Lie Detection Training: You Can't Learn From Your Mistakes

You cannot learn from your mistakes, despite what you’ve been told. We repeat the adage like scripture: experience makes you better; you learn from your mistakes. When it comes to lie detection in investigative work it’s mostly false. You cannot learn from a mistake you never measured. And anxiety-based, behavior-based lie detection is built—whether its sellers admit it or not—to ensure you never measure it.

Bad lie detection training creates investigative risk and emboldens confirmation bias

Every professional who actually gets better keeps score, otherwise, how would you know? A hitter knows his batting average. A closer knows his save percentage. A kicker knows exactly how many field goals he’s missed from forty yards and out. Those numbers exist because someone counted every outcome—the hits and the misses/mistakes. That is the entire point. You cannot have a real measurement or batting average if you only write down the hits.

Now ask the interrogator who swears they can read body language what his accuracy is. Not his confidence, but his accuracy. How many times were they certain and wrong? They can’t tell you. They never wrote the misses down. Now, let me show you what that costs, because it once cost me.

The Night We Had the Shooter

Years ago, the detectives assigned to the scene told me we had our shooter, and the evidence seemed to agree with them. There was a grainy video with the sound of "gunshots" as a lone white car drove by. There were shell casings on the ground; there were holes in a house. There was no doubt that a drive-by had occurred. The white vehicle had been identified by the victim at the house. The man they brought me was the guy the victim found creepy—tied to the house, pursuing a love interest with a woman who lived there, our victim, who had rejected his advances. The guy kept driving by, again and again. Our guy would drive by hoping to see her in the front yard so he could stop and talk.

I walked into the interview room armed with all of it. I was also armed with something else I believed was just as solid: training in Neuro-Linguistic Programming (NLP), delivered by a PhD, sold to me as scientific with years of law enforcement field use behind it. I got my baseline as trained; I knew where I stood. And as I worked through the suspects account, NLP told me, clearly, that he was lying about key facts. The cues were everywhere. He was building his story in real time, exactly the way I’d been taught a liar does, not pulling from memory. So I did what I’d been taught to do once I was sure of this individuals guilt: I moved for a confession. I have NLP lies and that was supported by physical evidence. I’d never learned the Strategic Use of Evidence on how to use the evidence I had. I had no research-supported way to test his statement credibility against the case facts. Folksy wisdom said confessions are the queen of all evidence anyways (besides, no one would ever confess to a crime they didn't commit). Confidence was high and I began using the tools I was trained on.

After speaking with our suspect for a while, I accused him, and he denied it. Despite my evidence, I just told him we had to know why. I was communicating in steps, of course, running a 50+-year-old interrogation playbook (no update needed, I guess). He said I had the wrong person in custody and the dance continued. I stepped out for a break for us both. I always kept my phone off during any interview or interrogation, so it was only then that my fellow detectives reached me. They’d been watching the video evidence some more. Several minutes after the white car passed, an entirely different vehicle—a different color, obvious even on bad footage—drove by, and that was where the real gunshots came from. What they’d first called gunshots, the sounds they had fed me going in, turned out to be the backfire of the man’s sporty car exhaust. The people in the sedan were most likely rivals of the men who lived at the house. The “creepy guy” was just a young man trying to find love in the worst possible way.

Physical evidence doesn’t lie to us. Pseudoscientific NLP did. And it had lied to me with total confidence—a baseline, a read, a certainty; all of it pointed at the wrong man. My NLP "tool" had given me a false positive.

Physical evidence can’t lie. NLP did — and it lied to me with total confidence.

That’s when the real question landed. Not “was I wrong this time,” but how many times had I been wrong before and never known it? I had no batting average. I had no way to know how effective this tool actually was. In fact, in cases with no physical evidence, I had no ground truth to measure from anyway. And here is the part that should bother you: neither do the people teaching it. They’ll tell you it’s near-perfect. They will never provide you with a number. There’s a reason for that. They don’t have one either. The only number that matters to them is the price of their class. The researchers who have tested these tools—carefully, with ground truth—keep finding the same thing: there’s nothing to NLP, and it isn’t effective. That is an investigative blindspot that creates risk for your investigations and your reputation.

You Can’t Tell a Hall of Famer From a Guy Below the Mendoza Line

In twenty-six-plus years, I have never once—not a single time—heard a law enforcement officer stand up and say, “I used my read on behavior, and I was wrong this many times.” Never. They tell you about the hits. Only the hits. Imagine judging a hitter that way: count the hits and politely ignore every ground out or strikeout. You could not tell a Hall of Famer from a guy who can’t stay above the Mendoza line. That is precisely the position the behavior reader is in—and he doesn’t know it, because the box score he keeps only has one column.

So please understand what you’re actually doing when you add these techniques or so-called "tools" to your proverbial "toolbox." You are building investigative error directly into how you approach your interviews. Not a little, not a lot — an unknown amount. And unknown is the whole problem. You cannot manage a risk you refuse to size.

You can’t learn from a mistake you never measured — and this is built so you never measure it.

Why We Love This Lie Detection Stuff Anyway

Let’s be honest about the appeal; it’s the whole problem. Anxiety-based deception detection is sexy. It feels like a superpower. Spot the shoulder shrug, the broken eye contact, the throat clear before the answer—and suddenly you’re the one in the room who sees what everyone else missed. Nobody signs up for that training to be humbled. They sign up to gain more insights.

And almost everyone who buys in can identify the moment it “worked.” One time, they caught a cue, they knew someone was lying, and they were right. That single memory does enormous work. It becomes proof. It is also textbook confirmation bias: you remember the hit and you never log the miss. Every success gets filed under “I’m good at this” and this "tool" works. Every failure gets filed under nothing at all — because you never noticed it was a failure. The feedback loop that would make you better is the exact thing these flippant approaches destroy.

No Scorekeeping, No Learning

Learning requires feedback. Feedback requires knowing when you were wrong. Knowing when you were wrong requires ground truth and honest scorekeeping. Anxiety-based lie detection offers none of it. The “liar” walks out of the room and you rarely find out what was true. The misses stay hidden, the hits get celebrated, and your confidence climbs while your accuracy sits still.

And the base rate is grim. The largest meta-analysis of human deception detection — Bond and DePaulo, pooling 206 studies and more than 24,000 judges — put average accuracy at about 54%. A coin flip is 50. People were slightly better at calling the truth than at catching the lie. That’s the ceiling most “read the body language” methods are quietly bumping against.

Now widen your view. When our field gets it catastrophically wrong—the false confession, the wrongful conviction, the innocent person who “looked deceptive” and lost years of their life—the pseudoscience crowd goes quiet. Those are the most significant most public misses we have, exactly the cases a real discipline would record and learn from. Instead, they’re waved off as extreme events, one-time occurrences, and that will never happen to me. They are not aberrations. They are what happens when confidence is cut loose from a scorecard.

A discipline that never studies its worst failures isn’t a discipline. It’s a belief system.

Now Point It at a Victim

Here is where it stops being an academic argument. Some national vendors still advertise that you can run these same techniques on victims and witnesses — presumably including people walking straight out of a traumatic event. Why? Because we are obsessed with credibility assessment. We’ve convinced ourselves we’re human polygraphs, and we bring that certainty to bear on the most vulnerable person in the room.

We are not human lie detectors. We never were. But we are confident—and confident-and-wrong, aimed at a trauma victim, is not a rounding error. It’s a second harm, delivered by someone who was told a nervous glance meant something it doesn’t.

They Call It “Science”

The tell is in the vocabulary. These techniques wear the language of science while behaving nothing like it. Real science makes falsifiable claims, tracks its error rate, and updates when the evidence arrives.

Neuro-Linguistic Programming (NLP)

The famous claim: look up and to the right, you’re lying; up and to the left, you’re remembering. It sounds precise. It sounds neurological. It is neither. When Wiseman and colleagues actually tested it — coding the eye movements of people lying and telling the truth, then checking real high-stakes press conferences — there was no relationship between eye direction and deception. Coaching people to watch for the “tell” didn’t help them either. The authors’ recommendation was blunt: abandon it. I learned that lesson the hard way, in a room with the wrong man.

Microexpressions

The idea that a true emotion leaks (the Pinocchio effect) across the face in a fraction of a second, betraying the liar, is cinematic—and mostly just that. In Porter and ten Brinke’s frame-by-frame study, genuine microexpressions appeared in roughly 2% of expressions, showed up in only a minority of people, and were nearly as common in truth-tellers as in liars. A “tell” that fires for honest people too is not a tell. It’s noise you’ve decided to read as signal.

Baselining

Baselining sounds rigorous — establish someone’s “normal,” then watch for deviation. The trouble is what the deviation actually measures. The underlying cues to deception are, per the definitive DePaulo meta-analysis, faint and unreliable — small effects, wildly inconsistent across people and situations. So you build a baseline, you spot a change, and you’ve captured… that the person is nervous, or caffeinated, or humiliated to be there, or simply not you. Deviation from baseline is real. Its link to lying is not. Mine told me exactly where I stood, and I was standing in the wrong place.

Behavioral Interviewing / Behavior-Provoking Questions

Scripted questions engineered to “provoke” a guilty reaction inherit every flaw above and add one more: they push the interviewer toward an accusatory posture that the evidence ties to false confessions. You’re not gathering information. You’re manufacturing anxiety and then interpreting the anxiety you manufactured. That’s a closed loop with no exit and no scorecard.

None of these is science in any sense that matters. They don’t make falsifiable predictions that survive testing, and — back to where we started — they don’t track when they’re wrong. They just sound like science. That’s the marketing, not the method.

Before You Pay for the Lie Detection Training — A Gut Check

This part is for you specifically if you’re thinking about spending money and career hours on anxiety-based lie-detection training or the interview and interrogation training packaged with it. Don’t take my word for anything. Just answer six honest questions first:

•       Do you actually know how to conduct a full cognitive interview?

•       Have you ever learned and can you fully implement the Strategic Use of Evidence (SUE) technique?

•       Do you know how to formulate memory-compatible questions?

•       Can you build an evidence disclosure plan?

•       Do you know the active listening techniques that maximize both the quality and the quantity of information you get?

•       Do you understand rapport and ORBIT, and how they increase information yield?

If you hesitated on any of those — and most people hesitate on most of them — you have not yet exhausted the evidence-based fundamentals. You’re reaching for the magic trick before you’ve learned the craft.

Sit with the second one bullet point. Nearly nobody in the United States has even heard of the Strategic Use of Evidence. That’s not a small gap. SUE has been around for nearly twenty years and is a disciplined way to handle your evidence, questioning strategies, credibility assessment, planning, and the suspect’s own statement—you let a person commit to an account, then measure it against what you already know. It’s how you gather a false exculpatory statement, the kind prosecutors and defense attorneys alike understand the value of (don't believe any instructor who says prosecutors just want confessions). If the best-tested credibility tool in the field is one you were never offered, ask who decided that for you.

And the fundamentals have receipts. When Hartwig and colleagues trained officers in SUE, accuracy jumped from 43% for untrained interviewers to 85% for trained ones. Meissner’s meta-analysis shows information-gathering approaches reduce false confessions without sacrificing true ones. Alison’s work on rapport-based interviewing—built from analysis of over 2,000 hours of real interrogations—shows that rapport increases information yield while accusatory tactics shut it down. This is what evidence looks like when it counts its misses. These techniques produce statement evidence you can actually testify to in court or a deposition.

The Choice You’re Actually Making As An Investigator

Let me put the real choice on the table. If the legacy ways of interviewing are good enough for you, then say the quiet part out loud: “I’m comfortable being below average. Nothing I do can ever be upgraded. I can’t get any better.” Because that’s the deal you’re signing. The batter who never measures can never know whether he’s improving — he’s frozen, by choice. Science-based interviewing is the opposite: it measures so it can improve. That is the entire difference between the two worlds.

If you’re not building toward the best interview you’re capable of, you’re not trying to be the best — you’re defending the right to stay exactly as good as you were the day you stopped counting. The day you left bad "tools" in your toolbox.

The Bottom Line

The pitch for anxiety-based lie detection is that it turns you into a human polygraph; at least you think so. The reality is that it turns you into a confident one—and confidence without a scorecard is the most dangerous thing an investigator can carry into a room. You’ll feel like you’re getting better while you get worse, because you’ve built a system that can only ever show you your wins. I know this because I took it into a room and pointed it at an innocent man.

So keep score. Count the misses. Study the disasters. And spend your training dollars on the methods that can prove they work—because they bothered to measure

Where to Start

If any of this landed, here’s the good news: better interviewing is a skill, not a gift, and it’s teachable. Bring our train-the-trainer program to your agency to build better interviewers with instruction from a researcher and practitioner. Book a course for your officers, host a class, or let's talk about an online/in-person combination course. It’s the evidence-based interviewing that law enforcement, HR, compliance, corporate investigators, or anyone conducting workplace investigations has gone years without being taught. You don’t have to keep guessing at your batting average. You can start keeping score.

Demand better training because mediocrity is a sin.

Lie Detection Resources and References

Alison, L. J., Alison, E., Noone, G., Elntib, S., & Christiansen, P. (2013). Why tough tactics fail and rapport gets results: Observing Rapport-Based Interpersonal Techniques (ORBIT) to generate useful information from terrorists. Psychology, Public Policy, and Law, 19(4), 411–431.

Bhatt, S. (2023). Neurolinguistic programming: A fancy name for a lot of nonsense. Project Aletheia.

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

DePaulo, B. M., Lindsay, J. J., Malone, B. E., Muhlenbruck, L., Charlton, K., & Cooper, H. (2003). Cues to deception. Psychological Bulletin, 129(1), 74–118.

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.

Meissner, C. A., Redlich, A. D., Michael, S. W., Evans, J. R., Camilletti, C. R., Bhatt, S., & Brandon, S. (2014). Accusatorial and information-gathering interrogation methods and their effects on true and false confessions: A meta-analytic review. Journal of Experimental Criminology, 10(4), 459–486.

Porter, S., & ten Brinke, L. (2008). Reading between the lies: Identifying concealed and falsified emotions in universal facial expressions. Psychological Science, 19(5), 508–514.

Wiseman, R., Watt, C., ten Brinke, L., Porter, S., Couper, S.-L., & Rankin, C. (2012). The eyes don’t have it: Lie detection and neuro-linguistic programming. PLoS ONE, 7(7), e40259.