Why People Trust Confident Speakers (Even When They Are Wrong)

In 2012, Cameron Anderson and his collaborators at UC Berkeley ran an experiment that has bothered me ever since I read it. They put MBA students into small groups and gave them a geography quiz with deliberately ambiguous items — places that looked plausible but did not exist, mixed in with real ones. After the quiz, group members rated each other's competence and confidence.

The students who scored worst on the quiz — the ones who confidently picked the fake place names as real — were also the ones the group rated as most competent. The relationship between actual accuracy and perceived competence was weak. The relationship between displayed confidence and perceived competence was strong.

Anderson titled the paper "A status-enhancement account of overconfidence." The argument is straightforward and slightly depressing: in social groups, confidence reliably earns higher status, even when it is unjustified — amplified by our tendency to overestimate how much others notice our hesitation. Other people read confident behavior as evidence of competence, and they often cannot distinguish it from the real thing.

This is the confidence-competence gap, and it shows up almost everywhere humans make group decisions.

Why we fall for it

The confidence heuristic — using how sure someone seems as a proxy for how right they are — works most of the time. In ordinary conversation, people who know what they are talking about generally speak with more fluency, fewer hedges, and more specifics. So treating confidence as a signal of competence is not crazy. It is a useful shortcut that gets us through most days.

The trouble starts in three specific failure modes:

When the speaker is performing rather than knowing. Confidence is trainable. Sales coaches, public speaking trainers, and political consultants spend careers teaching people how to sound certain without being more right. Once you can manufacture confidence, the heuristic breaks.

When the topic is genuinely uncertain. In fields where experts are honestly unsure — economics, medicine on the frontier, complex political situations — the most accurate experts often hedge the most. The confident speaker is often the one who has stopped updating their views. Philip Tetlock's decades of research on expert prediction (summarized in Superforecasting) found that "hedgehogs" — confident, ideological forecasters — were systematically worse than "foxes" who held uncertain, mixed views.

When confidence escalates inside a group. Groups tend to converge on the most confident voice, partly because hesitation reads as weakness in social settings. Once that voice becomes the group's default, dissenting opinions feel costlier to express, and the group loses access to its own quieter information.

The eyewitness research that should worry you more

If you want a specific example of how badly the confidence-accuracy gap can go, eyewitness testimony research is the place to look.

In the early years of eyewitness research, courts and juries assumed that confident eyewitnesses were more likely to be right. Gary Wells's lab and others showed, in study after study, that this is at best a weak relationship and at worst nearly absent. Confident eyewitnesses are sometimes wrong; uncertain ones are sometimes right. Yet juries consistently weight confident testimony more heavily.

This has real-world consequences. The Innocence Project, which has used DNA evidence to overturn convictions, found that mistaken eyewitness identification was a factor in roughly 70 percent of the cases it has reversed. In many of those cases, the witnesses were certain. They were also wrong.

This is the confidence-competence gap with the volume turned up. It is the same bias that makes a confident colleague seem credible in a meeting — except the cost of being wrong is years of someone's life.

What confidence is actually a signal of

If confidence is not a reliable signal of accuracy, what is it a signal of? A few things:

• The speaker's belief about how the audience will receive certainty. Confident speakers usually have an accurate model of which audiences reward certainty and which reward nuance, and they calibrate accordingly.
• The speaker's stake in the question. People who have committed publicly to a position tend to express more confidence in it than people considering it for the first time, because backing down is socially expensive.
• The speaker's level of detail. Real expertise produces specific, fluent answers. So does well-rehearsed confidence. From the outside, those can be hard to tell apart.

What confidence is not a reliable signal of: how much the speaker has actually thought about the question, how recently they have updated their view, or whether they would notice if they were wrong.

How to filter confidence in real time

You cannot stop the confidence heuristic from firing — it is automatic, and trying to suppress it just makes you tired. What works better is adding a few small checks that you run after the gut response:

• Ask for the specific source or example. "What study?" "Where did you see that?" "When was that decided?" Real expertise can usually answer these without pausing. Performed confidence often cannot.
• Notice the hedges that are missing. Genuine experts, in their area of expertise, tend to hedge in the right places — they will say "as of two years ago" or "in this subset of cases" without being asked. A speaker who never hedges anywhere is either an expert in nothing or performing confidence — and confidence, even when performed, produces real effects on how it's received.
• Pay attention to who is not speaking. In group settings, the most confident voice is usually not the most informed one — it is the one most willing to speak first. The person who has not said anything is often the one who knows the most and has not yet decided what is worth contributing.
• Be skeptical of your own confidence too. The bias is symmetric. You overestimate your competence in domains where you feel certain — especially after a long sequence of decisions when cognitive vigilance has already started to drift. The pattern is so robust that calibration training is a standard part of intelligence analyst education.

The version that matters most

Anderson's geography quiz study has a final detail that I have not been able to forget. After the experiment, when participants were told their group's competence ratings, they updated their self-views based on the group's ratings, not based on the actual quiz scores. The students who had been confidently wrong came away with even higher self-assessments. The students who had been correctly humble came away with lower ones.

This is the feedback loop that makes the confidence-competence gap so durable. Overconfident people get rewarded with higher status, which they interpret as evidence they were right to be confident. Underconfident people get punished with lower status, which they interpret as evidence they should hedge less. Over time, the system selects for confident bullshit and against humble accuracy.

Knowing this does not let you opt out — you are still going to be influenced by confident speakers, and so is everyone you work with. But it does let you put a small, deliberate filter between the gut impression of competence and the actual decision. Ask for the source. Notice who is quiet. Treat your own certainty as a hypothesis. None of these are clever tricks. They are just the discipline of refusing to use confidence as a substitute for evidence.