How to Tell If Someone Is Lying: What Actually Holds Up

Most of what you've been told about spotting a liar is wrong, and it's wrong in a way that's been repeated so often it feels like fact. Here's what the research actually supports, and a method that outperforms watching for "tells" entirely.

Start with the uncomfortable finding. When researchers test ordinary people — and even trained professionals like police officers and customs agents — on their ability to tell truth from lies, accuracy comes out barely above chance. Across dozens of studies pooled together, people get it right about 54% of the time. A coin flip gets you 50%. That gap is almost nothing, and it holds even when people are told specifically to watch for lying.

This matters because it means the popular checklist — avoiding eye contact, touching the face, crossing arms, fidgeting — isn't just imperfect. It's close to useless, and worse, it makes you confident in judgments that are no better than guessing.

Why the classic "tells" don't hold up

Eye contact is the biggest one. It shows up in nearly every article and every crime drama, and it has the least support of any cue tested. Some liars avoid eye contact from guilt or anxiety. Others deliberately hold eye contact for longer, because they know it's being watched and they're compensating. Cultural background changes baseline eye contact substantially. The result is a cue that points in both directions depending on the person, which means it points nowhere.

The deeper problem with all physical "tells" is that they measure anxiety, not deception. A nervous person telling the truth — worried about being disbelieved, aware of the stakes, uncomfortable being scrutinised — produces exactly the same physical signals as someone lying: faster speech, more movement, longer pauses, a higher pitch. You cannot tell those two states apart by watching a body. You need something that responds to the content of what's being said, not to how anxious the person feels saying it.

What actually has some support

Two things perform meaningfully better than physical tells, and both work by testing content rather than watching behaviour.

1. Compare against a real baseline

The single most useful shift is dropping the idea of a universal tell and instead learning how this specific person behaves when they're relaxed and telling the truth about something ordinary. Some people are naturally fidgety. Some people naturally avoid eye contact. What matters is a change from someone's own normal — a person who usually holds eye contact easily suddenly not doing so, in a specific moment, about a specific topic, is worth noting. The same behaviour in someone who never held eye contact to begin with tells you nothing.

This means the method only works with people you already know reasonably well. It is close to useless on a stranger, which is exactly why it outperforms techniques that claim to work on anyone in seconds.

2. Ask them to tell it out of order

This is the technique with the strongest research backing, used in real investigative interviewing: ask someone to recount events not in the order they happened, but backwards, or starting from the middle.

A true memory is stored as a real, lived sequence and can be traversed in either direction, if awkwardly. A fabricated story is usually rehearsed as a single forward narrative — a script — and scripts are dramatically harder to reverse on the spot. Liars asked to go backwards make more errors, hesitate more, and are more likely to contradict details they gave the first time.

Asked naturally, mid-conversation, this reads as simple curiosity rather than an interrogation, which is exactly what makes it usable in ordinary life.

Two more content-based signals worth knowing

Missing self-correction. Real memories come out messy — people correct themselves, add a detail they forgot, or say "actually, it was Tuesday, not Wednesday." A rehearsed account tends to arrive polished from the first telling and stays identical on every retelling, because it's a script being replayed rather than a memory being searched. Perfect consistency, oddly, is more suspicious than small natural corrections.

Unprompted denial. Volunteering detailed innocence before being asked — explaining in depth why they wouldn't have done something nobody accused them of — is a well-documented pattern. It doesn't prove anything on its own, but it's worth noticing when it happens.

What to do with a suspicion

Don't act on one gesture in one moment. Ask open questions and let the person add detail without interrupting — liars tend to give less spontaneous detail than truth-tellers, because inventing detail on the spot is cognitively harder than recalling it. Come back to the same event later in the conversation, or on another day, framed as a different question, and check whether the account holds together. A true story generally survives being asked about twice. A constructed one tends to shift.

Reading people is a bigger subject than lying

The Unfair Advantage covers the full mechanics of how people signal, persuade and get read — including where most popular psychology gets it wrong. Chapter One is free and complete.

Read chapter one free →