Can You Tell If an Employee Is Lying? What the Research Actually Says

You are sitting across from an employee who has just been accused of something serious. They fold their arms. They look at the corner of the ceiling before answering. They laugh at the wrong moment.

Everything you have absorbed from television, and even misinformed interview and interrogation training, tells you that you just learned something. You didn't. And the research on how to tell if an employee is lying is more settled than most people conducting workplace investigations realize. The good news is that what replaces it is teachable, defensible, and produces better information, and better opportunities to detect deception than pseudoscience or a hunch ever did.

The short answer: no, and neither can the experts

The largest synthesis of deception-detection research pooled 206 studies and found a weighted average accuracy of 53.46%. A second meta-analysis of 156 studies put it at 54.22%. Chance is 50%.

The finding that should stop you is the second one. Professionals whose jobs depend on this skill—police investigators, detectives, judges, and psychologists—averaged 55.51%, statistically indistinguishable from the undergraduates at 54.22%. In Bond and DePaulo's data, experts did not outperform laypeople at all.

Some studies find worse. In one, police officers scored around 47% while lay observers hit 59%. In another, 60 officers performed below chance while rating their own accuracy as high. Experience does not improve accuracy. It simply improves confidence.

The reason is not that investigators are unobservant. Three separate meta-analyses of nonverbal and paraverbal cues found only a handful of reliable differences between deceptive and truthful accounts, all of them small. There is no physical signature of lying—the signals are faint, inconsistent between people, obscured by lack of context, and swamped by everything else in the room.

And a workplace investigation is a room full of everything else. Being accused of harassment when you did nothing is frightening. Being accused when you did it is also frightening. The fear and the emotions may look identical from across the table.

Why "reading body language" makes workplace investigations worse

What should change the training you buy?

A meta-analysis of 30 deception-detection training studies found that training built on nonverbal cues, paraverbal cues, or feedback produced small or non-significant effects. Only training on verbal content—what the person actually says—produced a meaningful improvement. The authors' conclusion is blunt for a journal: practitioners "should not base their trainings on these unreliable cues." This includes all forms of interview and interrogation trainings.

It gets worse than "doesn't help." When researchers trained participants in the Reid Technique using Reid's own materials and then tested them, the trained group performed worse than untrained controls—a negative training effect. A follow-up with real investigators who had genuine training and experience found the mechanism: training produced a lie bias, not better discrimination. They judged more people as deceptive. They were not more often right. If you go looking for liars, you will find liars.

You have not gained a skill or a tool for the proverbial toolbox. You have acquired a bias—and lost the ability to notice you have one. It is the reason lie detection training is the one skill you cannot learn from your own mistakes—the feedback never arrives (The popular claim that eye movement direction reveals lying has been tested directly, too. Trained and untrained groups showed no difference in accuracy or confidence. There is nothing there.).

In an HR context, that bias costs you in three specific ways:

1. You disbelieve a truthful complainant. The person reporting harassment is often distressed, inconsistent about sequence, and reluctant to repeat details. Every one of those reads as "deceptive" to someone trained on demeanor.

2. You build a finding on an impression. Impressions do not survive a rebuttal, an appeal, or an arbitrator.

3. It ends up in the file. "The employee appeared evasive and would not maintain eye contact" is a sentence that will be read aloud back to you, slowly, by opposing counsel—which is the whole reason conducting a workplace investigation interview without creating legal liability starts with what you write down, not what you notice.

Meanwhile, the person actually lying to you is not fidgeting. They prepared. They rehearsed. They have an information strategy.

What is a decision-maker supposed to do with "I think he's lying"?

Push the demeanor claim one step downstream and it comes apart even further.

You finish the interview convinced the employee lied. You write it up. Now someone else has to act on it—an HR director, a general counsel, or a hearing officer—and the decision in front of them may end a career. Here are the three questions they should be asking yourself, and what happens to a body-language finding under each.

What is this based on? "He crossed his arms, he wouldn't hold eye contact, and I've been to a course." A course is not a measurement. The question was never whether you were trained; it is whether the thing you were trained to do works. On that, the evidence is not ambiguous—and for at least one widely sold method, the training measurably moved accuracy in the wrong direction.

How often is it right? This is the question almost nobody asks, and it decides everything. Give yourself the benefit of every doubt and call your accuracy 54%—the optimistic end of the research. Suppose one in five people you interview is genuinely being deceptive. Run 100 interviews through those numbers and you will flag about 48 people as liars. Roughly 11 of them will actually be lying.

Three out of four people you call a liar are telling you the truth.

That arithmetic is an illustration, not a measured rate. The real base rate of deception in your organization's investigations is unknown—which is itself part of the problem. But move the assumptions around and the conclusion barely shifts: when the behavior is uncommon and the detector is barely better than a coin flip, most of what it catches is noise. That isn't an investigator failing. It's simple math.

Would it survive review? Put a demeanor finding in front of an arbitrator, an appeals panel, or opposing counsel and it is one question deep. "What is the scientific basis for concluding that crossed arms indicate deception?" There isn't one, the answer is published in a dozen journals, and the rest of your report now inherits the credibility of its weakest sentence.

None of this makes you a bad investigator. It makes the tool a bad tool—and it is worth understanding how bias shapes investigative decision-making before trusting your own certainty. The fix is not to try harder at reading people. It is to hand the decision-maker something they can actually stand behind.

Stop reading the person. Start testing the account

The shift that makes investigative interviewing work is moving the burden of proof off the person's behavior, mannerisms, ticks, and onto the content of what they say.

The statement has details in it—times, sequences, people, actions, locations, capability, opportunities, what was said, and what came first. Those details can be compared against a badge log, a calendar, a Slack export, a shift schedule, video surveillance, or another witness. That comparison is corroboration. A crossed arm is not.

This is why investigators spend so much effort on getting a complete and uncontaminated account before doing anything else:

1. Planning. Start with an inventory of what you actually hold—not what you believe, what you can put in front of a person. An access badge log proves a card was used. It does not prove who carried it.

Interview objectives come out of that inventory of evidence. An objective is not "get him to admit it." An objective is a specific thing the account has to cover before a piece of evidence means anything: obtain her account of where she was between two and four, in her words, before she knows the badge log exists. That is checkable. That is achievable. And you know at the end of the interview whether you got it.

Then sequence it—what you ask first, what you hold, and in what order you would disclose if you disclose at all. This is where most of the Strategic Use of Evidence (SUE) actually happens. SUE is a planning technique and a questioning technique first; disclosure is the last move and an optional one. Plenty of good interviews end without you ever putting the document on the table, because the account either matched or it didn't, and you have that either way.

2. Free recall first. Ask an open invitation—"Tell me everything you remember about that afternoon, starting from before you got to the office"—and then stop talking. The uninterrupted account is the most reliable version you will ever get, and every question you ask before it narrows what you receive.

3. Do not reveal what you already have. The moment you say "we have the emails," you have handed over the shape of the answer. Hold it. This is the step your plan exists to protect through the use of the evidence framing matrix (EFM) and it is the one that collapses under pressure when there is no plan.

4. Then compare. With the account fixed and documented, disclose what you hold and let the person address the difference between the two. Sometimes there is no difference and you have just cleared someone properly. Sometimes the difference is the case through a statement-evidence contradiction.

Now look at what that sequence leaves behind. A plan. Stated questioning objectives tied to specific evidence. A recorded free narrative taken before any disclosure. A documented disclosure order. Every one of those is a decision you made on purpose and can explain, in plain language, two years later to a judge, an arbitrator, or a board that has never met you.

That is the whole point. This is the only way to test an account that survives an inquiry—not because it feels more rigorous, but because it produces a solid record. The behavioral read produces nothing. There is no plan, no objective, no artifact, and nothing to hand anyone except your say-so about what you noticed or what you think you noticed while using bad tools. Asked to justify it afterward, you have a memory of a moment and a conclusion you reached inside your own head.

And it is one of the very few interview interventions that has actually been shown to improve accuracy. In a controlled study, interviewers trained in the Strategic Use of Evidence classified subjects correctly 85.4% of the time, against 56.1% for untrained interviewers. Hold that number next to the 54% ceiling on demeanor judgment. The gain did not come from reading people better. It came from structuring the interview so the account could be tested.

The contradiction is the deliverable

Here is what the Strategic Use of Evidence produces that a demeanor judgment never can.

If the account contradicts a badge log, a timestamp, a calendar entry, or another witness, that contradiction is a fact. You can write it down. You can hand it to a decision-maker as a finding. You can be questioned about it under oath and the answer holds—because it does not depend on your read of anybody. The record says one thing, the statement says another, and both are in the file.

Compare the two sentences you could put in a report:

• "The employee appeared evasive and broke eye contact when asked about the meeting."

• "The employee stated he left the building at 4:00 PM and had no contact with the complainant afterward. The badge log records his exit at 6:12 PM, and two messages were sent from his account at 5:40 PM."

Only one of those survives a cross-examination. Only one of them is evidence at all.

This is not just a legal-defensibility argument, either. When researchers surveyed 127 California prosecutors on which interview techniques actually help them, "identify contradictions within the suspect's story" came back as the third most effective of 29 techniques. And when the same prosecutors rated interview outcomes, a statement contradicting the evidence was valued essentially as highly as a full confession.

A documented contradiction is worth about as much as an admission. And unlike an admission, you can obtain it without ever pressuring anyone.

That study asked prosecutors for their perceptions, using a self-selected sample. It tells you what experienced downstream consumers of interviews value—not an observed measurement of what works in court.

What this actually looks like in the interview

Four changes, all of which you can make on your next interview:

1. Plan the interview in writing before you walk in. What do you know, what do you need, what order do you disclose in, and what will you do if the account contradicts the record.

2. Open with the invitation, then be quiet. The silence after an open question feels much longer to you than to the person filling it. Let it run.

3. Ask about the account, not about the accusation. "What happened next?" produces information. "Why should I believe you?" produces defensiveness and nothing else. If you want the specifics, we keep a working list of HR investigation interview questions worth asking—and the ones to avoid.

4. Do your comparison afterward, on paper. Not live, not in your head, and not while the person is still in the room reading your face.

The questioning method that generates the most usable detail for exactly this purpose — the cognitive interview — has the deepest evidence base in the field. Across 46 studies it produced a large increase in correct details compared with a standard interview, with average accuracy differing by less than 1% and no significant increase in fabricated detail. More information, not worse information.

More correct detail is the entire ballgame. It is what you check. It is what makes your investigation findings hold.

Notice what is missing from that list of four. Nowhere do you decide whether they seemed truthful or were more likely guilty. That judgment isn't yours to make in the room, and making it early is how you create bias and where investigations go wrong.

The investigation trade-off nobody mentions

Science-based interviewing takes longer.

In a field study of investigators trained in these methods, interviews after training ran an average of 107 minutes against 72 minutes before—roughly half again as long. That is a real operational cost.

Two honest caveats on that research:

• Those were police interviews in criminal cases, not workplace investigations. The communication mechanism transfers; the setting may not.

• The benefits ran indirectly—training changed which tactics investigators used, and the tactic use is what drove cooperation and disclosure.

If your goal is a fast admission before lunch, none of this is for you. If your goal is a finding that survives an appeal, the extra 35 minutes is the cheapest insurance you will ever buy.

What you are actually after is not a confession

When those California prosecutors rated what they most wanted out of an interview, a true and accurate account outranked a full confession. The top nine techniques they rated most effective were all science-based ones—active listening, showing concern, finding common ground, confronting with actual evidence, and identifying contradictions. The techniques they rated worst were the confrontational staples: insulting the suspect, rapid-fire questioning, threatening consequences, and instilling hopelessness.

That is the whole case for information-gathering rather than confession-based interviewing, stated by the people who have to use the results.

The parallel in a workplace investigation is exact. Your deliverable is not an admission. It is a defensible, accurate, well-documented account of what happened—one that holds up when the employee lawyers up, when the finding is appealed, and when someone reads your notes two years later in a deposition.

You do not get there by deciding who looks guilty. You get there by running an interview designed to produce checkable information.

Where to start with better interviewing

Drop the demeanor question. Replace it with one habit on your next interview: ask one open question, then say nothing until they stop talking. That single change will get you more usable information than every body-language checklist in circulation. It is also the hardest one to do well, which is why active listening is not what you think it is.

If you want the rest of the method—rapport as trained behavior, listening as an evidence-gathering instrument, and how to structure an account so it can be tested—I'm teaching a free 30-minute session on Thursday, October 8 at 1:00 PM Central, including a live interview run in front of you so you can watch the technique work in real time.

Frequently asked questions on lie detection

Can you tell if an employee is lying from body language?

No. Across 206 studies, people identify lies and truths correctly about 53% of the time—barely above chance—and professionals whose jobs depend on it score no better than untrained observers. Three separate meta-analyses of nonverbal cues found only a few reliable differences between liars and truth-tellers, all of them small.

Does deception-detection training help investigators?

Training built on nonverbal or paraverbal cues shows small or non-significant effects across 30 studies. One widely sold method produced a negative effect: trained participants were less accurate than untrained controls. What training does reliably produce is a bias toward judging people as deceptive, which raises confidence without raising correctness. Training on verbal content is the exception—it works.

What should I do instead in a workplace investigation interview?

Focus on the account rather than the person. Get a complete, uninterrupted version of events before asking any narrow questions, and hold your evidence back while you do it. Then use the Strategic Use of Evidence approach: test the account against what you already know. It is not a gotcha moment—it is a way of putting a statement and a record side by side. If the two contradict, that contradiction is a fact you can document, report to a decision-maker, and testify to in a deposition. A body-language impression is none of those things. In controlled research, interviewers trained in this approach reached 85.4% accuracy against 56.1% untrained.

Is "the employee seemed deceptive" defensible in a termination decision?

It is very difficult to defend. There is no scientific basis for inferring deception from demeanor, so the finding cannot be supported under questioning, and it tends to damage the credibility of the rest of the report. A documented contradiction between an account and a verifiable record is defensible; an impression is not.

How long should a workplace investigation interview take?

Longer than most people expect. In field research with trained investigators, interviews averaged around 107 minutes compared with 72 minutes before training. The additional time is where the checkable detail comes from.

Lie Detection Sources

Bond, C. F., & DePaulo, B. M. (2006). Accuracy of deception judgments. Personality and Social Psychology Review, 10(3), 214–234.

Brimbal, L., et al. (2026). Prosecutors' perceptions of interview and interrogation techniques. Psychology, Crime & Law.

Hauch, V., Sporer, S. L., Michael, S. W., & Meissner, C. A. (2016). Does training improve the detection of deception? A meta-analysis. Communication Research, 43(3), 283–343.

Kassin, S. M., & Fong, C. T. (1999). "I'm innocent!": Effects of training on judgments of truth and deception in the interrogation room. Law and Human Behavior, 23(5), 499–516.

Hartwig, M., Granhag, P. A., Strömwall, L. A., & Kronkvist, O. (2006). Strategic use of evidence during police interviews: When training to detect deception works. Law and Human Behavior, 30(5), 603–619.

Memon, A., Meissner, C. A., & Fraser, J. (2010). The Cognitive Interview: A meta-analytic review and study space analysis of the past 25 years. Psychology, Public Policy, and Law, 16(4), 340–372.

Russano, M. B., et al. (2026). Legal and Criminological Psychology, 31, 273–300.