What the data actually shows
A core finding is what researchers call the closeness-communication bias. Work by Savitsky, Keysar and colleagues found that people are often no more accurate — and in some conditions less accurate — at understanding a close partner or friend than a relative stranger. The reason appears to be overconfidence: with people we know well, we assume we already understand them, so we invest less effort in actually decoding what they mean and lean more on what we expect them to mean.
A separate line of research on empathic accuracy — how well one person can infer another's specific thoughts and feelings in the moment — tells a related story. Studies in this tradition (associated with William Ickes and others) find that empathic accuracy is real but modest, and that long-married or long-partnered couples do not reliably outperform newer couples at reading each other. The sense of effortless mutual understanding tends to outrun the measured ability to actually do it.
A third piece is egocentric projection: when information about another person is missing, we tend to fill the gap with our own perspective, assuming they see, want, or feel what we would. With people close to us this feels especially justified — 'I know them' — which can make the projection harder to notice. The general pattern across these studies is an illusion of insight that grows with familiarity even when the underlying accuracy does not.
Why this feels different from how it actually is
Knowing someone for a long time produces a genuine, vivid sense of understanding, and that feeling is hard to separate from actual accuracy. Fluency is mistaken for knowledge: because thoughts about a familiar person come quickly and easily, the ease itself reads as proof that we have them figured out.
Shared history also gives us a large store of past examples to draw on, which is real information — but it can quietly become a substitute for present-tense attention. People change, moods shift, and circumstances move, yet a confident mental model built over years tends to update slowly. We can end up responding to who someone was rather than who they are today.
And there is a social reason the illusion is rarely corrected. With close people we often stop spelling things out, assuming we will be understood — and the other person, assuming the same, may not flag when we have misread them. The mutual assumption of understanding means misreadings can persist quietly, without the feedback that would reveal them.
Closeness tends to create confidence faster than it creates accuracy — and the gap usually points the same way.
What the research says to do about it
The most consistent practical implication is to treat understanding as something you keep checking rather than something you have achieved. Asking directly — what someone actually thinks, wants, or feels about a specific thing — outperforms inferring it, precisely because the inference is where the overconfidence lives. Perspective-getting (asking) tends to beat perspective-taking (guessing) in studies that compare them.
Curiosity that treats a familiar person as still partly unknown appears protective against the bias. Approaching a long-term partner or old friend as someone who can still surprise you keeps attention switched on, which is the resource the closeness-communication bias quietly turns off.
Slowing down and not assuming you already know what someone meant — especially in conflict or ambiguity — is the small, repeatable habit the research points to. The win is not a dramatic technique; it is reopening the question 'do I actually know this, or am I assuming it?' in the moments where assuming feels most natural.
What the research says does not help
Time alone does not reliably fix it. The data is fairly clear that years together build confidence faster than they build accuracy, so 'we've known each other forever' is not evidence of a good read — sometimes it is the very condition that produces the blind spot.
Assuming a close relationship means you can skip explaining yourself tends to backfire. The expectation that you will simply be understood, without being explicit, is part of the mechanism that lets misreadings go undetected. Familiarity earns trust, but it does not earn mind-reading.
Trying harder to guess what someone is thinking — investing more in inference rather than just asking — is a weaker strategy than it feels. Because the error is partly egocentric projection, more guessing can mean more confidently projecting your own view. Checking beats deducing.
Familiarity earns trust, but it does not earn mind-reading.
What this looks like in real life
Fluency mistaken for knowledge
Thoughts about a familiar person arrive quickly and easily, and that ease itself reads as proof you have them figured out. But quick isn't the same as accurate. The very smoothness of 'I know exactly what they'd say' is the feeling the closeness-communication bias runs on — it turns off the effort you'd spend actually decoding what they mean.
Responding to who they were
A shared history gives you a large store of past examples to draw on — real information that can quietly become a substitute for present-tense attention. People change, moods shift, circumstances move, but a confident mental model built over years updates slowly, so you can end up answering the person they were rather than the one in front of you today.
The misread that never surfaces
With close people we often stop spelling things out, assuming we'll be understood — and the other person, assuming the same, may not flag when we've misread them. Because both sides take mutual understanding for granted, an inaccurate read can sit undetected for a long time, missing exactly the feedback that would correct it.
Real numbers in context
There is no single tidy statistic for 'how well we know each other,' and it would be dishonest to invent one. What the research offers instead is a direction of error: in studies of the closeness-communication bias, people often performed no better at understanding close others than strangers, and sometimes worse, with the consistent driver being overconfidence rather than lack of care.
Empathic-accuracy research adds the texture that reading another person's specific thoughts is genuinely hard for everyone, that scores are modest across the board, and that long-term couples do not reliably do it better than new ones. Taken together, the honest takeaway is qualitative, not numerical: the feeling of deeply knowing someone tends to exceed the measured ability to do it — which is an argument for curiosity, not for cynicism about your relationships.