What the data actually shows
The classic demonstration comes from Peter Wason, whose selection-task and rule-discovery experiments in the 1960s found that people overwhelmingly test their hunches by looking for evidence that would confirm them rather than evidence that could prove them wrong. Given a chance to falsify a belief, most participants simply didn't try — they kept seeking cases that fit.
Raymond Nickerson's widely cited review of the literature pulled these threads together, describing confirmation bias as appearing across perception, memory, and reasoning, and as a strong candidate for the single most consequential cognitive bias. The pattern he documents is not one mechanism but several stacked together: we search selectively, interpret ambiguous evidence in our favour, and recall confirming instances more readily than disconfirming ones.
A related strand, often called motivated reasoning, helps explain why. Being right is comfortable and being wrong is not, so we tend to apply a tougher standard to evidence we dislike than to evidence we like. The research suggests this is less a deliberate dishonesty than an automatic asymmetry in how carefully we scrutinise things depending on whether they please us.
Why this feels different from how it actually is
It rarely feels like bias from the inside. When you notice yet another example that supports your view, it feels like the world simply confirming what is true — not like you filtering for it. The selection happens before the sense of having an open mind ever arrives, so introspection doesn't catch it.
Modern information environments amplify the effect rather than create it. Algorithmic feeds are built to show you more of what you already engage with, so the stream of evidence arriving in front of you is pre-sorted toward your existing leanings. You can spend hours 'researching' and end up only more certain, because almost everything you saw was selected to agree with you.
The bias also tends to make us overconfident and more polarised over time. Each confirming instance feels like fresh proof, the disconfirming ones quietly drop out of memory, and the accumulated result is a sense of certainty that is built largely from a filtered sample rather than a balanced one.
You are usually carrying a view and then noticing the world through it — quietly, without any sense of being closed-minded.
What the research says to do about it
The most consistently recommended counter is to deliberately seek disconfirming evidence — to ask not 'what supports my view?' but 'what would have to be true for me to be wrong, and is any of it true?' This runs directly against the natural search pattern Wason documented, which is why it has to be done on purpose rather than left to feel its way.
Considering the opposite is a specific, studied technique: before settling on a conclusion, explicitly generate reasons it might be mistaken or imagine how someone reasonable could see it differently. Studies of this prompt suggest it reduces overconfidence, though the effect is modest and tends to fade unless practised.
Diversifying your inputs helps mechanically. Deliberately reading sources that don't already agree with you, and treating an algorithmic feed as a filtered slice rather than the full picture, widens the evidence base before your own filtering narrows it again. The point is not to abandon your views but to test them against something other than their own echo.
What the research says does not help
Simply being intelligent or well-educated does not protect you, and may not even help. The research is clear that confirmation bias appears across ability levels; in some studies, more capable reasoners are better at constructing arguments for what they already believe, which can entrench a view rather than correct it.
Consuming a large volume of information on a topic does not, by itself, reduce the bias — if the information is selected by you or an algorithm to fit your priors, more of it mostly raises confidence. Feeling well-researched is not the same as having looked at the disconfirming side.
Trying to be 'unbiased' by force of will tends to fail, because the bias operates beneath awareness. Telling yourself to be objective does little; structured habits like actively listing counter-evidence outperform good intentions, because they intervene at the point where the filtering actually happens.
Don't ask 'what supports my view?' Ask 'what would have to be true for me to be wrong?'
What this looks like in real life
An afternoon of 'research'
You spend hours reading up on a question and come away more certain than when you started. The trouble is that you searched for what supports your view and an algorithmic feed served more of the same, so almost everything you saw agreed with you. Feeling well-researched isn't the same as having looked at the disconfirming side.
Considering the opposite
Before settling on a conclusion, you deliberately ask 'what would have to be true for me to be wrong, and is any of it true?' and try to see how a reasonable person could disagree. This runs against the natural search pattern, which is why it has to be done on purpose — and studies suggest it trims overconfidence, though the effect is modest and fades without practice.
Real numbers in context
There is no single clean statistic for confirmation bias — it is a robust qualitative pattern across decades of experiments rather than a number. What the research consistently shows is direction, not magnitude: in Wason-style tasks the great majority of participants test their beliefs by seeking confirmation rather than attempting falsification, and across studies people interpret ambiguous or mixed evidence in line with their prior view far more often than against it.
Treat any precise figure you see quoted about 'how biased people are' with caution. The honest summary is that the tendency is widespread, applies in some degree to nearly everyone, and is strong and consistent enough that Nickerson's review treats it as a default feature of human reasoning rather than an occasional glitch.