know how others see them. Their conclusion is specific and unwelcome. People work out how others see them mainly from their own self-perceptions rather than from the feedback they receive. They overestimate how consistently different people see them. And they are markedly better at knowing how others view them in general than how any particular person views them.

Erika Carlson, Simine Vazire and Michael Furr qualified that across three studies in 2011, and the qualification is real: people can genuinely distinguish how they see themselves from how they are seen, so meta-accuracy is not purely self-projection. We do have some insight into our reputation.

Now line that up with the question. Do I come across as flirty? is a question about reputation in general, which is the version people are relatively good at — so if several unrelated people have independently told you the same thing, that is information. Is this person reading me as flirting? is the other version, the one measured as weak. It is also the only version that matters on any given afternoon.

The upshot is blunt and oddly freeing. You cannot introspect your way to the answer, because the answer lives in someone else's head and your best guess at it is assembled largely out of your own self-image. The instrument that reaches it is a sentence, not a self-audit.

#8

The behaviours you would have to remove are the behaviours of liking

It is worth being concrete about what a successful suppression would cost, because the genre never prices it.

Matthew Montoya, Christine Kershaw and Julie Prosser pooled 309 effect sizes from 5,422 participants to find which enacted behaviours track attraction. Eleven reached significance: gaze duration and frequency, smiling, laughter, mimicry, head nodding, physical closeness, the amount of talking, the duration of talking, and two specific to speed dating. The overall association was z = .20.

Montoya's own reading of his result, in his university's summary of it, is the line this whole subject turns on: these are trust-and-rapport behaviours, and the same set appears between doctors and patients, parents and children, and salespeople and customers.

So the instruction stop doing the things that read as flirting decodes to: look at people less, smile less, laugh less, stand further away, and say less. That is not a flirtation filter. It is the complete list of ways humans signal that another person is safe and worth attending to, and removing it does not make you legible. It makes you cold — and it strips out the behaviours in which friendship itself is conducted, which is the subject of how people actually come to open up to each other.

There is a second reason it does not work, and it is already in the numbers above. Interest was detected in 7 of 25 real cases. The channel is noisy in the direction of missing things, so the signal you want to remove was being half-missed anyway, while the warmth you would be removing is read accurately most of the time. More on what that channel can and cannot carry under Body Language.

#9

Smiling is also something you are required to do

One of those eleven behaviours deserves its own section, because it is the one most often named when someone is told they are too friendly, and because the demand runs in two directions at once.

Marianne LaFrance, Marvin Hecht and Elizabeth Paluck pooled 448 effect sizes from 162 research reports on sex differences in smiling. Women and adolescent girls smiled more than men and adolescent boys, d = 0.41. That much gets quoted. The moderator analysis is the part that matters here.

The sex difference was larger when concerns about gender-appropriate behaviour were made conspicuous, larger when situational constraints were absent or ambiguous, and larger when emotion was salient. It varied by culture and by age. The authors' own summary is that the degree to which men smile less than women is contingent on rules and roles.

Read that as what it is. Smiling is not only a disposition. In many settings it is a social requirement, and the requirement is distributed unequally. Someone asked to smile less so that she will stop being misread is being asked to step out of a norm she is also penalised for failing.

That is the double bind sitting underneath this search, and it is why this page will not give the obvious advice. The behaviour singled out as the cause is, in a great many contexts, the behaviour being demanded. Someone is going to be unhappy with your face either way, and arranging your face around the reading of the least accurate person in the room is a poor principle to run a life on.

#10

Friends flirt, and being told you do it is not one accusation

There is also a definitional problem buried inside being told that you are flirting, and noticing it dissolves part of the charge.

Kori Egland, Brian Spitzberg and Michelle Zormeier asked people about flirtation in cross-sex platonic and romantic relationships and recovered four kinds of flirtation behaviour by factor analysis: display, stereotyped, attentiveness and conversational. People in both kinds of relationship reported flirting, to varying degrees. What differed between friendships and romances was how the flirtation was displayed, and how it related to judgements of appropriateness and communication competence.

Two consequences. First, flirtation inside a friendship is ordinary rather than a confession, so being told that you flirt is not the same as being told you are advertising availability. Second, and more usefully, one of those four factors is attentiveness — the behaviours of paying close attention to someone. A person describing you as flirtatious may be describing attentiveness, which you almost certainly do not want to stop.

That is why the single most useful response to being told you come across as flirty is a question about which behaviour, in which moment. Display and attentiveness are not the same complaint, and only one of them has an obvious fix.

The research on what flirting is for, and on the six motives people report for doing it — only three of which have anything to do with wanting the other person — is in flirting or just being friendly.

Further illustration for: When Friendliness Gets Read as Flirting, and What Can Actually Be Changed
#11

Some people read as flirtatious because of register, not message

Some people are read as flirtatious nearly everywhere they go. That is a measurable individual difference rather than a moral failing.

Jeffrey Hall, Steve Carter, Michael Cody and Julie Albright developed the Flirting Styles Inventory on 5,020 adults and found five distinct styles: traditional, physical, sincere, playful and polite. Hall and Chong Xing then recorded 51 pairs of strangers and had coders score 36 verbal and nonverbal behaviours. The visible register tracked the style. Playful flirts looked obviously engaging and flirtatious. Polite flirts, when they were genuinely attracted, became more reserved and more distancing.

Set that beside the selectivity research. Paul Eastwick, Eli Finkel, Daniel Mochon and Dan Ariely found, among 156 speed daters, that desire aimed at one particular person tended to be reciprocated, while people who wanted everybody were wanted less — and partners picked up on the lack of selectivity inside four minutes.

What the two imply together is not that general warmth is dangerous. It is that what reads as a message is contrast, not level. A person whose warmth is high and evenly spread is giving the room no contrast to read. A person whose warmth is ordinarily low and goes up for one individual has produced a signal without intending one.

So if you are told you seem flirtatious with everyone, the honest translation may simply be that your baseline is high — and the reader is comparing you against a baseline they have invented out of very little. The same logic explains why visible intensity is such poor evidence of anything, which what body language actually predicts goes through cue by cue.

#12

If you are read wrongly in several directions at once

A share of the people who ask this question are not being read as flirtatious in particular. They are being read wrongly in several directions at once — too intense, too cold, too familiar, too much — and flirtation is simply the reading that got said out loud.

There is relevant evidence, and it has to be labelled carefully, because it is not about flirting. Noah Sasson and colleagues ran three studies on first impressions of autistic adults formed from thin slices of real social behaviour. Impressions made by non-autistic observers were far less favourable across a range of trait judgements, and went with reduced intentions to interact. The effects appeared within seconds and did not improve with longer exposure. They disappeared when the impression was based on conversational content with the audio-visual cues stripped out — which is to say the judgement was tracking style rather than substance.

The honest statement of what that does and does not show: it concerns favourability, not perceived sexual interest, and nobody has measured whether misperception of flirting differs by neurotype. That study has not been run, and a page implying otherwise would be inventing a result.

What can be said is narrower and still useful. The rate at which a person's manner gets converted into confident conclusions by strangers varies between people, it varies in ways nobody can see from the inside, and the conversion happens fast enough that no amount of being oneself for longer corrects it. If that is the pattern, the lever is the one this page keeps arriving at — putting it in words — and the usual advice to simply act more normally is both unkind and unsupported. The same caution applies to the cue people are most often told to manage, and what eye contact actually predicts includes how far the comfortable duration varies by culture and by neurotype before anyone's feelings enter into it.

#13

What actually reduces the ambiguity is language, not demeanour

Everything above converges on one lever, and it is not how you carry yourself.

Kristen Jozkowski and colleagues asked 185 students, in open-ended questions, how they define, communicate and interpret sexual consent. Men and women defined it in the same way. They differed in how they signalled it — women reporting more verbal strategies, men more nonverbal ones — and in how they read a partner, with men relying more on nonverbal cues.

Now put that beside the detection rates. The nonverbal channel is the one being misread, and it is the channel you are proposing to adjust. Adjusting it means competing in the medium with the worst measured accuracy, against a reader whose sensitivity you cannot change, at the cost of the behaviours that make you pleasant to be around.

The verbal channel does not have that problem, and the evidence for it is the asymmetry quoted earlier: the absence of flirting was recognised 84% of the time. People read not that reliably when it is there to read. A sentence puts it there.

In practice this is much smaller than a conversation about feelings. It is naming the frame once, early, in the ordinary run of talk — a partner mentioned, a plan that has other people in it, a thing described as a work thing, an invitation declined specifically rather than vaguely. It is not a speech, and it is not addressed to anybody's hopes.

Two limits, because this site does not pretend otherwise. No study has tested whether saying the frame out loud reduces how often you are misread. That experiment does not exist. What exists is a measured difference in the reliability of the two channels, and what follows from it is labelled here as an inference rather than a finding. And when the question runs the other way — when you are the one who wants to know where you stand — the measured advice has the same shape, which is to ask once, in a form that is easy to decline.

#14

Warm and clear are not opposites

The move that follows is unglamorous: keep the warmth and add the frame. It is worth spelling out, because the intuitive alternative — becoming cooler — is the one people reach for and the one the evidence argues against.

What follows is reasoning from the measured parts above rather than a set of findings, and it is flagged that way on purpose.

Decline specifics, not people. A vague cooling of your manner is a signal with no content. I can't on Friday, I'm out with my sister has content. Research on rejection suggests people are far gentler in real situations than they predict they will be, so the cost of being specific is usually lower than it feels from the inside.

Say the frame early rather than correcting late. Impressions form within seconds and resist revision with more exposure, which is an argument for the first five minutes rather than the fifth meeting.

Do not run a campaign. Repeating the frame in different words at every meeting reads as a subject you want to keep discussing, which is the opposite of what you meant by it.

Expect to be read as colder than you feel. If your baseline is high, a small reduction is a large contrast, and the people you had no wish to redirect will notice it first.

And one thing not to do: asking a third person to pass the message along. That is the indirect route, and indirectness exists precisely to keep things deniable — which is the ambiguity you were trying to remove. Why that deniability is built into flirtation in the first place is covered in flirting or just being friendly.

#15

When the person telling you is your partner

If the person telling you that you come across as flirtatious is your partner, the question changes, and two very different things tend to arrive in the same sentence.

One is a request about a specific behaviour, in a specific setting, that someone found uncomfortable. That is ordinary relationship business. It is answerable and negotiable, and the negotiation has a shape: what happened, what it looked like from where they were standing, and what either of you wants to be different next time.

The other is monitoring. Needing to know who you were with and for how long. Checking your phone. Requiring an account of your tone. A rule about who you may be warm towards. An argument that restarts every time you are friendly in public. That is not a request about a behaviour. It is a claim on your conduct in general, and it does not resolve when you comply, because resolution is not what it is for.

The distinction has its own article here, and the evidence is specific: in the meta-analytic literature, jealousy turns up in a cluster alongside controlling behaviours rather than in place of them. The feeling is human and very common. What follows it is a separate thing, and that is where the line sits. The psychology of jealousy goes through what has been measured, including the famous sex-difference claim that is still in dispute.

If you find yourself editing how you greet people, reviewing your own tone afterwards, or narrowing in advance who you talk to in order to avoid an argument at home, that is worth taking seriously on its own, whatever either of you calls it. Find A Helpline, in the sources below, lists free and confidential services by country. In the United States the National Domestic Violence Hotline is 1-800-799-7233, or text START to 88788, around the clock. If what you have is a disagreement rather than that, our free conflict style test is a two-minute way to look at how the two of you argue, and it says plainly what it cannot tell you.

#16

At work the question is a different question

At work this stops being about romance and starts being about consequences, and being read as flirtatious has a measured professional cost.

David Henningsen and colleagues found that instrumental flirting — flirting in order to get something — is reported more often in work settings than in social ones. In 2023, Henningsen and Mary Lynn Miller Henningsen put 120 people into four-person mixed-sex decision-making groups and measured how members saw one another: the more someone was seen as using charm, the less task-competent the rest of the group judged them to be. One study, students, short-lived groups. It still points where everything else here points — at work, warmth is readily reclassified as tactics.

So the reason to manage the reading at work is professional rather than moral, and the levers are mundane. Keep the frame explicit: the meeting, the deadline, the deliverable. Prefer written channels for anything that could be read two ways, since tone is what travels worst. Decline specific invitations specifically.

Two things are not your job. The first is the reading itself. Under US federal law, harassment is unwelcome conduct based on a protected characteristic, sex included; isolated minor incidents are generally not unlawful, while conduct that is severe or pervasive, or that is made a condition of keeping the job, can be. The Equal Employment Opportunity Commission's guidance is to tell the person directly that the conduct is unwelcome and must stop, and to report it early. Other countries have their own frameworks, and an employer's policy may go further than the law does.

The second is absorbing it quietly when the person outranks you. A reading that arrives attached to power is not a signal problem, and someone who reports to you has very few safe ways to decline anything — which is the reason the lopsidedness runs the way it does.

#17

When it is not a misreading at all

There is a version of this situation that is not a perception problem, and it needs saying plainly, because the vocabulary of mixed signals is routinely used to disguise it.

If someone has been told no and continues; if they explain afterwards that you were sending signals; if the account of your behaviour arrives after the thing they wanted was declined — that is not a report of a perception. It is a justification assembled backwards.

The research is the reason to say so with some confidence. Misperception frequency is predicted by attitudes: in the study of 356 college men, by hostile masculinity, an orientation towards impersonal sex, and drinking in dating situations. Kristen Lindgren, Michele Parkhill, William George and Christian Hendershot's review describes the sex difference as replicated across scores of studies using varied methods, and links it to sexual harassment and assault.

So the empirical picture is not one of symmetric honest confusion. Some of it is honest confusion, measured at around one friendly photograph in eight. Some of it clusters with a set of attitudes and is not confusion at all.

The line this site holds does not depend on which it was. Warmth is not agreement. A long conversation is not agreement. Laughing at someone's joke, accepting a lift, or moving somewhere quieter is not agreement. Agreement is given for a specific thing, can be withdrawn, and is established the plain way: by asking and listening to the answer. Someone who treats earlier friendliness as standing permission is not misreading a cue.

If this has happened to you, it was not caused by how friendly you were, and there is nothing on this page to find yourself guilty by. Find A Helpline lists free, confidential services by country, and in the United States RAINN's National Sexual Assault Hotline is 1-800-656-4673, with online chat at rainn.org.

#18

The cost of the fix, and who is being asked to pay it

The advice this article declines to give — be less warm, be careful, dial it down — has a name in the research literature, and the name is worth knowing.

Fiona Vera-Gray and Liz Kelly use the term safety work for the continuous, largely invisible adjustments women make in public space: the route changed, the headphones left out, the conversation not started, the smile withheld. Their argument, from UK research, is that this work limits women's space for action and responsibilises them for preventing violence — it moves the job of prevention onto the person who might be harmed. This is qualitative feminist scholarship rather than an experiment, and it should be read as a concept with evidence behind it rather than as an effect size.

It names something the search query hides. How do I stop coming across as flirty is, much of the time, a request for instructions in exactly this work. The instructions exist in abundance, they are free, and nobody has ever tested whether they reduce the thing they are sold for.

Put the halves together. The behaviours you would remove are the behaviours of ordinary liking. The requirement to perform some of them is unequally distributed, per the smiling meta-analysis. The reading you would be managing is driven substantially by the reader. And the measured cost of the management falls entirely on you.

None of which means you may not change anything. People adjust how they come across constantly, for all sorts of reasons, and wanting to be read a particular way in a new job or a new town is a perfectly good reason. It means the adjustment is a preference rather than a remedy — and that treating it as a remedy has you paying for a repair nobody has shown works.

The same untested-remedy problem turns up in the adjacent question of accommodating too much, where the measured penalties for asserting yourself are also distributed unevenly: what has actually been measured about being too nice goes through the negotiation experiments in which perceptions of niceness were the mediator.

#19

What to do with the information that someone read you wrong

If somebody has told you that you come across as flirtatious, here is the whole of what the evidence supports doing about it.

Ask which behaviour, and when. The Egland factor structure is the reason: display and attentiveness are different complaints with different fixes, and you seem flirty names neither. A specific answer is usable. A vague one usually means the person is reporting an impression rather than an observation — which is still information, about the impression.

Count the sources. The meta-accuracy work found people reasonably good at knowing their general reputation and poor at knowing how a specific person sees them. Several unrelated people saying the same thing is a reputation, and a reputation is the version you can actually learn about. One person saying it, especially one with a stake in the answer, is a reading.

Decide what you want, rather than what would stop the error. Stopping the error is not available at any price worth paying, because the error is largely in the reader. Being read as a colleague at work, or as a friend by a friend, is available, and the route to it is the sentence rather than the manner.

Then let it go. The one thing you cannot do is run a continuous audit of your own warmth, because you are the single person who cannot see it from outside, and the attempt costs the attention the actual conversation needs. If that loop is already running — if every friendly exchange gets replayed afterwards for evidence — then the loop is the thing worth looking at. Our free attachment style test takes about two minutes and says plainly what a result like that cannot tell you, and where love and anxious attachment come apart covers the machinery underneath that kind of replaying.

#20

What nobody has measured

The limits belong on the page, because this is a subject where the confident version is always the wrong one.

Nobody has tested the fix. There is no study in which people reduced their warmth and researchers measured whether they were misread less often afterwards. Every piece of advice on the first page of results for this question, the restrained version here included, is reasoning from adjacent findings.

The prevalence figures are retrospective self-report. People counting incidents from memory, in student samples, with no way to check the other person's account of the same event. How far self-reports of one's own behaviour can sit from the record is a standing problem in this field, and the clearest demonstration of it is in what phone behaviour on a date actually signals.

Almost all of it is heterosexual, Western and young. Bendixen's replication is Norwegian undergraduates. Farris, Haselton and Buss, Perilloux and Jacques-Tiura are American students. The questions are framed around opposite-sex encounters throughout. How any of this works in same-sex contexts has essentially not been studied, which is a gap of decades rather than an oversight.

The sibling result is a vignette study. It shows that the reading moves with the reader when the described behaviour is held fixed. It does not quantify how much of any real misreading is yours and how much is theirs, and no design currently in the literature does.

And one claim here is explicitly not evidence. That naming the frame in words beats adjusting your manner follows from a measured difference in the reliability of the two channels. It has not been tested directly, and it is labelled that way above rather than quietly upgraded on the way down the page.

What survives is still more than the genre offers. The error is common across a life and modest per occasion, it is driven largely by the reader, you are badly placed to audit your own signal, the behaviours you would remove are the behaviours of liking, and the only instrument with measured reliability is a sentence. More of this under Signs, and the claims this site has had to retract are listed on the corrections page.

Pulling it together

The short version is that you are asking the wrong component to change.

Being misread is common across a lifetime and uncommon on any given occasion. Ordinary friendliness is read as ordinary friendliness most of the time, and when it is not, the best-measured sources of the error sit on the other side of the conversation: sensitivity rather than thresholds, attitudes that cluster in some people and not others, and a reading that moved a full point in the laboratory when nothing about the behaviour changed at all.

You cannot see your own signal from inside. You are worst at predicting how one specific person reads you. And the behaviours you would be removing are the same ones that make you someone people want to be near — several of which you are penalised for dropping.

So keep the warmth and add the words. Name the frame early, decline specifics specifically, and say it once rather than campaigning. And when the reading arrives attached to pressure, or to monitoring at home, it has stopped being a question about signals: a reading of anybody's behaviour is a guess, and it is never permission.

More of this under Signs.

Frequently asked questions

How do I stop coming across as flirty?

Not by becoming cooler, which is the advice the evidence argues against. The behaviours that read as flirtatious — eye contact, smiling, laughter, standing close, talking at length — are the same eleven behaviours Montoya and colleagues found related to liking across 309 effect sizes, and the author's own reading of that result is that they are trust-and-rapport behaviours that appear between doctors and patients and salespeople and customers. Removing them does not make you legible, it makes you cold. What the research supports instead is working in the channel people read accurately: naming the frame once and early — a partner mentioned, a plan with other people in it, an invitation declined specifically rather than vaguely. That is an inference from the measured difference between the verbal and nonverbal channels, not a tested intervention, and this page says so.

Why do people think I'm flirting when I'm just being friendly?

Mostly because of how they read, not how you act. Farris and colleagues separated two possible explanations in a sorting task with 280 undergraduates and found no evidence that men had a lower threshold for calling something sexual interest; what differed was sensitivity, the ability to tell the categories apart. The most direct demonstration is Haselton and Buss's sibling comparison: 131 people rated identical described behaviours — smiling repeatedly at a party, touching an arm — and men's ratings of sexual intent dropped from M = 0.02 for an unrelated woman to M = −1.03 when the person performing them was their sister. The smile never changed. The reader did.

Is it my fault if someone misreads my friendliness?

No, and the research is specific enough to say why. Misperception frequency is predicted by the perceiver's attitudes: among 356 male college students, by hostile masculinity, an orientation towards impersonal sex, and drinking in dating situations. Perilloux and colleagues found the magnitude of a man's overperception predicted by the woman's physical attractiveness, which is not a behaviour at all. And in Bendixen's replication, self-rated mate value did not change the pattern of being misread while the participant's age and the other person's orientation did. Separately from all of that: if someone has treated your friendliness as permission, that is not a perception problem. Agreement is given for a specific thing and established by asking, and nothing earlier in an evening substitutes for it.

Should I stop smiling at people so much?

Consider what you would be stepping out of. LaFrance, Hecht and Paluck pooled 448 effect sizes from 162 reports and found women and girls smile more than men and boys, d = 0.41 — and that the difference is larger when concerns about gender-appropriate behaviour are made conspicuous. Their summary is that it is contingent on rules and roles. In many settings smiling is not a disposition but a requirement, and the requirement is unequally distributed, so a woman told to smile less in order to stop being misread is being asked to drop out of a norm she is also penalised for failing. Smiling is also one of the eleven behaviours that reliably track liking of every kind. You can turn it down. It will cost you more than it buys.

How do I know if I'm actually flirting without realising it?

You are the worst-placed person to answer that, and the finding is old. Kenny and DePaulo reviewed eight studies and 569 participants and concluded that people work out how others see them mainly from their own self-perceptions rather than from feedback, that they overestimate how consistently different people see them, and that they are much better at knowing their general reputation than how one particular person views them. Carlson, Vazire and Furr later showed the insight into reputation is genuine rather than pure projection. So the useful move is to count sources: several unrelated people saying the same thing is a reputation, which is the version you can learn about. One person saying it is a reading, and the reading may be mostly theirs.

What should I say to someone who has misread me?

Something specific, once. The asymmetry in the detection data is encouraging here: in Hall, Xing and Brooks's recorded conversations, the absence of flirting was recognised 84% of the time, while flirting itself was caught in only 7 of 25 cases. People read "not that" reliably when it is available to read, and a sentence makes it available. In practice that is naming the frame rather than making a declaration — a partner, a plan with others in it, a thing described as a work thing, a particular invitation declined with a reason. Avoid repeating it in different words at every meeting, which reads as a subject you want to keep discussing, and avoid routing it through a third person, since indirectness exists precisely to keep things deniable.

My partner says I'm too flirty with other people. What do I do?

Separate two things that arrive in one sentence. A request about a specific behaviour in a specific setting is ordinary relationship business and is answerable: what happened, what it looked like from where they stood, what either of you wants to be different. Monitoring is not that — needing to know who you were with and for how long, checking your phone, requiring an account of your tone, rules about who you may be warm towards. In the meta-analytic literature jealousy appears in a cluster alongside controlling behaviours rather than in place of them, and the psychology of jealousy goes through what is measured. If you are editing how you greet people or narrowing who you talk to in advance to avoid an argument at home, that is worth taking seriously on its own. Find A Helpline lists services by country; in the United States the National Domestic Violence Hotline is 1-800-799-7233.

Does being friendly send mixed signals?

The phrase assumes there is a clean signal underneath, and at the level of observable behaviour there mostly is not. The behaviours that track liking are the same whether the liking is romantic or not, which is why Egland and colleagues found flirtation reported inside platonic friendships as well as romantic ones, across four factors — display, stereotyped, attentiveness and conversational. One of those four is simply paying close attention to someone. So "mixed signals" often describes a reader resolving ordinary ambiguity in a particular direction rather than a sender doing two things at once. Flirting or just being friendly covers why that ambiguity is built in and what it is for.

Is this different at work?

The consequences are, so the calculation is. Henningsen and colleagues found instrumental flirting — flirting to get something — reported more often at work than in social settings, and a 2023 study of 120 people in four-person decision-making groups found that the more someone was seen as using charm, the less task-competent the rest of the group judged them. That is a professional reason to keep the frame explicit: the meeting, the deadline, the deliverable, and written channels for anything that could read two ways. What is not your job is absorbing unwelcome attention quietly, particularly from someone senior. Under US federal law harassment is unwelcome conduct based on a protected characteristic, and the EEOC's guidance is to say the conduct is unwelcome and must stop, and to report it early.

Why does this keep happening to me and not to other people?

Partly it happens to nearly everyone: 88.3% of women and 70.6% of men in Bendixen's sample reported it at some point, and the total amount of being misread did not differ significantly between the sexes — only its direction. Partly it is style. Hall and colleagues' Flirting Styles Inventory, built on 5,020 adults, separates five measurably different registers, and the playful and physical styles look obviously flirtatious while polite flirts become more reserved when they are actually attracted. If your baseline warmth is high and evenly spread, you are giving the room no contrast to read, and what reads as a message is contrast rather than level. That is a fact about your register, not a signal you are sending.

Sources

What this article draws on. For the broader bibliography see our sources page.

Keep reading