What Actually Gives Lies Away

Part 2 of 8 in a series on lies and deception at work.

Liars are rarely caught by their faces. They’re caught by paperwork, timestamps and details that can be checked, and that’s where to look instead.

In 1995 Jonathan Aitken, then a cabinet minister, called a press conference to announce he was suing the Guardian and Granada Television for libel. He promised to fight with “the simple sword of truth”. The line is still quoted thirty years later.

The case turned on who had paid for his stay at the Ritz in Paris in 1993. The Guardian said a Saudi businessman had. Aitken said his wife had. In June 1997 the Guardian produced records showing she had spent that weekend in Switzerland. The case collapsed, and in 1999 Aitken was jailed for 18 months for perjury and perverting the course of justice.

Nobody caught him by watching his hands or his eyes. He was caught by paperwork.

Aitken lied under oath and was undone by documents. More than a decade ago, I worked in the criminal courts for the Ministry of Justice, and I watched smaller versions of that story play out in live courtrooms. People took the oath, then told stories the CCTV had already disproved.

In the first essay in this series I argued that faces and body language tell us very little about who’s lying. This one is about what does. The short answer is words, a little, and evidence, a lot.

Found out later

In 2002 Hee Sun Park, Tim Levine and colleagues asked students to describe a lie they’d uncovered and how they’d found out. Only 2.4% had caught the lie from the liar’s behaviour alone. Most, 62%, found out from something outside the conversation. Someone else told them, physical evidence turned up, the story clashed with something they already knew, or the liar owned up. Most lies came out hours, weeks or months after they were told.

Some researchers argue this undersells behaviour. An odd manner might make you suspicious, and suspicion sends you looking for proof. That’s plausible. But suspicion isn’t proof, and the first essay showed how often the suspicious-looking person is innocent.

I’ve seen this outside a courtroom too. One place I used to work went through a regular external audit. The IT department’s job was to pull reports from our data for the auditors. Those reports showed lower compliance than the story the organisation told about itself. It was nothing that would close the place down, but it put a grade the organisation wanted at risk.

The reports didn’t go straight to the auditors. A team of senior staff had been assembled as the go-between, collecting whatever the auditors asked for from each department and passing it on. We sent our reports to them. Because we’d pulled the data, we knew what score it should give. When the final result came back, it didn’t match.

It wasn’t anyone’s manner that made us suspicious. It was the score. I’ll come back to what we found.

Words beat faces, but not by much

In Bond and DePaulo’s pooled analysis of lie detection, the one behind the famous 54%, people did better when they could hear a speaker than when they could only see them. Words carry more than faces.

What do they carry? Bella DePaulo’s 2003 review of 158 possible cues found liars tend to be less forthcoming and tell less convincing stories. Their accounts also contain fewer of the ordinary imperfections of real memory: fewer self-corrections, fewer admissions that they’ve forgotten something, fewer odd details that don’t go anywhere. Real memories are untidy. Invented ones are often neater.

That turns a lot of popular advice on its head. Hesitation and “um”s are often treated as warning signs, and a very detailed story as a smokescreen. Hesitation isn’t a reliable sign of anything much. Correcting yourself is, if anything, a small sign of honesty. Detail depends on what kind, as we’ll see.

Computers can count words more reliably than we can. A 2015 meta-analysis of 44 studies using software found liars used slightly more negative emotion words, fewer sensory words, and more language distancing themselves from events. The effects were small and shifted with the situation. There’s a kernel of truth in “watch for distancing language”, but it’s a kernel, not a method. Machines that claim more are the subject of the next essay.

The method that points the wrong way

To see how far a kernel can be stretched, look at SCAN, or Scientific Content Analysis. Avinoam Sapir, a former Israeli polygraph examiner, developed it to find lies in written statements through things like pronoun choice and changes of tense. In 2019 ProPublica reported that his company listed 417 US agencies it had trained, with police in the UK and elsewhere among its other customers.

When researchers asked 117 people to write one true and one invented account of a bad experience, SCAN couldn’t tell them apart. A 2021 analysis of 71 studies reached the same verdict. A US government review went further. Two of SCAN’s warning signs, gaps in memory and spontaneous corrections, are actually signs of truth. The method doesn’t just fail. It points the wrong way.

The same 2021 analysis found two older methods do better than chance. Criteria-Based Content Analysis and Reality Monitoring both score the quality of a statement: its detail, structure and sensory texture. But they need trained assessors, most of the evidence comes from laboratories, and results vary widely between studies. Like any test, where you draw the line decides how many honest people you wrongly accuse. I’ll come back to that at the end of this series.

Two kinds of consistency

It’s widely believed that a true story holds together because it’s based on real memories, while liars trip over their own accounts. Researchers who’ve tested this found something less comfortable.

Pär Anders Granhag and Leif Strömwall interviewed liars and truth-tellers more than once. Liars were just as consistent as truth-tellers, and sometimes more so. Their explanation is that liars repeat and truth-tellers reconstruct. A liar has rehearsed a story and sticks to it. An honest person goes back to the memory each time and pulls out something slightly different, because that’s how memory works. A 2019 study from the University of Portsmouth found truth-tellers kept recalling new details between interviews, which made the liars’ accounts look slightly more consistent.

So “your story’s changed” is weak evidence. Adding or dropping details between tellings is normal. Flatly contradicting yourself on something central is different, and worth a question.

I’ve been on the wrong end of my own memory. I was once hunting for a photograph in my Lightroom library. I was fairly sure when I’d taken it, give or take, so I thought I knew where to look. It wasn’t there. I was sure where I’d taken it too: Soho. So I checked Lightroom’s map, which shows where your pictures were taken if your camera records GPS. Still nothing.

Days, perhaps weeks, later I stumbled on some pictures I recognised from the same day. I searched that folder and there it was. I’d been out by about a year, and the picture I’d have sworn I took in Soho had been taken in Shoreditch.

If anyone had asked me about that picture, I’d have given them a confident, coherent and completely wrong account. I wouldn’t have been lying. I’d just have been wrong.

Deception researchers separate two kinds of consistency. Coherence asks whether a story hangs together. Correspondence asks whether it matches the world. J. P. Blair, Tim Levine and colleagues found correspondence is the better guide. When Blair’s team gave people relevant background knowledge to judge statements against, accuracy across ten experiments rose from the familiar 54% to around 75%.

A story that matches the facts beats a story that matches itself.

I saw this in court more than once. A defendant’s account could hang together perfectly well and still collide with the CCTV, or with several independent eyewitnesses. Some kept insisting on their innocence anyway. Their stories were coherent. They just didn’t match the world.

Details you can check

Galit Nahari, Aldert Vrij and Ronald Fisher noticed that liars face a dilemma. A thin story sounds suspicious, so they need detail. But detail that can be checked can sink them. The way out is detail nobody can verify.

Compare two accounts of an afternoon. “I was at my desk working on the report.” Or: “I had a call with the supplier at two, then sent the draft to finance just before four.” The second can be checked in ten minutes.

In their 2014 study, liars gave as many uncheckable details as truth-tellers but fewer checkable ones. A 2021 meta-analysis of 20 studies found the same pattern. The gap grows when people are told at the start that their details may be checked. Truth-tellers respond by offering more. Liars don’t.

Two cautions. An independent replication in 2021 found the effect was less than half the size of the original, and that simply counting all details worked better. Almost none of this has been tested outside the lab. Still, the idea is cheap and sound: ask for things you can check, then check them.

Photographers have a head start here, because every digital photograph carries checkable details. The camera writes the date and time into the file, and often the location too.

I once did a job for a company in London. I delivered the pictures, got paid, and everyone was happy. Later I saw them using one of my photographs that hadn’t been part of the job. When I raised it, they stood their ground. As far as they were concerned, it had come as part of the package.

In the end I emailed my contact there the raw file and asked him to look at the metadata. It showed the picture had been taken long before I’d ever worked for them. It turned out a junior member of staff had taken it from my website. It was an innocent mistake, and it was resolved.

Nobody had lied. The file settled it without anyone having to call anyone a liar.

Keep your cards until the end

The most promising method comes from Maria Hartwig, Pär Anders Granhag and colleagues, who developed it at the University of Gothenburg. It’s called the Strategic Use of Evidence, and the idea is almost embarrassingly simple. Don’t show your hand first.

Guilty and innocent people handle questioning differently. The guilty tend to avoid or deny anything that might incriminate them. The innocent tend to tell it straight and trust the truth to come out. Open by revealing what you know, and a liar simply builds a story around it. Ask for their account first, then specifics, and only then reveal the evidence, and liars tend to contradict what you already hold.

In a 2006 study, 82 police trainees interviewed mock suspects. Some had committed a staged theft. Others had done something innocent that left similar traces. Trainees taught to hold the evidence back judged correctly 85% of the time. Those without the training managed 56%.

In the first essay, training in reading body language made students less accurate and more confident. Here, training made people much better. The difference is where the training points you: not at the person’s face, but at the gap between what they say and what can be shown.

Courts build a version of this into their procedure without calling it anything. Witnesses wait outside until they’re called, so they can’t shape their evidence around what they’ve heard. I watched a side effect of that several times. A defence witness would take the stand and stick loyally to the defendant’s version, unaware that the prosecution had already shown the court CCTV that contradicted it. It was a witness defending a story they didn’t know had already collapsed.

Not all of them were necessarily lying. Some may have heard the defendant’s version so often that it had become their memory too. Either way, it wasn’t their manner that showed the story was wrong. It was the footage.

Honest caveats. A 2016 study with 59 US law-enforcement officers found a smaller gain, 65% against 43%. Suspects who knew about the technique adapted by volunteering more. Innocent people misremember too, so a clash with the evidence is a reason to ask another question, not a verdict. And the method assumes innocent people speak freely. Someone who has learnt that the truth doesn’t always protect them may not.

Email is where lies get caught

It’s easy to assume email is ideal for lying: no face to give you away, and time to polish the story. The evidence suggests roughly the opposite.

In 2004 Jeff Hancock and colleagues at Cornell had 30 students log every conversation and every lie for a week. Lies turned up in 37% of phone calls, 27% of face-to-face conversations, 21% of instant messages and 14% of emails. Hancock put it down to recordability. Email sits on a server and can be forwarded to the person you’re lying about. He also pointed out that you can phone in sick dressed for skiing, which you can’t do in person.

That isn’t the whole picture. In a 2010 study, business students dividing a pot of money lied about its size more often by email than on paper, 92% against 64%, and felt more justified doing it. Compared with a handwritten note, email doesn’t feel like signing your name.

So email is neither a liar’s paradise nor a truth serum. It’s a record. Popular tips for reading email, like watching reply times and shifts in tone, are Brokaw hazards waiting to happen: the trap, described in the first essay, of mistaking someone’s normal habits for signs of lying. Some people reply slowly because they’re in meetings. Some are terse because it’s Friday. The value of email isn’t in reading between the lines. It’s in the lines themselves, dated and timestamped, ready to be checked against everything else.

Back to that audit. The mismatch made us suspicious, so we went looking for the record. We compared the reports IT had sent to the senior team with what that team had then sent to the auditors. We also checked the Microsoft 365 audit logs to see what had happened in between. Lo and behold, some of the data had been changed. The logs showed who had changed it, and exactly when.

Nobody had to read a face or decode the tone of an email. The two versions simply didn’t match.

We reported it to the IT director, who was on the board. Not long afterwards, I left. It didn’t sit right with me that information from my department had been manipulated and misrepresented.

The guilty knowledge test, done properly

The Guilty Knowledge Test, now usually called the Concealed Information Test, deserves a proper explanation, because it’s one of the cleverest ideas in the field.

David Lykken published it in 1959. Asking “did you do it?” makes guilty and innocent people equally nervous, so Lykken tested recognition instead. A suspect hears multiple-choice questions, each containing one real detail of the crime among plausible alternatives. Was the weapon a knife, a rope, a pistol or a bat? An innocent person has no idea which option matters. Someone who was there does, and their body responds, measured through sweat, heart rate or breathing.

Researchers regard it as far better grounded than the standard polygraph. It’s rarely used in the West, but Japanese police use it routinely, with about 100 examiners handling around 5,000 cases a year.

It has a weakness that matters at work. It only works if the key details haven’t leaked to innocent people, and knowing something isn’t the same as having done it. You might simply have heard. In most offices, anything interesting has leaked by lunchtime. Gossip gets an essay of its own later in this series.

It’s tempting to turn the idea into office conversation tactics, such as deliberately getting a detail wrong to see who corrects you. Without the alternatives, controls and measurement that make the real test work, that isn’t a test. It’s a trick, and judging the reaction brings us straight back to reading faces. The machinery behind the real test, polygraphs included, is covered in the next essay.

What this means on a Monday morning

None of this adds up to a way of catching colleagues out, and I’d be wary of anyone who says it does. It adds up to some habits. Ask for someone’s account before you tell them what you know. Ask for details you can check, and then check them. Don’t treat a changed story, a slow reply or an “um” as proof of anything. When an account and the evidence clash, ask another question before reaching a conclusion. Remember that most people, most of the time, are telling the truth.

Whether you should be doing any of this to colleagues at all is a separate question, and it’s where this series will end.

Jonathan Aitken had the simple sword of truth. The Guardian had the paperwork.

Sources and further reading

Armstrong, K. & Sheckler, C. (2019). Why are cops around the world using this outlandish mind-reading tool? ProPublica.

Blair, J. P., Levine, T. R. & Shaw, A. S. (2010). Content in context improves deception detection accuracy. Human Communication Research, 36, 423–442.

Blair, J. P., Reimer, T. O. & Levine, T. R. (2018). The role of consistency in detecting deception: the superiority of correspondence over coherence. Communication Studies, 69, 483–498.

Bogaard, G., Meijer, E. H., Vrij, A. & Merckelbach, H. (2016). Scientific Content Analysis (SCAN) cannot distinguish between truthful and fabricated accounts of a negative event. Frontiers in Psychology, 7, 243.

Brennen, T. & Magnussen, S. (2022). The science of lie detection by verbal cues: what are the prospects for its practical applicability? Frontiers in Psychology, 13, 835285.

DePaulo, B. M. et al. (2003). Cues to deception. Psychological Bulletin, 129, 74–118.

Granhag, P. A. & Strömwall, L. A. (1999). Repeated interrogations: stretching the deception detection paradigm. Expert Evidence, 7, 163–174.

Hancock, J. T., Thom-Santelli, J. & Ritchie, T. (2004). Deception and design: the impact of communication technology on lying behavior. Proceedings of CHI 2004.

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, 603–619.

Hauch, V., Blandón-Gitlin, I., Masip, J. & Sporer, S. L. (2015). Are computers effective lie detectors? Personality and Social Psychology Review, 19, 307–342.

Hudson, C. A. et al. (2019). The devil is in the detail: deception and consistency over repeated interviews. Psychology, Crime & Law.

Luke, T. J. et al. (2016). Training in the strategic use of evidence technique. Journal of Police and Criminal Psychology, 31, 270–278.

Lykken, D. T. (1959). The GSR in the detection of guilt. Journal of Applied Psychology, 43, 385–388.

Nahari, G., Vrij, A. & Fisher, R. P. (2014). Exploiting liars’ verbal strategies by examining the verifiability of details. Legal and Criminological Psychology, 19, 227–239.

Naquin, C. E., Kurtzberg, T. R. & Belkin, L. Y. (2010). The finer points of lying online. Journal of Applied Psychology, 95, 387–394.

Oberlader, V. A. et al. (2021). Validity of content-based techniques for credibility assessment. Applied Cognitive Psychology, 35, 393–410.

Osugi, A. (2011). Daily application of the Concealed Information Test: Japan. In B. Verschuere, G. Ben-Shakhar & E. Meijer (eds), Memory Detection: Theory and Application of the Concealed Information Test, 253–275. Cambridge: Cambridge University Press.

Palena, N., Caso, L., Vrij, A. & Nahari, G. (2021). The Verifiability Approach: a meta-analysis. Journal of Applied Research in Memory and Cognition, 10, 155–166.

Park, H. S., Levine, T. R. et al. (2002). How people really detect lies. Communication Monographs, 69, 144–157.

Verschuere, B. et al. (2021). The verifiability approach to deception detection: a preregistered direct replication. Applied Cognitive Psychology, 35, 308–316.