Why We Bite on the Bullshit

There has never been a better time to be confidently wrong.

We carry phones capable of reaching more information in a few seconds than most of us could once have tracked down in hours. Google can answer a question before we’ve finished typing it. AI can take a complicated subject and explain it in plain English before the coffee’s ready. Social media, meanwhile, has managed to put experts, journalists, politicians, neighbours, lunatics and somebody’s cousin who “did his own research” into the same scrolling feed. You’d think access to all that information would make bullshit harder to sell. Somehow, we’ve managed the opposite.

I don’t think the most interesting question is why people lie. People have been doing that forever. What interests me is why perfectly intelligent people sometimes believe things that, with a little distance, they might question. More uncomfortable still, why do we believe them? Because if the whole explanation is that gullible people believe nonsense and sensible people don’t, we get to feel superior for a few minutes and learn absolutely nothing.

I’m much more interested in what happens before we decide something is true. Some claims make us suspicious right away, while others seem to walk straight through the front door. Confidence can sound a lot like knowledge, familiarity can feel like proof, and when thousands of other people appear to believe something, our own belief can suddenly feel a little safer. None of that started with AI, Google or social media. They’ve just given some very old human tendencies a much better distribution system.

Which brings me to the question I think we need to ask more often:

How do we know this?

Bullshit Usually Doesn’t Look Like Bullshit

The obvious stuff isn’t particularly difficult. If somebody announces that German Shepherds are secretly controlling the Bank of Canada, I’m probably going to require some documentation. Kai’s reasonably bright, but monetary policy hasn’t come up.

The stuff worth worrying about usually looks much more respectable. There might be a statistic attached to it, or a study, a chart, a screenshot or somebody with impressive credentials explaining why the claim is obviously true. Better yet, some of it probably is true. The claim gets through the door because it connects with something we already know, something we’ve seen ourselves or something we were already inclined to believe.

And to be fair, those signals aren’t meaningless. Credentials can matter. Sources matter. Experience matters. Other people’s judgment can matter. We use shortcuts because we have to; none of us has the time or expertise to investigate every statement from first principles before deciding whether to believe it. If somebody tells me it’s going to rain this afternoon, I’m not conducting a peer review before I take a jacket.

Most of the time, sensible shortcuts work perfectly well. The trouble begins when we confuse the things that usually accompany reliable information with proof that the information itself is reliable. A confident voice, a citation and an impressive title may all be good reasons to pay attention, but they’re not the same thing as evidence. That distinction matters more now because the signals themselves are becoming remarkably easy to manufacture.

Confidence Does a Lot of Heavy Lifting

I’ve always found it interesting how much credit we give confidence. Someone speaks clearly, answers immediately, never seems uncertain and appears to know exactly where the conversation is going, and we tend to read that as competence. Sometimes that’s exactly what it is. Sometimes the person simply has tremendous confidence and very little else getting in the way.

We see it with politicians, salespeople, television experts and people online who’ve discovered that putting a microphone in front of themselves somehow adds twenty points to an opinion. We also see the opposite. Somebody who says, “I’m not completely sure, but this is what the evidence seems to suggest,” may actually be behaving more responsibly than somebody offering an absolute answer. Unfortunately for them, uncertainty doesn’t sound nearly as impressive.

Confidence and accuracy are two different things, even though our brains have a habit of bundling them together. Presentation influences how much scrutiny we give the underlying claim. If somebody sounds uncertain, we naturally start checking. If they sound like they couldn’t possibly be wrong, sometimes we stop checking when we should probably be doing exactly the opposite. AI makes that especially easy to see because confidence can now be produced on demand.

AI Didn’t Create This Problem

Ask a modern AI system almost any reasonable question and you can get a clean, organized explanation in seconds. It may include examples, qualifications and enough detail to make you think, Well, that certainly sounds right. Quite often, it is right. That’s why these systems are useful, and I use them every day.

But the more I use AI, the more important one distinction has become to me: the quality of an explanation and the quality of the evidence behind it are separate questions.

An answer can be wonderfully clear and still contain a factual mistake. It can take uncertain evidence and explain it as though the issue is settled. It can misunderstand a source and then describe that misunderstanding so smoothly that nothing in the language itself warns you something has gone sideways. Humans do all of that too, of course. The difference is that AI can produce polished explanations almost instantly and at enormous scale, which makes our old tendency to trust fluency and confidence a little more consequential.

That’s why I don’t think the usual warning that “AI can be wrong” gets us very far. So can Google, a newspaper, an expert, somebody we trust or, for that matter, me. The useful question isn’t whether a source is capable of making a mistake, because everything is. What matters is what we’re using as our reason for believing the answer in front of us. If the reason comes down to it sounded really convincing, we probably haven’t finished the job.

Google Gives Us a Good Example

Google makes this particularly interesting because most of us have spent years treating search as a route toward evidence. Traditionally, Google showed us pages. We still had to decide whether those pages were any good, but there was at least a visible separation between the search engine finding information and somebody else making the claim.

AI Overviews blur that separation. Google can now give us a synthesized answer at the top of the page and place citations beside it. Visually, that’s powerful because we see the statement and the source together, and it’s very easy to make the small mental jump from there’s a citation to this has been verified. The problem is that those two things don’t always mean the same thing.

A 2026 study by Haofei Xu, Umar Iqbal and Jacob M. Montgomery examined 55,393 Google searches over a 40-day period and broke the resulting AI Overviews into 98,020 individual factual claims. Their analysis found that 11% of those claims weren’t supported by the pages Google cited for them, with omission—the cited material simply not establishing the claim—being the dominant failure mode. (Xu, Iqbal and Montgomery, 2026)

That finding needs to be handled carefully because unsupported doesn’t mean false. The researchers didn’t prove that 11% of AI Overview claims were wrong. What they showed is that the presence of a citation beside an AI-generated statement doesn’t automatically mean the cited source supports that particular statement. That’s a narrower conclusion, but I think it’s an important one.

The citation itself is an authority signal. We see it and relax a little. I do it too. The answer looks sourced, so unless the subject really matters, opening the source can begin to feel unnecessary. Sometimes that’s perfectly reasonable; I’m not going to turn checking a dinner recipe into a federal inquiry. But when the claim matters, clicking through and seeing what the source actually says is still a remarkably useful little habit.

Then We Bring Ourselves Into the Conversation

Even perfect sources wouldn’t solve the whole problem, because information doesn’t arrive in an empty room. We’re already there, carrying opinions, experiences, loyalties, fears, suspicions and a fairly impressive collection of things we’d like to be true.

If a claim supports something I already believe, I don’t always examine it with quite the same enthusiasm I’d bring to a claim telling me I’m wrong. I’d like to say I do. I don’t. Something that fits my existing view can feel plausible almost immediately, while contradictory evidence somehow arrives facing a much more demanding entrance exam.

That’s confirmation bias in ordinary clothes. It isn’t necessarily a conscious refusal to look at evidence, and it certainly isn’t something only “other people” suffer from. It’s one of the reasons I’ve learned to become a little suspicious whenever an explanation fits my existing opinion almost perfectly. Maybe I was right. I’d just prefer to know why.

There’s another signal I don’t think we should dismiss entirely, and that’s gut instinct. Most of us know the feeling when something just seems off, even before we can explain why. Sometimes that instinct is experience recognizing a pattern faster than the conscious mind can put words around it, and I’ve learned not to ignore that completely. But I don’t think “it feels right” is where the thinking should end either. A gut reaction can be a reason to look closer. It isn’t evidence by itself.

Social proof deserves the same kind of treatment because other people’s behaviour really does contain information. If thousands of people are discussing something, that tells me it has attracted attention. If several people whose judgment I trust independently think something is worth reading, I’m more likely to read it. There’s nothing irrational about that, and pretending other people’s judgment never matters would be just as silly as blindly following the crowd.

The mistake comes when we quietly swap one kind of evidence for another. Imagine, purely as an example, that I see a dramatic post claiming AI will replace most office jobs next year and the post has 40,000 shares. That number is hypothetical; I’m not sneaking a statistic past you. Those shares tell us that a lot of people found the claim interesting, frightening, convincing or worth passing along. They might tell us something about the reach of the person who posted it. They don’t tell us whether AI is actually going to replace most office jobs next year.

Popularity tells us something real. It just doesn’t tell us everything we sometimes pretend it does.

Authority Matters, but It Doesn’t Settle the Question

None of this means authority is useless. Quite the opposite. Expertise is real, and pretending everybody’s opinion deserves equal weight would be a strange way to solve the problem.

If my furnace quits, I’d rather hear from somebody who’s repaired furnaces for twenty years than somebody who watched three YouTube videos yesterday. If I’m trying to understand cancer treatment, an oncologist’s opinion should carry a great deal more weight than something I overheard at the gas station. We rely on expertise because nobody can know everything, and most of the time that reliance is entirely sensible.

The complication is that authority has context. Someone can be highly qualified and still be speaking outside their specialty. Experts can disagree. Evidence can change. A careful scientific conclusion can also travel through enough headlines and social posts that, by the time it reaches us, the researcher who produced it might barely recognize it.

So I don’t think the answer is to stop trusting experts. Usually, genuine expertise deserves substantial weight. I just want to know what kind of weight I’m giving it and why. Is the person actually qualified in this area? Are they describing evidence or giving an interpretation? Is there broad agreement among people who know the subject, or is there serious disagreement that somebody has politely forgotten to mention? Asking those questions doesn’t weaken expertise. It helps us use it properly.

The Answer Isn’t to Distrust Everything

This is where scepticism can become every bit as lazy as gullibility. Once you start noticing how often confidence, authority and social proof can mislead us, it’s tempting to decide that experts are biased, studies can be wrong, Google makes mistakes, AI hallucinates, journalists get things wrong and governments certainly aren’t immune either. All of that can be true, but it doesn’t get us anywhere if the conclusion is that nothing deserves trust.

Modern life depends on reasonable trust. I don’t inspect the engineering calculations on a bridge before I drive across it. I don’t reproduce clinical trials before taking medication. I don’t personally test the municipal water supply every morning. Somewhere along the line, I’m relying on expertise, institutions, accumulated evidence and people doing jobs I couldn’t possibly reproduce myself.

So the useful skill isn’t permanent suspicion. It’s recognizing when a claim matters enough that the usual shortcut isn’t good enough, and the threshold changes with the consequences. Somebody tells me a movie is terrible and I believe them? Worst case, I miss a decent movie. Somebody tells me an investment is guaranteed to double, a medical treatment is useless, a person committed some terrible act, or that I should completely change my business because Google supposedly changed something yesterday? Now I’d like a little more than confidence and a screenshot.

I’d add one other signal I’ve learned to watch in myself: how badly I want the claim to be true. If something makes me instantly furious, frightened, vindicated or delighted because it proves exactly what I already believed, that’s probably not the moment when my critical thinking is operating at some previously undiscovered level of perfection. Quite often, that’s when I need to become slightly more annoying and ask myself where the claim came from, what the evidence actually shows and whether I’d apply the same standard if the conclusion went the other way.

That’s really what the second question is for:

How do we know this?

I’m not suggesting we demand absolute proof every time somebody opens their mouth. I’m talking about deciding whether we have enough reason to believe something for the purpose at hand. Sometimes the evidence is strong and nothing changes. Sometimes it’s good but much more qualified than the headline suggested. Sometimes what looked convincing five minutes ago gets considerably less impressive once you follow it back to the source. Any of those results is more useful than never asking.

It’s Also Changed How I Use AI

This is one reason I’ve become much more interested in using AI as a conversation than as an answer machine. Give me an answer and I’ll work with it, but I don’t want to stop there when the subject matters. I want to know what supports it, where the uncertainty is, whether there’s another credible interpretation and what evidence would change the conclusion. Sometimes I’ll challenge the answer simply because I want to see whether it survives being challenged.

That part is important to me because I tend to do the same thing before I make up my own mind. I want to look at both sides of the coin if there genuinely are two sides worth looking at. If the other argument has a point, I’d rather acknowledge it than pretend it doesn’t exist. Sometimes considering the other side changes my conclusion; sometimes it just makes me more confident that the original one holds up. Either way, I’d rather arrive there after testing it.

AI can be useful in that process, but there’s an obvious limit. If AI tells me something and then I ask the same AI whether it was right, I haven’t suddenly created two independent sources. When the stakes justify it, eventually I need to leave the conversation and look at the evidence itself. What the conversation can do is stop me from treating the first answer as the finish line and get me examining the answer instead.

For me, that’s a much more interesting use of the technology, and I suspect it matters more than finding some mythical perfect prompt.

When Answers Get Cheap, Judgment Gets Expensive

For most of human history, getting information was the difficult part. You had to know where to look, find somebody who knew the subject, track down a book or spend considerable time doing the research yourself.

We’ve almost reversed that problem. Information is everywhere, explanations are everywhere and opinions are certainly everywhere. Generative AI can now produce a polished explanation of almost anything in seconds, and those systems are going to get better at sounding convincing, not worse.

That’s an extraordinary advantage if we use it well. I can learn faster, explore an unfamiliar subject, compare ideas and ask questions I might not have thought of on my own. I wouldn’t use these tools every day if I didn’t think they were useful.

But once everybody can get a convincing answer, the real challenge shifts. We have to decide whether the answer deserves confidence, whether the source actually supports the claim, whether we’re being persuaded by evidence or by presentation and whether we’d apply the same standard if the conclusion contradicted something we already believed. That isn’t a neat little checklist so much as a habit of slowing down when slowing down actually matters.

There’s no formula that will make us right every time. Evidence can be incomplete. Experts can disagree. New information can change a conclusion that looked perfectly reasonable yesterday. Sometimes two intelligent people can examine the same evidence and still come away disagreeing, and I’m increasingly comfortable with that. “I don’t know yet” is a perfectly respectable answer when the evidence doesn’t justify anything stronger.

What worries me more is certainty we haven’t earned, especially as producing persuasive language, images, audio and video keeps getting easier. Search engines will answer more questions directly. AI will explain almost anything we ask. People who want our attention, our money, our vote or simply our outrage will have access to the same tools.

I don’t think the answer is to become cynical enough to distrust everybody. That would replace one bad shortcut with another. I think we need to get better at noticing the moments when belief has become a little too easy: when a claim matters, when it fits our existing beliefs almost perfectly, when everybody around us seems to be repeating it, or when somebody gives us a wonderfully confident answer to a complicated question.

That’s when I want the second question.

How do we know this?

Maybe the answer holds up exactly as it was presented. Great. Now we have a reason to believe it that’s better than confidence, popularity or a citation we never opened. And in a world where convincing answers are getting cheaper by the day, being able to make a decent judgment about them is becoming a hell of a lot more valuable.

Ask the second question first.