Why You're Worse At Reading People Than You Think

Most of us are far worse at reading other people than we feel, and the gap is not evenly spread. Research on empathic accuracy suggests we judge stable surface qualities — warmth, confidence, status — reasonably well from very little information, while our ability to infer what someone is specifically thinking or feeling in a given moment sits close to chance. Strangers land near 20% accuracy; long-term partners reach roughly a third. Worse, confidence in a read does not track its accuracy. The single intervention with strong evidence behind it is not a better technique. It is asking, and then believing the answer.

That last sentence sounds too obvious to be a finding. It is a finding, and this article is about why it took the field decades to establish it — and why almost nobody acts on it.

The part of reading people that genuinely works

Start with what the evidence supports, because it is real and it is the reason the illusion holds.

In 1993, Nalini Ambady and Robert Rosenthal showed participants silent thirty-second video clips of college teachers they had never met. Strangers rated those teachers on qualities like warmth and competence, and the ratings predicted the teachers' end-of-term student evaluations with striking accuracy. Cut the clips to six seconds and the predictive power barely dropped. The effect — now known as "thin slicing" — has since been replicated with physicians, salespeople and job candidates.

So something in human social perception works, and works fast. What it reads well are relatively stable, outward-facing traits: warmth, dominance, confidence, status, whether someone is comfortable in a room. These leak continuously through posture, vocal tone and facial behaviour, whether or not the person intends them to. They are, in a real sense, on the surface.

The trouble is that this success does not generalise, and nothing in the experience of reading someone tells you when you have crossed from the thing you are good at to the thing you are not.

Where it collapses: specific thoughts and feelings

William Ickes spent much of his career measuring what he called empathic accuracy. Two people have an unscripted conversation on video. Afterwards each watches the tape alone and writes down what they were actually thinking at specific moments, then watches again and writes down what they believe the other was thinking. Independent coders score the match.

The numbers are humbling. Strangers infer a specific thought correctly around 20% of the time. Close friends and long-term partners reach roughly 30–35% (Ickes, Everyday Mind Reading, 2003).

The implication is easy to skate past. Two people who have shared a home for fifteen years correctly identify what the other is actually thinking about one time in three. Not because either is careless. Because the information was never on the surface. Specific thoughts are not broadcast the way warmth is.

The confidence problem

Here is the part that turns a curiosity into a practical problem.

As relationships deepen, accuracy improves only modestly. Confidence, on the other hand, climbs steeply. Partners become considerably more certain that they know what the other is thinking, while their actual hit rate moves far less. The consequence is that the longer you know someone, the wider the gap tends to grow between how well you read them and how well you believe you read them.

The same disconnect shows up starkly in deception research. In 2006, Charles Bond and Bella DePaulo published a meta-analysis pooling more than 200 studies of deception detection. Mean accuracy across all of them: 54%. Police officers, customs officials, judges and clinicians clustered within a few points of undergraduates. Training programmes moved the figure very little.

And the finding that should genuinely unsettle you: across that literature, confidence and accuracy were essentially unrelated. The people who felt certain were not the people who were right.

This suggests something uncomfortable about what certainty actually measures. It is not a readout of the other person. It appears to be a readout of how easily your mind assembled an explanation — and explanations come easily when they are simple, familiar, and flattering to the person doing the explaining. Fluency feels, from the inside, exactly like evidence.

But what about microexpressions?

This is the standard objection, and it deserves a straight answer.

The popular claim is that suppressed emotion flickers across the face for a fraction of a second, and that a trained observer can catch it. The evidence supporting it as a practical deception-detection tool is weak: independent reviews have repeatedly failed to show that brief facial leakage reliably distinguishes liars from truth-tellers under realistic conditions.

The most expensive test of the idea was not a laboratory study. The US Transportation Security Administration spent roughly $900 million training airport personnel in behaviour detection. In 2013 the Government Accountability Office reviewed the programme, concluded the available evidence did not support its effectiveness, and recommended Congress limit future funding (GAO-14-159).

A technique can be famous, expensive, widely taught, and still not do what it claims.

The finding that reorganised the field

If better observation is not the answer, what is?

In 2018, Tal Eyal, Mary Steffel and Nicholas Epley published a series of 24 experiments in the Journal of Personality and Social Psychology testing the single most recommended piece of advice in social psychology: perspective-taking. Put yourself in their shoes. Imagine the world from their point of view.

They tested it on strangers, on friends, on married couples. On reading emotion from faces, eyes and voices. On predicting another person's preferences, opinions and attitudes. Perspective-taking failed to improve accuracy in essentially every study. In several, it made participants measurably worse — while making them more confident.

The proposed mechanism is straightforward once stated. When you imagine another person's viewpoint, you are not accessing them. You are running a simulation assembled entirely from your own stored material: your assumptions, your generalisations about their group, your memory of who they were years ago. The simulation feels vivid, and vividness gets misread as evidence.

One thing in that same body of work did raise accuracy, consistently and substantially. The authors call it perspective-getting: asking the person directly, and listening to what they say. Not inferring harder. Not studying the face more closely. Obtaining the information from the only source that reliably holds it.

Why the obvious answer gets skipped

If the fix is this simple, why is it rare?

Partly because it does not feel like a skill. "Ask them" has none of the appeal of a technique you could be good at. Partly because asking is genuinely uncomfortable — a guess protects you from an answer you might not want to hear. Some of the accuracy most of us are missing is accuracy we are quietly choosing to miss.

And partly because of a second, mirrored error. We also tend to overestimate how visible our own internal states are to other people — a well-documented effect that Thomas Gilovich, Kenneth Savitsky and Victoria Medvec named the illusion of transparency (1998). If you assume your feelings are obvious, you will not state them; if the other person assumes the same, neither of you says anything, and both of you infer.

What to actually do

Three changes, in rising order of difficulty.

Downgrade the vocabulary. Replace "he's annoyed" with "I'm reading annoyance." The sentence is awkward, and the awkwardness is the point: it restores the uncertainty the original sentence silently deleted. A read held as a hypothesis produces different behaviour than a read held as a fact.

Ask a question that cannot be answered with "fine." "Are you okay?" is, functionally, an invitation to close the topic. "What's taking up most of your head right now?" is not.

When the answer contradicts your read, believe the answer. This is the hard one. Your model was wrong — which is precisely what new information feels like from the inside.

Two honest limits. Where someone has a genuine incentive to mislead you — a negotiation, a sale — stated answers deserve less weight and repeated behaviour over months deserves more. And where someone treats your questions as material to use against you later, the appropriate response is distance, not a better technique.

FAQ

Are some people naturally good at reading others? There is measurable individual variation, but it is smaller than most expect and it does not rescue the general picture. Believing yourself to be good at it is not a reliable indicator of being good at it.

Does knowing someone for longer help at all? Modestly, for some kinds of inference. The problem is that confidence rises much faster than accuracy does, so the error in your self-assessment tends to grow with familiarity rather than shrink.

Is body language useless, then? No. It carries real information about relatively stable qualities — comfort, confidence, engagement, status. What it does not reliably carry is the content of a specific thought, or whether a specific statement is true.

Why did perspective-taking fail when it's such standard advice? Because imagining someone's viewpoint draws on your own stored assumptions rather than on them. The vividness of the resulting simulation is easily mistaken for accuracy.

Connect

Questions or thoughts? Reach out anytime.

Email

subscribe to our newsletter

hello@minddecoded.com

© 2026. All rights reserved.

Terms of Service Privacy Policy Disclaimer