At 11:40 p.m. you are lying in the dark writing a novel. The protagonist is the person breathing beside you, and you are composing their inner monologue with real skill: the sigh at dinner meant this, the short text meant that, the silence on the drive home was a verdict on you. You will never show them this draft. You don't need to. You already believe it.

Here is the uncomfortable part. The longer you have known someone, the more fluent that fiction becomes, and the research on how well we read other minds says fluency is not the same as being right. It is closer to the opposite. The more confidently we imagine our way into another person's head, the less likely we are to check, and the one person who could correct the draft is the one we have quietly stopped asking.

The mind reading you do without noticing

Humans are compulsive mind readers. Psychologists call it mentalizing, or theory of mind: the automatic inference of what someone believes, wants, and feels from the thin evidence of their face, their word choice, and their timing. It runs constantly and mostly beneath awareness, which is why it feels less like guessing and more like seeing. You do not experience yourself inferring that your friend is annoyed. You simply see an annoyed friend.

That seamlessness is the problem. A guess that feels like perception never gets treated as a guess. It gets treated as data, and everything downstream, the careful wording, the pre-emptive apology, the week of polite distance, gets built on top of it.

What the experiments found

In 2018, social psychologists Tal Eyal, Mary Steffel, and Nicholas Epley published a series of more than twenty experiments testing the advice every one of us has been given: to understand someone, put yourself in their shoes. Participants tried to predict other people's feelings, preferences, and attitudes. Some were explicitly instructed to take the other person's perspective first, to imagine their situation and their emotions before answering.

Perspective taking did not make people more accurate. In several of the studies it made them slightly less accurate. What it reliably did was make them more confident. People finished the exercise feeling they understood the other person better while understanding them no better at all, and sometimes worse. The effort of imagining had been mistaken for the achievement of knowing.

What worked was almost embarrassingly plain. The researchers called it perspective getting. Instead of imagining the other person's mind, you ask them about it, or you let them tell you, and then you take what they say seriously. In one study, romantic partners predicted each other's preferences. Couples who were simply given a chance to talk before guessing outperformed couples who had been told to imagine. Getting a perspective worked. Taking one did not.

Why imagining fails: you only have one head to imagine from

The failure is not laziness. Imagining another mind has to begin from the only mind you have direct access to, which is yours. Epley and colleagues described the mechanism years earlier as egocentric anchoring and adjustment. You start from your own reaction and adjust outward toward the other person, and the adjustment tends to stop too early, usually the moment an answer feels plausible.

So when you imagine how your partner felt about the comment at dinner, what you mostly generate is how you would have felt, lightly edited. When you barely know the person, imagining reaches for a different shortcut, the stereotype, which is why putting yourself in the shoes of someone unlike you can produce a picture that is both vivid and wrong. In neither case does the simulation have access to the one thing that would make it accurate: the other person's actual experience.

There is a second trap layered on the first. Imagining feels like work, and it feels generous. You tried to see it their way. That sense of having done something virtuous is exactly what inflates the confidence the studies measured, and confidence is what shuts the door on checking.

The closeness trap

You might assume this applies to strangers and colleagues, not to the people you love. The research on empathic accuracy, pioneered by William Ickes, suggests otherwise. Even close partners, asked to infer what the other was thinking and feeling moment by moment during a recorded conversation, score far below what they would predict for themselves. Knowing someone helps, but much less than it feels like it should.

One study of married couples found that empathic accuracy was highest in the earliest years of marriage and lower among couples who had been together longer. One reasonable reading is that early on, you still know you don't know, so you watch and ask. Later, a stored model of the person takes over. You stop collecting evidence because you feel you already have it. Meanwhile the actual person keeps changing, and the model does not.

The night-time novel is that stored model talking. It is articulate because it has had years of practice. It is wrong in the specific ways you cannot see, because you are the one who wrote it.

What getting a perspective actually sounds like

Perspective getting is not an interrogation, and it is not the therapeutic cliché of asking how someone feels in a tone that already knows. Three things distinguish it.

First, it is specific. Not "are we okay," which invites a reflexive yes, but "what was going on for you in the car last night." A specific question is answerable, and it signals that you noticed.

Second, it arrives empty. "Were you upset with me?" carries your draft inside it, and people often agree to a draft rather than write their own. "What was that like for you?" leaves room for an answer you did not predict, which is the only kind worth getting.

Third, and hardest, you accept the answer provisionally. It will sometimes contradict the story you have been living inside for a week. The temptation is to decide the person is hiding something, or doesn't know themselves. People do have limited insight into their own minds, but their report still beats your simulation, because it is drawn from the inside.

There is a reason the other person has not volunteered any of this. Thomas Gilovich and colleagues documented what they called the illusion of transparency: we believe our internal states leak out far more visibly than they do. Your partner thinks the hurt is obvious, so there is no need to say it. You think you have read them, so there is no need to ask. Two confident people, each certain the other already knows, sit in a quiet that neither of them chose.

Your next moves

  • Audit one draft today. Pick a person whose mind you have been "reading" this week. Write your theory in one sentence. Then ask them one open question about that exact thing before the day ends, and compare their answer to your sentence.
  • Swap one statement for one question. The next time you are about to say "you seem annoyed" or "I know you're disappointed," say instead: "What was that like for you?" Then stop talking.
  • Run the prediction game with someone close. Each of you writes five guesses about the other: what they would order tonight, what they most want this weekend, what is weighing on them, what they wish you would ask. Swap, score, and talk about the misses.
  • Write the question, not the answer. When you catch yourself composing someone's inner monologue at night, write down the question underneath it, such as "What did the silence in the car mean?" Keep it to ask in daylight.
  • Say one thing you assume is obvious. Name a feeling you believe the other person already knows about, out loud, today. You are probably the only one who can see it.

The question you meant to ask

Most of the questions that would repair a misunderstanding are not hard to ask. They are easy to lose. They occur at the wrong moment, in the car or at midnight, and by morning they have hardened into an assumption that no longer feels like a question at all. Curiokit was built for that gap: a place to catch a question the moment it appears, keep it until you can actually ask it, and build a small habit of getting other people's perspectives instead of inventing them. If you would like to stop writing other people's inner monologues and start reading the real ones, you can begin at curiokit.lumenlabs.works.