Coworker: friendly or flirting? How men misread warmth at work

Friendly. That's the default, and it holds until multiple signals stack against her personal baseline — over weeks, in ways that cost her something. One laugh, one arm touch, one warm goodbye is not information. The research is embarrassingly one-directional about which sex needs to hear this. And the highest-leverage fix isn't studying her harder — it's calibrating the instrument doing the reading: you.

Saturday night, your coworker Kyle's housewarming. You anchor yourself to the kitchen island because it gives your hands a job: opening bottles, finding cups. Kyle's girlfriend introduces her friend Maya, and for ten minutes it's easy — her dog, the trail you hiked last weekend. She laughs three times. You counted. Once, she touches your forearm. Then a group photo pulls her away.

For two hours afterward you track her position with peripheral vision and lose four straight rounds of the same debate: go back over, or is that weird now? At the door she says "you too!" — pitch rising — and you spend fifteen minutes of the drive home replaying that audio through every decoder you own.

Monday, 11 p.m., the party group chat floods with 87 photos. One catches you mid-sentence with Maya, and for the first time in twenty-seven years you see yourself talking to a woman from the outside: shoulders rolled forward, T-shirt bunching at the waist, smile stiffer than it felt. You stare for twenty minutes: which conversation was she actually in — the one you remember, or the one in this photo?

Swap the kitchen island for the Tuesday standup and this is the most common decoding problem men bring us: coworker — friendly or flirting?

You've already read the answer. What follows is why your gut refuses to accept it — and the narrow set of conditions under which you're allowed to move off the default.

Why do men misread friendly as flirting?

Because male perception runs on firmware tuned for a cost structure that no longer exists. Men don't over-read warmth out of arrogance; the bias is built in, and it shows up on camera. If you're asking "is she interested or just nice?", the honest prior is: probably nice.

The cleanest demonstration is Abbey's 1982 experiment. A man and a woman talked for five minutes while another man and woman observed; everyone then rated the interaction. Both the male participant and the male observer rated the woman as more seductive and more promiscuous than either woman rated the same five minutes. Same room. Same behavior. Two different movies.

Haselton and Buss named the mechanism in 2000: error management theory — when two possible errors carry asymmetric costs, selection tunes perception toward the cheaper one. For ancestral men, missing genuine interest cost more than an awkward false alarm, so the male read skews positive the way a smoke alarm skews jumpy. (The same theory predicts women under-read men's commitment — everyone's software is biased, just in different directions.)

Your own state contaminates the read further. In Dutton and Aron's 1974 bridge study, men who had just crossed a swaying suspension bridge pursued the female experimenter far more often than men who crossed a stable one — misattribution of arousal: the brain files its own adrenaline as evidence about her. Post-party buzz, a good review, a third drink — all booked to the wrong account.

Two conclusions. First, your gut's "she's into me" signal is the least trustworthy instrument you own — it fires on baseline warmth, on professional politeness, occasionally on a barista spelling your name right. Second, and worse: the modern office inverts the ancestral cost matrix. A false positive at work — acting on interest that isn't there, with someone who shares your Slack channels and your manager — now costs vastly more than a false negative. Your alarm was calibrated for the wrong building.

Caveat: biased doesn't mean always wrong — coworkers do fall for each other. It means your unaided read overstates the probability in a known direction, so the evidence bar has to sit higher than your gut insists.

Friendly vs flirting: what actually separates the two?

One principle does nearly all the work: judge her behavior toward you against her behavior toward everyone else, never against a generic list of signs. Warmth is a personality trait — and at work, partly a job requirement. Deviation from her own baseline is information. Behavior she shows everyone is just her.

"Baseline" means exactly what it sounds like: how she normally treats colleagues she has zero romantic interest in. That's the control group. A signal only exists as a difference from it.

This is why "signs she likes you" listicles fail. They score the behavior and never score the baseline — and the work baseline is inflated by design: friendliness is professionally mandated, and client-facing roles are literally paid to be warm. A smile at the office is the least informative smile she produces all day.

Behavior	Reads as friendly when...	Worth noticing only when...
Laughs at your jokes	She laughs generously with everyone; it's her social style	She laughs at yours specifically — including the weak ones — and nowhere else
Brings you a coffee	She's the one who does snack runs for the whole pod	You're the only recurring recipient, unprompted, more than once
Asks about your weekend	Monday small talk is workplace liturgy	She follows up Thursday on a detail you mentioned once
Touches your arm	She's expressive and tactile with friends of both sexes	Touch shows up only with you, across separate occasions
Long eye contact	Normal conversational attention; people look at people	Held noticeably longer with you than with others, repeatedly
Messages you	It's about work, on work channels, at work hours	She opens non-work threads, off-channel, on her own time
One-on-one time	Project structure puts you together	She creates it when nothing about the job requires it

Read the right-hand column again. Every entry is a deviation — from her baseline, from what the job requires, from what the politeness script prescribes. That's the entire method. Not "did she laugh," but "does she laugh like that at Dave from finance."

One honest limit: you can't see her full baseline. You get one context, a few hours a week, filtered through your own bias. Which is one more reason the friendly default deserves to outlive your patience with it.

Which coworker signals actually carry information?

The signals worth weighting share three properties: she initiates, it costs her something, and it repeats. Warmth, laughter, politeness, one party-adjacent arm touch — none of it clears that bar. What does:

1. Off-channel initiation. She starts non-work conversations in non-work spaces — texting your phone, not your Teams — on her own time. This costs her deniability, which is exactly why it carries signal.
2. Engineered proximity. One-on-one time the org chart doesn't explain, created by her, more than once.
3. Held, repeated eye contact. Not a glance. In Kellerman's 1989 experiment, two minutes of mutual gaze between strangers measurably raised reported attraction. Gaze is potent — which is why adults ration it, and why someone repeatedly spending it on you is spending something real.
4. Asymmetric memory. She retains and resurfaces small details of your life, unprompted, weeks later.

The rule this article stands behind: three or more of these, initiated by her, sustained across weeks — before you're allowed to move the needle off "friendly." Notice how much higher that bar sits than wherever your gut just put it. That gap is the calibration.

A full stack still isn't certainty — it's a probability shift. It earns you no green light; at most, the standing to make one respectful, easily-declinable move outside work. Which is the part most articles refuse to spell out.

What should you do at work — and what should you never do?

At work, behave as if the answer is "friendly" even while your private model updates. Not because your read must be wrong — because the office is the one environment where testing it isn't yours to do cheaply. She can't freely exit the interaction the way she could at a bar. Her politeness is load-bearing: team dynamics, reputation, reviews. Every ambiguous move you make is something she has to manage at her job.

So the never-list, in order of importance:

• Never run tests. Escalating touch, loaded jokes, engineered late nights to see how she reacts — that machinery outsources the cost of your uncertainty onto her, and it's exactly what makes workplaces worse for women.
• Never leverage structure. If saying no is expensive for her — you're senior to her, you control something she needs, she's cornered at a work event — the question is disqualified before it's asked. Full stop.
• Never renegotiate a no. "I'm so busy right now!" is a no. It does not get appealed at the holiday party.

If — and only if — the signals have genuinely stacked, no reporting line runs in either direction, and a genuinely social context exists: one clear, specific, low-pressure invitation she can decline with zero professional residue. "Coffee Saturday?" is the entire move. Her answer, at any level of vagueness, is final. Anyone selling you a longer playbook is selling you risk and billing it to her.

Underneath all of it sits the actual thesis: the costs of misreading are asymmetric. Read friendly as flirting and act on it — she pays first, you pay second, and the bill can run for years. Read flirting as friendly and do nothing — the cost is a maybe that quietly expires. When one error is catastrophic and the other is survivable, you don't need certainty. You need a default. Friendly is the default.

Are you reading her signals — or misreading yourself?

Mostly yourself. Every interpretation of her warmth silently assumes you know how you land in her eyes — and that's the variable you have the least data on and the most distortion about. Getting ground truth on your own first impression does more for the friendly-or-flirting question than any amount of decoding her.

Go back to the party photo: twenty minutes staring at rolled shoulders and a stiff smile. That instinct was right — just twenty-seven years late and one frame short. Her read of him didn't form during the conversation he remembers. Willis and Todorov showed in 2006 that people form stable trait judgments from a 100-millisecond glimpse of a face, and those snap judgments track closely with ones made under no time limit. Extra exposure mostly adds confidence, not revision.

First impressions also don't average upward with each pleasant chat. Perception works more like a threshold than a meter: the first glance either registered you or filed you under harmless-colleague — and her warmth afterward looks identical from your side either way. The woman who noticed you and the woman being professionally kind both laugh at the Tuesday joke. That's why signal-decoding without self-knowledge always terminates in the same place: a 20-minute stare at one photo, no verdict.

You can't ethically A/B test a coworker. You can get the other side's data. Get an honest read on how your face and presence land in that first glance — the one that formed before you ever spoke — plus which lever is holding the read down. Cross-check it against the signals you already get in the wild and a working model of what women actually register in men. With that ground truth, "is she interested or just nice" mostly answers itself: you'll know whether her warmth is you-specific or just weather.

He opened Google that night, typed "how do women actually see—" and deleted the whole line. The question has an answer. It was never going to come from her signals.

Key numbers

• 5 minutes — the length of the conversations in Abbey's 1982 study, after which male participants and observers rated the same woman as more seductive and more promiscuous than female raters did.
• 2000 — the year Haselton and Buss formalized error management theory: men's over-perception of sexual interest as an evolved default error, whose cost structure the modern workplace has since inverted.
• 100 milliseconds — face exposure after which trait judgments already track judgments made with unlimited time (Willis & Todorov, 2006). Her first read of you predates every conversation you're now decoding.
• 2 minutes — of induced mutual gaze between strangers measurably increased reported attraction (Kellerman, 1989). Held, repeated eye contact earns its place on the signal list; a glance does not.
• 3+ — deviations from her baseline, initiated by her and sustained over weeks, before this article licenses re-reading warmth as possible interest.
• 1 — invitation a full signal stack earns you: specific, low-pressure, outside work, declinable at zero cost. Any no is final.

The bottom line

"Coworker: friendly or flirting?" is a well-formed question pointed at the wrong side of the interaction.

The evidence says your instrument over-reads — put men and women in front of the same five minutes and the men see more intent, reliably. So default to friendly. Score deviations from her baseline, not behaviors from a listicle. Require the stack — initiated by her, costly to her, repeated across weeks — before you update. At work, keep the friendly assumption in your behavior even while the model updates privately: misreads there are asymmetric, and she pays first.

Then put your effort where the leverage actually is: not decoding her warmth, but learning what her first glance already concluded about you. That read exists. It formed in a tenth of a second, and you've never seen it. Take the test — and spend the next kitchen-island conversation talking instead of decoding.

---

Studies referenced: Abbey, A. (1982). Sex differences in attributions for friendly behavior: Do males misperceive females' friendliness? Journal of Personality and Social Psychology, 42(5), 830-838. Haselton, M. G., & Buss, D. M. (2000). Error management theory: A new perspective on biases in cross-sex mind reading. Journal of Personality and Social Psychology, 78(1), 81-91. Willis & Todorov (2006), Psychological Science, 17(7), 592-598. Kellerman, Lewis & Laird (1989), Journal of Research in Personality, 23(2), 145-161. Dutton & Aron (1974), Journal of Personality and Social Psychology, 30(4), 510-517.