Two people have a conversation at a work event, or during orientation week, or at a dinner where they happen to be seated next to each other. By most objective measures, it went well. Both were engaged, they found things to talk about, there were at least a couple of moments that could be described as warm. When they part ways, one of them thinks: I talked too much, that probably came across badly, they were likely just being polite.
The other person walks away thinking: that was a genuinely nice conversation.
The discrepancy, according to researchers at Cornell, Harvard, Yale, and the University of Essex, is not a coincidence. It is a consistent and measurable pattern.
What the liking gap is
A study published in Psychological Science in 2018, led by Erica Boothby, Gus Cooney, Gillian Sandstrom, and Margaret Clark, ran five studies testing how well people could estimate how much their conversation partners liked them. The settings ranged from strangers in a laboratory paired for a 5-minute icebreaker, to first-year university students getting to know their dormitory roommates over several months, to adults at professional development workshops. In every context, the finding was the same.
People consistently underestimated how much their conversation partner liked them and enjoyed their company. Not slightly. Measurably. Both people in every conversation tended to rate their partner as more likable than they believed their partner rated them — which is logically impossible unless at least one person in each conversation was making a systematic error. The data suggested that person was almost always the one doing the self-assessing.
What made this finding particularly difficult to explain away was that the researchers also analyzed video recordings of the conversations. The footage showed genuine positive signals: smiling, eye contact, engaged body language. The participants weren’t misreading distant or disinterested partners. They were misreading warm ones.
What the internal monologue is doing
The researchers identified the driver. While one person is present in the conversation, the other person is simultaneously running a self-critical commentary on how the conversation is going — monitoring their own word choices, mentally replaying what they just said, cataloguing moments they felt fell flat.
As Margaret S. Clark, the John M. Musser Professor of Psychology at Yale, noted: “They seem to be too wrapped up in their own worries about what they should say or did say to see signals of others’ liking for them, which observers of the conversations see right away.” The person on the other side of the conversation, by contrast, is not conducting the same audit. They are receiving the conversation rather than grading their own performance within it.
Boothby and Cooney described the result: “When it comes to social interaction and conversation, people are often hesitant, uncertain about the impression they’re leaving on others, and overly critical of their own performance.” The internal monologue, in this account, is not just noise. It is producing a genuinely inaccurate picture of what is happening in the room.
Why this runs opposite to how we usually see ourselves
The liking gap stands out for a specific reason: it runs against the direction of most self-perception biases. In most domains, people rate themselves above average. Better drivers than most. More competent in their jobs. Less likely than average to face serious illness or relationship failure. Self-serving bias is one of the most robust findings in social psychology.
But in conversations, the direction reverses. People become systematically pessimistic about a domain where they have more direct, real-time feedback available than almost anywhere else. As Boothby and Cooney noted, “In light of people’s vast optimism in other domains, people’s pessimism about their conversations is surprising.”
The researchers’ hypothesis is about social risk. When the assessment involves another person’s opinion of you, the stakes of an overestimate feel different. Getting the read wrong in a social context means potential rejection, not just personal error. Clark described it simply: “We’re self-protectively pessimistic and do not want to assume the other likes us before we find out if that’s really true.”
The protection is understandable. Its cost, the study suggests, is that people routinely underinvest in relationships that were already going well.
What doesn’t fix it on its own
One of the study’s less obvious findings is that the liking gap is not a first-impression effect that resolves as people get to know each other. The tracking study of university roommates showed it persisting over months of regular contact. Familiarity, on its own, does not close the gap — at least not at the rate most people would assume.
The distortion also appears independent of conversation quality. People who had objectively warm exchanges — the ones captured on video with visible engagement and laughter — were just as likely to underestimate how their partner felt about them. The problem isn’t in how the conversation went. It’s in the assessment that follows it.
I’m not a psychologist, and this isn’t professional advice — these are findings from a peer-reviewed study conducted in specific experimental and naturalistic settings, which the researchers note may not generalize to every relationship context. But what the research does offer is a challenge to one of the more common post-conversation habits: the quiet re-adjudication of how things went, which the data suggests almost always ends up more negative than the other person’s experience of the same exchange.
The person you just spoke with probably liked talking to you more than you think they did. That finding has held, across five studies and multiple countries, and the researchers have not yet found a context where it reliably reverses.
If this is something that shows up heavily in your life, speaking with a therapist about social anxiety is genuinely worth considering.
