How to Tell When Someone Is Lying: What Research Actually Shows

You are worse at spotting lies than you think, and so is everyone else. Across hundreds of studies, people land around 54 percent accuracy, which is a coin flip with extra confidence. The reason is not that lies are undetectable. It is that almost everything you were taught to watch for does not work, and the things that do work are verbal rather than visual. Here is what the research actually supports.

Fair warning: this article will make you less certain, not more. That is the point. The people who are dangerous with this topic are the ones who came away from a documentary convinced they can now read anyone.

Why Is Eye Contact Not the Signal You Think It Is?

Because it has been tested repeatedly, and it does not hold up.

The definitive work here is a meta-analysis by Bella DePaulo and colleagues, published in Psychological Bulletin in 2003. They gathered every behavioural cue that had been examined in at least five deception studies, 50 cues in total, and asked which ones actually distinguished liars from truth-tellers.

Only 14 of the 50 showed a significant relationship with deception. That is 28 percent. And the average effect size across those was around d = 0.25, which in plain terms means small: a real but faint difference that would not let you call any individual case with confidence.

Gaze aversion was not among the reliable cues. The relationship between lying and how much someone looks at you came out non-significant. Blinking did not work either. Neither did most speech disturbances.

Now hold that against what people believe. The Global Deception Research Team surveyed beliefs about lying across 58 countries. In 51 of them, the majority named gaze aversion as a sign of lying. It is close to a human universal, and it is close to universally wrong.

Why the myth survives. Three reasons, and they reinforce each other.

• It is taught. Parents say look at me when I'm talking to you. The link between shame and looking away gets installed early, and shame does often produce gaze aversion. Lying just does not, reliably.
• Liars know about it. This is the part that flips the whole thing over. Someone deliberately deceiving you knows the folklore too, so they make a point of holding your gaze. If anything, deliberate steady eye contact is the more suspicious behaviour, and there is research examining exactly that inversion.
• It confirms itself. You suspect someone, you watch them, you notice a glance away, you feel confirmed. You do not count the times honest people looked away, because you were not watching them.

The same demolition applies to fidgeting, touching the face, crossed arms and most of what gets sold as body language. These are anxiety behaviours at best. Anxiety is not deception. An innocent person who thinks they are about to be disbelieved produces plenty of it.

What Do Trained Deception Researchers Actually Watch?

The field moved away from faces some time ago. The productive line of work, associated most closely with Aldert Vrij, starts from a different premise: lying is mentally harder than telling the truth, so instead of hunting for nervousness, you make the task harder and watch who struggles.

The reasoning is straightforward. A truth-teller retrieves. A liar has to invent a plausible account, keep it consistent with everything already said, monitor your reaction to see if it is landing, suppress the truth, and remember the invention for later. That is several jobs running at once, and Vrij set out the case for exploiting it in the Journal of Investigative Psychology and Offender Profiling in 2008.

What that produces, in ordinary conversation:

• Pauses in the wrong places. Not long pauses generally, which everyone does. Pauses before questions that should require no thought at all. If someone hesitates before naming who else was in the room, the hesitation is doing work.
• Answers that arrive too complete. A rehearsed account comes out smooth, ordered and finished. Real recall arrives in pieces, doubles back, and corrects itself.
• Over-explanation of things nobody questioned. Unprompted justification for a detail you had not asked about is a sign that the detail has been thought about in advance.
• Deteriorating quality under load. Ask something unexpected and the difference shows. A truth-teller adjusts and keeps going. A rehearsed account tends to stall, deflect, or repeat.

One honest caveat, because this literature has its own critics. The cognitive load approach has been challenged in recent methodological reviews, which argue the evidence base is thinner and more variable than early enthusiasm suggested. It is a better bet than watching eyes, not a solved problem. Treat it as improving your odds, not as a technique that works.

And the rule that matters more than any single cue: look for clusters. No individual behaviour means anything on its own. Several of these appearing together, at the same point in a story, on more than one occasion, is the only thing worth taking seriously.

How Does the Story Structure Test Work?

This is the most useful thing in the article, and it costs nothing to apply.

Real memories are a mess. When you genuinely remember something, you do not retrieve a clean narrative. You retrieve fragments in the wrong order, with sensory junk attached, and you assemble the story as you talk. That is why honest accounts wander. Someone remembers the smell of the room halfway through describing a conversation, mentions it, then has to find their place again.

Rehearsed accounts are the opposite. They were built as stories, so they come out as stories: beginning, middle, end, in order, no detours, nothing irrelevant. Everything present is load-bearing.

So the signature you are looking for is not nervousness. It is tidiness.

What genuine recall tends to contain:

• Irrelevant detail. The waiter had a bad cold. There was a mark on the table. None of it matters to the account and that is exactly why it is there.
• Self-correction. No wait, it was Thursday, because I had the dentist. People edit themselves mid-sentence when retrieving.
• Admitted gaps. I don't remember what happened after that. A liar tends to have full coverage, because gaps feel like weakness.
• Non-linear order. Jumping forward, then back to fill in.
• Unflattering material. Details that make the speaker look bad or confused. Invented accounts trend towards making the teller reasonable.

The reverse order question. One specific technique from the cognitive load research is easy to use and quite hard to beat: ask someone to tell it backwards. Start at the end and work back to the beginning. Retrieval from real memory copes with this reasonably well because the memory is not stored as a sequence. A rehearsed story is stored as a sequence, and running it backwards breaks the thing that made it easy to tell.

Do not do this as an interrogation. In ordinary life it looks like curiosity: hang on, back up, what happened just before that? Which brings us to the version you can use without sounding like a detective.

Which Verbal Cues Actually Signal Deception?

Language does more work than behaviour here, though the same caution applies: these are weak signals that matter in combination.

Distancing language. People putting linguistic space between themselves and an event. That woman rather than Sarah. The car rather than my car. Pronouns thinning out, especially I. It is not proof of anything, but a sudden shift from first person to passive constructions at one specific point in a story is worth noticing.

Tense shifts. An account of a past event drifting into present tense, or the reverse. Rehearsed material sometimes gets told the way it was practised rather than the way it was lived. A shift that lands on one particular section is more interesting than one that runs throughout.

Specific denials versus categorical ones. This is the most practically useful distinction on the list. Compare:

• "I did not send that message to her on Tuesday night."
• "I would never do something like that."

The first is a claim about a fact and can be checked. The second is a claim about character and cannot. People avoiding a direct lie often reach for the second, because a statement about the kind of person you are is not technically falsifiable. Watch for answers that assert character where a fact was requested.

Answering a question you did not ask. Related move. You ask where they were; you get told how busy work has been. The response is topically adjacent and factually empty.

Qualifiers stacking up. As far as I recall, to the best of my knowledge, I'm pretty sure. Everyone uses these. What is notable is a cluster of them arriving suddenly, on one topic, from someone who was speaking plainly a minute earlier.

What is common to all five is that the change matters more than the trait. You are not looking for someone who uses passive voice. You are looking for someone whose language changes shape at one specific point.

What Happens When You Ask for More Detail?

This is the single best test available to a normal person in a normal conversation, and it works because it exploits the difference between having a memory and having a script.

Ask for more about a specific part of the account. Not a challenge, just interest. What was the room like. Who spoke first. What were you doing with your hands.

Genuine memory expands. The person goes back to the experience and finds material that was there all along but had not come up. New sensory detail appears. They may correct something they said earlier. The story gets richer and slightly messier, because the source is a real event with unlimited detail available.

A rehearsed story repeats or flattens. They restate what they already said in slightly different words, add generic filler that could apply to any occasion, or push back on why you are asking. The script has a fixed amount of content in it, and asking for more does not produce more, because there is nothing behind it.

That contrast is far more informative than anything on a face, and it has a useful property: it is nearly impossible to defend against without preparing infinite detail in advance, which is not something people do.

How to use it without turning the conversation into an interview:

1. Pick one narrow point. Broad questions get broad answers. Ask about a single moment rather than the whole evening.
2. Ask about sensory or peripheral things. What was playing, how cold was it, who else was around. Peripheral detail is the part invented accounts do not bother to build.
3. Leave gaps in the conversation. Silence invites elaboration. People fill it, and what they choose to fill it with is informative.
4. Come back to it later. Days later, casually. Real memory tells it slightly differently each time because it is being reassembled. Rehearsed accounts tend to come back word-perfect, and word-perfect is itself a signal.
5. Stay warm. The moment you sound like you are testing them, you get defensiveness from honest and dishonest people alike, and defensiveness tells you nothing.

What Should You Do With This Information?

Hold it loosely. Here is the number that should govern everything above.

Charles Bond and Bella DePaulo pooled 206 documents covering judgments by 24,483 observers in Personality and Social Psychology Review in 2006. Overall accuracy was 54 percent. Broken down, people correctly identified 61 percent of truths and only 47 percent of lies, which means on lies specifically, observers did worse than guessing.

There is a bias hiding in that split worth understanding. We tend to believe people by default, so we pass most truths and miss most lies. Now notice what happens when you start suspecting someone: the bias inverts. You begin reading ambiguous behaviour as guilt, and your accuracy on honest people collapses. Suspicion does not sharpen you. It relocates your errors.

So the rules:

1. Treat everything here as a signal, never a verdict. The strongest cues in DePaulo's meta-analysis had small effect sizes. Nothing on this page justifies certainty about a single conversation.
2. Judge patterns over time, not moments. One inconsistent story is life. The same account changing across months, or promises that never once turn into anything, is data. That distinction is the whole basis of our guide on future faking.
3. Weigh what actually happened above what was said. Verifiable facts beat verbal analysis every time. If something is checkable, check it instead.
4. Do not accuse from an analysis. Saying your story sounded rehearsed produces a defensive person and no new information. Ask another question instead.
5. Watch what suspicion is doing to you. If you are running structural analysis on your partner's account of their Tuesday, the state of the relationship is already the finding, whatever the answer turns out to be.

One more thing, and it is the situation this article gets read for most often. If someone is not just lying but steadily rearranging what you remember, telling you that conversations did not happen, that you are too sensitive, that you are imagining it, that is a different problem with a different name and lie-spotting will not solve it. Our guide on the signs of gaslighting covers it. And if you want the honest limits of reading faces, our companion piece on micro-expressions explains what facial leakage can and cannot tell you.

What Else Do People Ask?

How accurate are people at detecting lies?

About 54 percent, which is barely above chance. Bond and DePaulo pooled 206 documents covering 24,483 observers in 2006 and found people correctly identified 61 percent of truths but only 47 percent of lies. On lies specifically, the average observer performs slightly worse than guessing, largely because we default to believing people.

Does avoiding eye contact mean someone is lying?

No. DePaulo and colleagues examined 50 behavioural cues across deception studies in 2003 and found the relationship between gaze aversion and lying was not significant. The belief is close to universal, with majorities in 51 of 58 surveyed countries endorsing it, and it is close to universally wrong. Deliberate liars often hold eye contact deliberately because they know the folklore too.

What is the most reliable sign that someone is lying?

There isn't one single reliable sign, which is the honest answer. The strongest available approach is asking for more detail about a narrow part of the account. Genuine memory expands and gets messier because a real event has unlimited detail attached. A rehearsed story repeats, adds generic filler, or resists the question.

Why do real stories sound messier than lies?

Because memory is not stored as a narrative. Genuine recall arrives in fragments, out of order, with irrelevant sensory detail attached, and gets assembled while the person talks. Rehearsed accounts were built as stories, so they come out linear and complete. Suspicious tidiness is a more useful signal than nervousness.

Should I confront someone if I think they are lying?

Not on the strength of behavioural analysis alone. The cues that survive research have small effect sizes and cannot support certainty about one conversation. Accusing produces defensiveness from honest and dishonest people alike, and it teaches the person to manage their behaviour around you. Ask another question, verify what is checkable, and judge patterns over time.

Sources: Bond, C. F., and DePaulo, B. M. (2006), "Accuracy of deception judgments," Personality and Social Psychology Review, 10(3), 214-234, pooling 206 documents and 24,483 observers (journals.sagepub.com/doi/10.1207/s15327957pspr1003_2). DePaulo, B. M., Lindsay, J. J., Malone, B. E., Muhlenbruck, L., Charlton, K., and Cooper, H. (2003), "Cues to deception," Psychological Bulletin, 129(1), 74-118. Vrij, A., Fisher, R., Mann, S., and Leal, S. (2008), "A cognitive load approach to lie detection," Journal of Investigative Psychology and Offender Profiling, 5(1-2), 39-43 (onlinelibrary.wiley.com/doi/10.1002/jip.82).

For self-reflection and educational purposes only. Nothing here is a diagnosis or a substitute for professional mental health support. Only a qualified clinician can assess or diagnose any condition. If you feel unsafe, please contact a local domestic abuse or crisis helpline.