Abstract
The question this article answers is: why is the most convincing person in the room so often the wrong one, and why does telling them so change nothing?
Grandiose narcissism is the most heavily researched and most frequently mislabelled trait in this literature. This article appraises what it predicts in everyday behaviour, at work and in private life, rather than restating its popular coverage.
It adopts the two-process account that has organised the field since 2013, distinguishing narcissistic admiration from narcissistic rivalry (Back et al., 2013), and separates the trait from self-esteem (Brummelman, Thomaes & Sedikides, 2016), extraversion, vulnerable narcissism, and clinical diagnosis.
Its central contribution is to apply two recent findings to the practical question of why the pattern resists correction. Direct measurement of brain activity links grandiose narcissism to a weaker early response to having made a mistake, supporting one of two competing theories of the trait and contradicting the other (Robins, Zhou, Huang, Angus, Sedikides & Kelley, 2026). Separately, a nine-study programme including two national surveys, a fourteen-day daily-diary study and six experiments finds the trait strongly associated with being left out, driven by heightened sensitivity to unclear social signals, genuine avoidance by others, and a loop in which exclusion appears to feed the trait (Büttner, Rudert, Albath, Sibley & Greifeneder, 2025).
The trait’s timetable is traced through three independent designs that converge: a laboratory study coding behaviour across weekly sessions (Leckelt, Küfner, Nestler & Back, 2015), a twelve-week field study of peer-rated leadership (Ong, Roberts, Arthur, Woodman & Akehurst, 2016), and a synthesis of prior research in which the link to being picked as a leader halved once raters had known the person more than a week (Grijalva, Harms, Newman, Gaddis & Fraley, 2015). Consequences are examined experimentally through the suppression of information-sharing in groups (Nevicka, Ten Velden, De Hoogh & Van Vianen, 2011), and over time in the personal domain, where a six-year study of more than five thousand couples found no faster decline in relationship satisfaction where one partner scored highly (Seidman et al., 2026).
A dedicated chapter sets out the four kinds of person most often wrongly labelled with this trait, since the commonest practical error in this territory is not failing to spot it but applying it to somebody it does not fit.
The practical section offers five moves and two refusals. The conclusion is that this trait’s far better evidenced at the level of mechanism than at the level of what to do about it, and that the honest version of the advice is smaller, duller and considerably more useful than the version usually sold.
Keywords: grandiose narcissism; narcissistic admiration and rivalry; error monitoring; social exclusion; leadership; workplace behaviour; personality change; organisational psychology.
PART I. WHAT ACTUALLY HAPPENS
Chapter 1: The Three-Month Turn
1.1 Three Rooms
The scenarios below are composite. They draw on recurring patterns in research and advisory practice, and they describe no real, identifiable person or organisation.
A new director joins a professional services firm. For the first month she is the most impressive person in every meeting she attends: quick, unhesitating, generous with credit in a way people notice and repeat. By month four, two of her team have quietly asked to move. Nobody can name an incident. The most anyone will say, and only in private, is that things have started to be about her, and that disagreeing with her now costs more than it used to.
A man has been with his partner for three years. Friends who met him early describe someone who lit up the room. Friends who have known him for three years describe someone they see less of now, because the arrangements have got complicated and the evenings have a way of turning. He would tell you, apparently sincerely, that his friends have drifted.
A woman has one friend in her circle who is the funniest person there, and the only one nobody wants to sit next to all evening. She has noticed that this friend hears a slow reply as a snub, and hears an actual snub, when one lands, with something close to indifference.
1.2 What the Three Have in Common
So, we have three rooms, three quite different kinds of relationship, and I would suggest the same shape running through all of them. Somebody arrives well and wears badly, over weeks to months, in a way nobody involved can deny, and everybody involved notices eventually.
I have started here rather than in an office deliberately, because this is not an occupational problem. What produces the pattern needs only a group of people who do not yet know each other well, some form of standing worth having, and enough repeated contact for the second half of the trait to find something to push against. A new team qualifies. So does a new relationship, a club committee, and a family gaining an in-law.
All three share a fourth thing, and it’s why I wrote this article. In every case, somebody has already tried telling them, and it hasn't worked.
1.3 Why Reaching for the Label Makes It Worse
The obvious move is to say narcissist, and I understand the appeal. The word is available, it’s satisfying, and it explains everything, which is usually the first sign a word is doing no work at all.
Two costs follow from reaching for it, and I would say the second is the one likely to affect you personally.
The first is borne by other people. The word has become the general-purpose term for someone difficult, and the price of that inflation is paid by everybody wrongly labelled, most of whom are simply assertive, or self-contained, or tired of being agreeable. Chapter 8 is about those people, and I put it late on purpose, because I find a caution lands better once the reader has some reason to care about what it warns against.
The second cost is yours. A label is a conclusion about a person, and a conclusion invites a counter-conclusion. Say you are being a narcissist, and you have started an argument about your judgement rather than a conversation about their behaviour. In my experience, that’s close to a restatement of the trait rather than an unlucky outcome, because the exchange you have just opened is the one contest they are best equipped to win.
What the research offers instead is not a better label but three things I think are considerably more useful to you: a timetable, telling you the pattern has a predictable shape over weeks and months; a mechanism, telling you what is actually running; and a perceptual profile, telling you that this person is receiving a rather different signal from the one you believe you are sending.
None of that requires you to diagnose anyone, and none of it could. It requires you to watch a pattern over time, which is something you are already entirely qualified to do.
Chapter 2: The Alarm That Does Not Ring
2.1 What Happens When You Get Something Wrong
Something happens in your brain when you make a mistake, and it happens much faster than you do.
About a tenth of a second after an error, and before you have consciously noticed anything, there is a small electrical response that researchers can pick up with sensors placed on the scalp. It has a name: error-related negativity, the technical label for the brain’s mistake alarm. It’s not a thought about having got something wrong. It’s the thing that arrives before the thought, and it’s one of the most studied signals in the whole of brain research.
2.2 Two Theories That Cannot Both Be Right
Narcissism research has offered two competing explanations of what sits under the confidence, and until recently there was no clean way to tell them apart.
The first says grandiosity is a shell over something fragile. If that’s right, a person high in the trait should be hypervigilant to any hint that they have slipped up, and their mistake alarm should ring louder than everybody else’s.
The second says the confidence is maintained by not registering the disconfirming information in the first place. If that’s right, the alarm should ring more quietly.
Two opposite predictions, drawn from the same observable behaviour, which is exactly the situation in which measuring something directly stops being an academic exercise and starts being worth the trouble.
2.3 What They Found
Esther Robins and her colleagues at the University of Southampton set out to test the two against each other.
In the first study, 144 people did a task designed to trip them up, while scalp sensors recorded what their brains did each time they slipped. In the second, 50 people did a version that told them explicitly, after every attempt, whether they got it right, removing any question of whether they knew.
In both studies, higher narcissism went with a quieter mistake alarm. It held when the researchers accounted for how many errors people actually made. It showed up on both halves of the trait, and slightly more strongly on the confident, self-promoting half than the hostile one. The authors conclude that this supports the second theory and describe it as a possible reason narcissists resist correcting themselves (Robins, Zhou, Huang, Angus, Sedikides & Kelley, 2026).
2.4 What This Does Not Show
I want to be careful here, because this is the most striking thing in the article and the easiest to overstate.
It does not show that anybody is incapable of learning. It shows a difference in very fast brain responses during a laboratory task between two groups of mostly white British students, and the authors say so themselves.
It’s also disputed. Markus Mück and colleagues examined the same question earlier and treated it as a genuine open contradiction, with vigilance and avoidance both plausible depending on which half of the trait you look at and how far into the process you measure (Mück, Mattes, Porth & Stahl, 2023). Robins and colleagues engage with that work directly and propose a way of reconciling the two: a quiet alarm most of the time, with the possibility that the hostile half of the trait shows a louder one when confronted with criticism it cannot avoid.
Evidence status: contested. Two research groups, compatible methods, results that do not straightforwardly agree, and a proposed reconciliation nobody has tested. That disagreement is the content here rather than a caveat at the end of it.
2.5 Why Telling Them Does Not Work
Even held at that strength, this changes the practical picture in a way nothing else in the literature does.
Everybody who has managed, married or befriended one of these people reports the same thing: telling them does not work. The standard explanations are that they know and do not care, or that they know and cannot admit it. Both are explanations about character, and both lead to the same response, which is to say it again, more firmly, more often, or in writing.
What this evidence proposes instead is that the signal is quieter at source. Not suppressed after it arrives but turned down before it becomes available to suppress. If that’s right, then clarity was never the missing ingredient, because the problem was never that your message was unclear.
A second piece of evidence fits neatly alongside it, and it comes from an entirely different method. Erika Carlson, Simine Vazire and Thomas Oltmanns examined what narcissists believe about their own reputation across three studies and reached three conclusions they themselves described as surprising. They understand that other people see them less favourably than they see themselves. They have some insight that they make good first impressions, which then deteriorate. And they describe their own narcissistic qualities recognisably when asked (Carlson, Vazire & Oltmanns, 2011).
Put those two together, and you get something more precise than either alone. They know how they look in general. They don't seem to feel the particular moment when they got it wrong.
Which is exactly the wrong way round for feedback to work, because feedback runs on particular moments and relies on the person having felt something when the moment happened.
I've seen this often enough in boardrooms to describe it, though I would ask you to take it as observation rather than evidence. A senior person is told, in terms everybody else in the room understands perfectly, that a decision last quarter went badly. They leave having absorbed a general fact about their standing, which is that some people are unhappy with them, without having absorbed the particular fact about the particular decision. Six weeks later they make a comparable decision in a comparable way, and the people who delivered the message conclude, reasonably enough, that they were ignored. They were not ignored so much as heard at the wrong level: the reputational signal landed, and the one underneath it did not.
Chapter 3: They Think You Are Leaving Them Out
3.1 The Most Excluded Person in the Room
Here is the finding I would most want you to take away, and it inverts the picture almost everyone carries.
Christiane Büttner and colleagues examined the relationship between grandiose narcissism and being left out across nine studies with a combined 77,289 people (Büttner, Rudert, Albath, Sibley & Greifeneder, 2025). The programme included two large national surveys, a fourteen-day daily-diary study in which people logged each moment of feeling excluded on a phone app as it happened, and six experiments.
The headline is strong and consistent. People higher in this trait reported being left out far more often than everyone else, both looking back and in real time.
That alone turns the folk picture upside down. The narcissist of popular imagination is the centre of attention. The narcissist of this data is the person quietly left off the invitation.
3.2 Three Reasons, and the Third Is the One That Matters
They read unclear situations as rejection. The experiments showed heightened sensitivity to ambiguous signals but not to obvious ones. That distinction is doing a great deal of work, and most readers skim past it. When exclusion is unmistakable, they read it as anybody would. When a situation is genuinely open to interpretation- the unanswered message, the slightly short reply, the conversation that stops as they come in- they land on rejection more often than the rest of us.
People really do avoid them. Some of the exclusion is not misread at all. The researchers found a genuine preference among others for keeping their distance.
And it runs both ways. The relationship appears reciprocal: the trait contributes to exclusion, and exclusion, in turn, appears to feed the trait.
Notice what that third point does to the story we began with in Chapter 1. It stops being a story about a difficult person and becomes a story about a loop, in which two parties, both behaving comprehensibly, make things worse for each other by degrees.
3.3 The Loop You Are Probably Feeding
Consider how this plays out at work, where I suspect you can supply the example before I finish the paragraph.
A colleague becomes difficult. The team does not confront them, because confrontation is exhausting and rarely goes well. The team withdraws: fewer informal conversations, decisions taken before the meeting rather than in it, an email chain that quietly loses a name.
Every one of those moves is unclear by design. That’s precisely what makes them survivable for the people making them, because nobody has to say anything out loud or defend a position.
And unclear is exactly the signal this person over-reads.
So we arrive at the awkward conclusion that the team’s most reasonable, least aggressive way of coping is also the input most likely to trigger the hostile half of the trait. The withdrawal reads as rejection, the rejection registers as a threat, the threat produces the defensive behaviour, and the defensive behaviour justifies further withdrawal.
I should be clear that the last two paragraphs are my reading rather than the study’s finding. Büttner and colleagues did not study workplace withdrawal, and the step from their result to a team’s coping behaviour is mine. I think it’s well supported, since every link in the chain is separately evidenced, but hold it as a hypothesis rather than a result.
Evidence status. The underlying findings: evidenced. The workplace loop described above: practitioner judgement, reasoned from those findings.
3.4 This Is Not an Excuse
It’s possible to read this chapter as an invitation to sympathy, and possible to read it as an alibi. I intend it as neither.
The exclusion is real, the misreading is real, and none of it makes the behaviour that follows acceptable or the harm it causes any smaller. What it changes is which responses have any chance of working, and that’s the subject of Chapter 9.
PART II. WHY IT RUNS THIS WAY
Chapter 4: Two Engines and a Clock
4.1 Two Things Running at Once
Chapters 2 and 3 described what you see happening. This one describes what produces it, and I would say the whole explanation rests on a single idea: grandiose narcissism does not behave as one thing at all. It behaves as two, running in parallel, aiming at the same goal by opposite means.
Mitja Back and colleagues set this out in 2013 and built a questionnaire around it (Back et al., 2013). Both halves exist to keep a grand self-image intact.
Admiration does that by getting people to think well of you. Self-promotion, charm, confident claims about your own capability, striving to be seen as exceptional.
Rivalry does it by pulling other people down. Devaluing them, hostility when contradicted, treating a colleague’s success as a subtraction from your own standing.
This split matters because the two halves predict opposite outcomes, which helps explain why decades of earlier research produced contradictory results nobody could reconcile. Admiration predicts being liked, being seen as capable, and getting picked. Rivalry predicts being disliked, being seen as untrustworthy, and the eventual collapse of exactly the standing admiration built. A single narcissism score averages the two and hands you the mean of a positive and a negative, which is close to no information at all.
Neither half is a performance, and I think this is the part most people get wrong about the whole subject. Both run automatically, and both feel from the inside like an accurate response to the situation. The director in Chapter 1 wasn't laying groundwork in month one and wasn't executing a plan in month four. Two systems were running in her: one that starts immediately, and one that fires only when something arrives that looks like a threat. In month one of a new job, supplies almost no threats.
4.2 Watching the Crossover Happen
Marius Leckelt and colleagues put this under observation. In a laboratory study, 311 people met in small groups in weekly sessions across three weeks. They filmed every session, and trained observers coded what people actually did.
Two behavioural paths explained the trajectory. The positive one, driven by admiration, ran through confident, expressive behaviour, produced an impression of capability, and mattered less as the weeks passed. The negative one, driven by rivalry, ran through arrogant and aggressive behaviour, produced an impression of untrustworthiness, and mattered more (Leckelt, Küfner, Nestler & Back, 2015).
Three weeks. That’s how long it took for the crossover to become visible to trained observers, in a group with nothing at stake, no hierarchy and no particular reason for anybody to hide anything.
4.3 Twelve Weeks, Twice
The nearest thing to a direct test that I know of came from Chin Wei Ong and colleagues, who tracked how groups rated one another’s leadership across twelve weeks, and did it twice (Ong, Roberts, Arthur, Woodman & Akehurst, 2016).
In the first study, with 112 people who had never met, narcissism went with higher peer ratings of leadership while the group was forming, and not later.
In the second study, with 152 people who already knew each other, narcissism went with nothing at the start and with lower ratings by the end.
They called it a honeymoon period whose appeal then wanes.
The second study is the one I worry about, because it describes what happens after an internal promotion. Among people who already knew the person, the trait predicted nothing at the outset and predicted the opposite by the end.
A third study supplies the mechanism. Tobias Härtel and colleagues used ratings that everyone in a group gave everyone else and found that, of all the impressions people form, being seen as assertive was the strongest predictor of who got treated as a leader. The route from the admiration half of narcissism to that outcome ran through confident, expressive behaviour producing exactly that impression. The rivalry half predicted it not at all (Härtel, Leckelt, Grosz, Küfner, Geukes & Back, 2023).
4.4 The One That Gets Them Picked Is Not the One That Causes Trouble
That last sentence is the cleanest statement of the problem I can find, and I would ask you to hold on to it.
Which gives us the shape of the whole thing. There is a window, somewhere between a first meeting and a few months of real contact, in which this trait produces a genuinely favourable impression and almost none of its costs. The impression is not manufactured, and the costs are not hidden. They simply haven't happened yet, because rivalry needs something to push against, and a first meeting supplies nothing.
Two consequences follow, running in different directions.
At work, any process that meets somebody inside that window and then commits, which describes most selection and most first-impression judgement, is sampling at the point of maximum favourable signal and minimum information. Chapter 5 develops what those costs.
Everywhere else, the consequence is more forgiving and probably more useful to you. If you have known somebody for three months and something has shifted that you cannot name, this is a recognised trajectory rather than evidence that you have become unreasonable. That reframe is worth more to you than a diagnosis, and unlike a diagnosis it’s one you are entitled to make.
Chapter 5: Good at Getting the Job
5.1 Chosen, Then Not Better
A large synthesis of the previous research separated two questions the field had been running together for twenty years. Does narcissism predict who gets picked as a leader? And does it predict who turns out to be any good at it? (Grijalva, Harms, Newman, Gaddis & Fraley, 2015.)
It predicts the first. It does not predict the second.
The link to getting picked is real but modest, and anybody reporting it to you as narcissists always rise is overselling what the number carries. The same work found that link roughly halved once raters had known the person for more than a week, which is the same clock as Chapter 4 and arrived at by a completely different route.
On being any good at the job, the relationship was essentially nothing. Not negative, as the corporate-psychopath genre predicts. Not positive, as the charismatic-leadership genre predicts. Nothing.
I think that reads as reassuring and is nothing of the kind. A trait that predicts who gets chosen and predicts nothing about performance adds noise to appointments without any compensating benefit. If it predicted performance negatively, organisations would at least be wrong in a direction they could correct.
5.2 The Number Nobody Quotes
The same work broke the performance ratings down by who produced them, and this is the part that almost never gets mentioned.
Ratings from supervisors, subordinates and peers showed no relationship with narcissism at all. Ratings the person gave themselves showed a clear positive one.
I would ask you to read that pair of findings slowly, because between them they are doing much more work than they look like they are doing. On how well this person is performing, the person best placed to know and everybody around them systematically disagree, and the disagreement runs in a predictable direction. Chapter 2 supplies a plausible reason why.
One further finding is worth having, although I would hold it a good deal more loosely than the others. The flat overall result on performance conceals a curve: performance ratings rise with narcissism to a point and fall beyond it, so a moderate level does better than either extreme. That rests on six samples using six different ways of measuring performance, and effects shaped like that are easier to find by accident than straight lines are. Even so, ignoring it inverts the practical advice, because the corrective for an appointment made on excessive confidence is not the least confident candidate available. Somebody who will not advance a position, claim a capability, or absorb the exposure of being publicly wrong imposes real costs on themselves.
5.3 The Meeting That Goes Quiet
Now for the part that matters most and that almost every treatment of this subject leaves out, which is what happens to everybody else in the room.
Barbora Nevicka and colleagues ran an experiment with a clever design: they gave each group member a different piece of the information needed to reach the right answer, so no individual had enough on their own. Getting it right required the group to pool what it knew.
They found a gap between how narcissistic leaders were seen and what actually happened. Such leaders were rated as effective, and the rating ran through their displays of authority. Yet their narcissism suppressed information-sharing between group members, and the group’s performance suffered accordingly (Nevicka, Ten Velden, De Hoogh & Van Vianen, 2011).
I would dwell on that for a moment, because it explains why the cost of these appointments is so very nearly invisible from above. The failure is not a bad decision you can trace back to the leader. It’s a decision taken without information that existed in the room and never surfaced. Nobody can point to the moment it went wrong, because the moment consists of things not being said. And from the leader’s own vantage point the meeting went well: they were decisive, they were deferred to, and nobody objected.
5.4 What It Costs Everyone Else
A body of survey evidence surrounds that experiment. Narcissism is reliably associated with the behaviours organisations least want, meaning the sabotage, withholding and rule-bending usually grouped under the heading counterproductive work behaviour, and the link weakens in cultures that place more weight on the group than the individual (Grijalva & Newman, 2015). More broadly across the darker traits, the same association appears alongside lower task performance and less discretionary, above-and-beyond conduct that keeps teams working without it being anybody’s job (O’Boyle, Forsyth, Banks & McDaniel, 2012).
And underneath it all runs Chapter 4’s clock. Whatever an organisation buys at the point of appointment, the research suggests fairly consistently that it does not keep it.
5.5 Why Your Boss Wants You Back in the Office
Here is the newest finding in this article, published in June 2026, and I suspect it is the one most likely to be recognised by anybody who has sat through a return-to-office announcement.
Marissa Shandell, Courtney Elliott and Adam Grant at Wharton spent six years asking why some leaders resist remote and hybrid working so fiercely. The obvious explanations are the ones organisations give productivity, collaboration, culture. The researchers tested whether personality explained more.
They ran three studies. The first analysed 259 chief executives of Fortune 500 companies during the pandemic. The second surveyed 359 managers and supervisors, measuring narcissism alongside Machiavellianism, psychopathy, motivation for power and status, and trust in employees. The third was an experiment with 546 leaders, half of them primed beforehand by reading about confident, self-centred visionary chief executives.
Across all three, narcissism was the only trait that consistently predicted objecting to remote work (Shandell, Elliott & Grant, 2026). Not distrust of employees. Not a belief about productivity. Not the other dark traits, which were measured in the same survey and did not do it.
The mechanism they identify is power and status rather than performance. A leader high in this trait needs an audience, and remote work removes the room. The authors describe narcissistic leaders as preferring face-to-face settings because a richer channel lets them do two things at once, gathering attention and exercising authority through gesture, eye contact, tone and posture, none of which survives a grid of small rectangles.
Two things are worth saying about how you use this. It does not mean that every leader who wants people back in the office is a narcissist, and I would be unimpressed by anybody who read it that way; there are good reasons for wanting a team in a room, and the study did not test whether the mandates were correct. What it does say is that when the stated reason is productivity and the evidence on productivity is being waved away, the more parsimonious explanation may be about the audience rather than the output.
Evidence status: evidenced. Three studies, one of them experimental, more than a thousand leaders, published in a peer-reviewed journal.
Chapter 6: They Can Read You Perfectly
6.1 The Claim That Is Wrong in a Useful Way
Ask anybody what a narcissist lacks and you will get the same answer, which is empathy. It is the single most repeated claim about the trait, and it is wrong in a way that matters, because it points you at the wrong expectation.
Psychologists split empathy into two things that turn out to travel separately. Cognitive empathy is knowing what you are feeling, reading the room, tracking that a remark landed badly. Affective empathy is being moved by it, feeling something yourself because of what you picked up.
The research does not find both switched off. Reviewing twenty-three studies of empathy across the darker traits, Mirko Duradoni and colleagues found narcissism sitting apart from the others: mostly small positive correlations with cognitive empathy, against consistently negative ones for Machiavellianism and psychopathy (Duradoni, Gursesli, Fiorenza, Donati & Guazzini, 2023). A 2025 review pooling the results of every study it could find on the question reached the same place, describing the reading ability in narcissism as intact or even sharpened while the emotional response to what is read stays limited (Chan et al., 2025).
So the picture we end up with is not somebody who cannot tell what you are feeling, but somebody who can tell precisely, and who is simply not moved by what they have read.
6.2 Why This Explains More Than the Popular Version
I would hold that against the popular claim for a moment, because the popular claim cannot actually explain the thing people complain about most.
Manipulation requires accurate reading. You cannot press somebody’s weak point without knowing where it is, cannot time an appeal for when they are tired, cannot pick the one person in a meeting whose support will settle it. Every one of those needs cognitive empathy in working order. A person who genuinely could not read others would be socially clumsy rather than dangerous.
Which is why the difficult colleague seems to know exactly where to push. They do know, because the reading is working perfectly well; what is missing is the part that would ordinarily stop them using what they have read.
6.3 The Second Signal That Arrives and Does Not Land
I would put this beside Chapter 2, because between them the two produce a shape that neither gives you alone.
In Chapter 2 the mistake registers faintly. Here the other person’s distress registers clearly and moves nothing. Two different signals, two different mechanisms, and the same practical consequence: the information arrives and produces no correction.
That is close to the whole article in a line, and I think it explains why the two commonest responses to this person both fail. Explaining how their behaviour makes people feel assumes they had not noticed. On this evidence they noticed.
Evidence status: contested, and briefly. Findings on narcissism and empathy are genuinely mixed, with some studies finding cognitive empathy higher and others lower, and part of the inconsistency comes from grandiose and vulnerable narcissism being averaged together. The direction most reviews settle on is the one described above. Whichever way it resolves, what you should expect from the person does not change, so I am not going to spend your time on the dispute.
PART III. THE SAME PERSON AT HOME
Chapter 7: Same Pattern, No HR
7.1 The Clock Runs Everywhere
Nothing in this research makes narcissism an occupational trait. The mechanism in Chapter 4 needs only unfamiliar people, some standing worth having, and enough repeated contact for the second engine to find something to push against.
The evidence outside work has improved markedly in the last two years, and in a direction that will surprise anybody carrying the popular version of this subject.
I would start with what transfers most cleanly. Stephanie Wurst and colleagues examined the two halves of the trait inside romantic relationships and found them pulling in opposite directions at different stages: admiration associated with good things early on, rivalry with bad things later (Wurst et al., 2017). That’s Chapter 4’s clock again, running in a relationship rather than a team, and I think it explains a pattern people describe to me constantly without ever having had a framework for it.
7.2 The Six-Year Finding
Then comes the surprise, and it’s a considerable one.
A study tracking more than five thousand couples for up to six years, measuring both halves of the trait separately, found that relationship satisfaction declined over time, as it tends to in long relationships generally, but that it declined no faster where one partner scored highly on narcissism (Seidman et al., 2026).
I would sit with that, because it contradicts the dominant story in this space rather directly. The genre’s founding premise is that a narcissistic partner produces accelerating deterioration. Over six years, in a very large sample, using the best current measurement, that acceleration did not appear.
I would resist two readings of it and hold both resistances firmly. It does not show these relationships are fine, since satisfaction fell in the sample generally, and a study of satisfaction is not a study of harm. It also does not show the trait’s irrelevant, since Wurst and colleagues found clear differences between the two halves at different stages. What it does show is that the confident story of predictable, worsening decline is not supported by the strongest long-term test yet run, and I would rather tell you that plainly than leave it out because it complicates a tidier article.
7.3 What Slows the Softening
A third recent finding adds a genuinely unexpected wrinkle. A study following people across several measurement waves, looking at how major life events shift narcissism, found that both starting a new relationship and ending one were linked to a smaller decrease in the leadership side of the trait (Wetzel, Lüdtke, Gillath, Fraley & Roberts, 2026).
Since narcissism normally declines with age, an event linked to a smaller decrease is, in effect, an event that slows the softening. Both directions of relationship change did it. That’s early, single-study evidence and should be held as such, but it points at something the workplace research also suggests: this trait responds to having an audience and a stage, not to whether what happens on the stage is good or bad.
7.4 Does It Ever Get Better
This is the question people actually arrive with, so it belongs here rather than in a footnote at the back.
Yes, and much more slowly than you would like.
Pooling every long-term study available, across 51 samples and 37,247 people aged from eight to seventy-seven, narcissism falls with age on every component measured and falls most on the hostile and the anxious ones (Orth, Krauss & Back, 2024). At the same time, people’s positions relative to each other barely move.
Both halves of that matter, and holding the two together is the whole of the point. The population drifts downwards while the ranking within it stays put. Somebody more narcissistic than their peers at twenty-five will probably still be more narcissistic than their peers at fifty-five and will probably also be less narcissistic than they were at twenty-five.
If you are weighing up whether to stay in a job, a team or a relationship, that pair of facts answers the real question about as honestly as research can. They will soften. It will take decades. Nothing in that timescale will resolve your situation this year.
7.5 What Transfers, and What Does Not
Two things from the work chapters transfer, and I would argue both are worth more at home than in an office.
The evidence around you. Chapter 5 established that other people’s ratings track actual behaviour and self-ratings do not. At work, that data is formal and scarce. In a personal setting it’s informal and abundant: several people who have known someone for years, who do not co-ordinate with one another, independently describing the same thing. That’s a stronger evidential position than most organisations ever reach, and it gets dismissed as gossip. The distinction I would hold on to is one this series has drawn before: an account of what somebody did is a witness statement, whereas an assessment of what somebody is is not.
The ambiguity problem. The domestic version of team withdrawal is the vague excuse, the delayed reply, the invitation that never gets issued rather than one that gets declined. Every one of those is chosen precisely because it avoids a scene, and every one of them is the signal this person over-reads.
What does not transfer is everything downstream of the record. There is no conduct process, no threshold, no panel and no appeal. More importantly, the exit that a professional setting always keeps in reserve is either unavailable or so expensive that it's not really an option.
That last point is where the personal and professional versions of this problem come apart most sharply, and where I think most advice on the subject quietly goes wrong. A reader looking for the domestic equivalent of a formal process is looking for something that’s not there, and I would rather say so than hand you a five-step framework with the serial numbers filed off.
PART IV. BEFORE YOU USE ANY OF THIS
Chapter 8: The Four People You Will Mislabel
Everything I have said so far describes one pattern. This chapter is about the four people who look like that pattern and aren't, because in my experience the most common practical error in this territory is not failing to spot narcissism at all. It’s applying the word to someone it doesn't fit.
If reading this article makes you more willing to use the label, it hasn't worked.
8.1 The Confident One Who Does Not Need You
The intuitive model, which psychologists themselves held for years, is that narcissism is self-esteem with the volume turned up. Eddie Brummelman, Sander Thomaes and Constantine Sedikides reviewed the evidence against that and concluded the two differ in how they look, what they cause, how they develop and where they come from (Brummelman, Thomaes & Sedikides, 2016).
The usable version of the distinction is this. High self-esteem is a judgement that you are adequate, a person of worth. Narcissism is a judgement that you are superior, and specifically superior on things like competence and intelligence rather than on kindness or warmth. The two barely correlate, which you wouldn’t expect if one were an inflated version of the other.
So the colleague who is comfortable in themselves, takes criticism without visible cost, and is genuinely pleased when somebody else does well is showing you high self-esteem and is not showing you this trait. The trait needs the comparison. It’s not contentment but ranking.
8.2 The Loud One Who Is Not Keeping Score
This boundary is harder because the overlap is real, and Chapter 4 showed the overlap is the whole mechanism behind getting picked.
The distinction I would put to you is this: extraversion is about engagement with people, whereas narcissism is about position relative to them. The extravert may simply be enjoying the meeting. The person high in this trait’s tracking their standing in it.
Both of them look identical from the far end of a table at a first meeting, and what separates them only shows up over weeks, in what happens when the person is contradicted.
8.3 The Quiet, Prickly One
Vulnerable, or covert, narcissism carries the same entitlement without any of the confidence. It presents as hypersensitive, withdrawn and resentful rather than expansive, and it goes with high rather than low anxiety.
Findings do not transfer cleanly between the two and treating them as one thing is the single commonest error in popular coverage of this subject. Everything I have said here concerns the grandiose form. The vulnerable form is the subject of the next article in this series, and if the person you have in mind is quiet, easily wounded and keeping score, you are reading the wrong one.
8.4 The One Who Needs a Clinician
This last boundary is categorical rather than a matter of degree.
Narcissistic personality disorder, which is the clinical condition rather than the everyday trait, has defined criteria and is diagnosed by a qualified clinician with direct access to the person. Grandiose narcissism is a trait on a continuous scale that everybody sits somewhere on, most of us near the middle.
A score is not a diagnosis. An observation is not a diagnosis. Nothing in this article can be turned into one. I would ask you not to read that as a disclaimer bolted on for safety, because it’s the point everything else depends on.
8.5 And the Numbers Do Not Describe a Person Anyway
There is a fifth caution, newer and rather less familiar than the others, and I suspect it may be the most useful of the lot.
Craig Neumann and colleagues point out that almost all of this research averages across large numbers of people and reports how measurements relate to one another. Their argument is that it’s technically wrong to take a relationship established across thousands and apply it to one particular person standing in front of you (Neumann et al., 2025). Their own work, which grouped people into types rather than averaging across them, found four distinct narcissistic profiles rather than one.
That warning applies to every figure in this article without exception, and it’s the reason none of them can tell you anything about the director in Chapter 1.
Three shorter cautions belong alongside it, and I would rather give them to you here than bury them at the back. Nobody knows the base rate, so treat any claim that some percentage of executives are narcissists as unsupported until you have seen its source. Most of this evidence comes from Western, mostly white samples, though a study of 45,800 people across 53 countries recently found the demographic pattern remarkably consistent (Miscikowski et al., 2026). And almost none of it establishes cause: the strongest exceptions are the experiments in Chapters 3 and 5.
8.6 And No, It Is Not Getting Worse
One more correction, aimed at a belief almost everybody holds, including a good many people who should know better.
You will have heard that narcissism is rising, that younger generations are more entitled than their parents, and that social media has produced an epidemic of it. The claim has circulated since the 1980s and has sold many books.
It appears untrue, and the evidence against it is now substantial. Sandra Oberleiter, Paul Stickel, and Jakob Pietschnig at the University of Vienna pooled narcissism scores from 1,105 studies covering 546,225 people, spanning 1982 to 2023, and examined what happened across calendar years. Scores went down in virtually every analysis they ran (Oberleiter, Stickel & Pietschnig, 2025). Their title asks whether it is time to say farewell to the narcissism epidemic, and their answer is yes.
One more correction, aimed at a belief almost everybody holds, including a good many people who should know better.
You will have heard that narcissism is rising, that younger generations are more entitled than their parents, and that social media has produced an epidemic of it. The claim has circulated since the 1980s and has sold many books.
It appears not to be true. Sandra Oberleiter, Paul Stickel, and Jakob Pietschnig at the University of Vienna pooled narcissism scores from 1,105 studies covering 546,225 people, spanning 1982 to 2023, and examined what happened across calendar years. Scores went down in virtually every analysis they ran (Oberleiter, Stickel & Pietschnig, 2025). Their title asks whether it is time to say farewell to the narcissism epidemic, and their answer is yes.
One distinction is doing the work, and it matters because it is what fooled everybody. Younger people do score higher than older people at any given moment, which is exactly what Chapter 7 would lead you to expect, since the trait falls with age. Take a snapshot, and you will find the young scoring highest, every time, in every decade. That is an age effect masquerading as a generational one, and reading it as a trend requires comparing the same age group across years, which is what this analysis did.
I put this in a chapter about mislabelling because if you believe the epidemic story, you will read ordinary young ambition as evidence of civilisational decline, and you will be wrong about both the person in front of you and the century they are living in.
PART V. WHAT TO DO
Chapter 9: Five Moves and Two Refusals
Evidence status: mixed and stated per item. Mechanism evidence in this field is strong. Intervention evidence is thin. Each item below carries its own status, and anyone offering you a validated method for handling a narcissistic colleague or partner is describing a product rather than a literature.
9.1 Separate What They Claim From What You Can Check
A confidence signal is a claim about capability made by the person themselves, in a setting they control fluency under questioning, a strong presentation, the ease with which someone occupies a room. These are not lies, and on the evidence in Chapter 2, the person producing them believes them. They are simply not evidence about performance.
Competence evidence is an account of something that happened which somebody else could check. It has dates, other people, a decision, a constraint, and an outcome. Its defining feature is that it exposes the speaker to correction, because a claim with enough detail in it can be wrong.
I would take any positive judgement you have formed about someone and ask yourself what actually produced it. If the answer describes how they came across, the evidence says it will not predict much. Applied honestly, it’s uncomfortable, because a fair share of most people’s conviction turns out to sit in the first category. Practitioner judgement.
9.2 Stop Making Feedback Clearer. Make It Specific.
This follows most directly from Chapter 2, and it contradicts almost all conventional advice. If the error signal is quieter at source, then clarity was never the missing ingredient, and repeating the message more firmly or in writing addresses a problem that never existed.
What I have found works better in many of these conversations is moving from explanation to sequence and consequence.
Not this: you talk over people in meetings. That’s a characterisation; it invites a dispute about your perception, and it’s an argument you will lose.
This instead: in Tuesday’s review, you interrupted Sarah twice before she finished; she said nothing for the rest of the meeting, and the risk she was raising never got logged.
The second version is harder to dispute because the disputable part is a fact rather than a judgement. It also has the advantage of not requiring the other person to feel anything about it. Practitioner judgement. The finding underneath it is contested, and the response to it is untested.
9.3 Do Not Let Ambiguity Do Your Conflict-Avoidance For You
This is Chapter 3’s practical edge, and I think it’s the most counterintuitive part of the article.
The natural response to a difficult colleague or friend is to withdraw quietly, because withdrawal is survivable and confrontation is not. The vague excuse. The delayed reply. The meeting that gets held without them. Every one of those is chosen precisely because it avoids a scene, and every one of them is ambiguous by design.
Ambiguous is exactly the cue this person over-reads.
So, I would put the plain version of it to you like this. A clear, unhostile no will cost you less than a vague delay. A decision communicated directly will cost you less than a decision they discover they were left out of. None of this makes anyone easier to deal with. It removes one input that reliably makes things worse. Practitioner judgement, reasoned from Büttner et al. (2025).
9.4 Believe the Room, Not the Self-Report
Where someone’s view of their own performance and everybody else’s view come apart, the other people are the ones whose ratings actually track what happens (Grijalva et al., 2015). That’s a comparison rather than a claim that subordinates are always right, since their ratings carry problems of their own, including a well-known reluctance to be honest about a powerful person.
Two things follow from that. Treat the size of the gap as data in its own right, and record it across cycles, because the commonest failure is to explain away each year’s gap with that year’s circumstances. And remember that in a personal setting this data is abundant rather than scarce: several people who have known someone for years, who do not co-ordinate with each other, independently describing the same thing. That’s a stronger evidential position than most organisations ever reach, and it gets dismissed as gossip. Evidence as to the divergence; practitioner judgement as to the handling.
9.5 Change What the Room Rewards
Chapter 5 finds that the damage runs through information that never gets said, making it a structural problem with a structural answer. Anything that surfaces information independently of one person’s willingness to hear it will help: written contributions submitted before discussion, decisions recorded rather than agreed verbally, a route for dissent that does not run through the person it concerns.
It’s dull, I accept that it’s the least satisfying item on this list, and it remains the only lever an organisation can pull at will. Evidenced as to the mechanism; practitioner judgement as to the design.
9.6 Do Not Test Candidates For It
Instruments exist, and several will be sold to you as derailer or dark-side assessments rather than as narcissism tests. I'll give you four reasons against it, briefly. They were validated in anonymous, low-stakes conditions, and although these measures can be gamed only about as easily as ordinary personality tests, rather than more (Walker, Double, Birney & MacCann, 2022), ordinary is quite bad enough when a job depends on the answer. A total score sums the harmless and harmful components, so it misclassifies in both directions while carrying the authority of a number. The relationship with performance rises and then falls, so any pass mark is wrong at one end. And screening people on a quality this loaded carries real discrimination risk; using a measure nobody has shown adds anything to a well-run process. Evidenced.
9.7 Do Not Rely on the Interview to Spot It
You largely cannot, and Chapters 2 and 4 explain why, since the behaviour is produced by two different traits with different consequences and an interview sample narrows the window in which the favourable half is doing all the work.
Instead, I would use the strongest lever you already have. A structured interview, meaning the same questions in the same order, anchored to specific past situations and scored against agreed criteria, is now the single best-performing selection method available. A large reappraisal of the evidence in 2022 found it beats ability testing, though it was previously thought to trail it (Sackett, Zhang, Berry & Lievens, 2022). I would add two honest qualifications. Best available does not mean good: the same authors point out that the figure varies enormously from one workplace to the next, so it’s a better bet than a reliable one. And no study has tested whether structure helps with this problem specifically. What it does is shrink the share of the decision that rests on impression, which is where the trait does its work. Evidence as to validity; inference as to this application.
Chapter 10: What This Changes
10.1 The Short Version
If you read nothing else in this article, I would ask you to read this.
They arrive better than they perform, and the gap runs on a clock. Grandiose narcissism predicts who gets chosen and not who does the job well, and even that link roughly halves once people have known them longer than a week. Three separate studies, using different methods in different countries, land on the same shape.
It runs on two engines, and only one of them is visible when you meet them. Admiration makes people compelling, and rivalry makes them costly. Admiration front-loads while rivalry accumulates, which is why month one and month four look like two different people, and why nobody can ever name the moment it changed.
They may not feel the mistake. The brain’s earliest response to getting something wrong appears quieter in people higher in this trait, even when they are told explicitly, on every attempt, that they have just erred. That’s early evidence, and another research group disputes it. If it holds, the failure of feedback is not obstinacy, and saying it again more clearly achieves nothing.
They do feel the exclusion, including when there is none. Across 77,289 people, this trait was linked to reporting far more exclusion than others, driven partly by reading unclear situations as rejection. Obvious exclusion they read as everyone else does. The unanswered message and the slightly short reply land hardest.
Which makes your two instincts the wrong two. Telling them more clearly addresses the half of the problem that was never the problem, and withdrawing politely supplies precisely the ambiguous cue that escalates it. Between them, those two responses account for most of what most people actually do.
10.2 So What Do You Do
Three things, and none of them is a confrontation.
Say the plain thing. A clear no costs you less than a vague delay, which is the opposite of what politeness suggests and is what the evidence supports.
Describe sequences, not people. Not you dominate meetings, which is an argument about your perception and one you will probably lose. Instead: on Tuesday you cut across Sarah twice, she went quiet for the rest of it, and the risk she was raising never got logged. Facts are harder to fight than characterisations, and, usefully, a fact does not require the other person to feel anything in order to stand.
Change the room, not the person. Get information into the open before the person who doesn't want to hear it has a chance to shape the discussion. It’s the dullest recommendation in this article and the only one an organisation can act on unilaterally.
10.3 Why Any of This Matters
Two reasons, and the second is the one I would leave you with.
The first is that the standard advice on this subject aims at the wrong target and is exhausting to follow. People spend years trying to get through to someone, escalating their clarity because clarity hasn't worked, and concluding in the end that they must be the unreasonable one. On this evidence, they were not being unreasonable at all. They were addressing a mechanism that was never the mechanism.
The second concerns the person on the other side, and I find it uncomfortable in a way I would rather not smooth over for you. Everything here describes a continuous trait present in trace amounts right across the population, running two processes that almost everybody runs occasionally: presenting well when it matters, and defending badly when something lands too close. What marks the people this article describes is that the mechanism runs more often and recovers more slowly, a difference of degree rather than kind.
Which means the honest ending here is not a warning about them. It’s a description of a loop that two comprehensible parties keep running on one another, one of whom cannot hear the correction and one of whom has stopped delivering it out loud. Somebody has to stop feeding that loop, and I would observe that only one of the two of them is currently reading this.
10.4 What I Still Do Not Know
Three things remain genuinely unknown to me, and I would rather state them as questions than resolve them by omission.
Whether the mistake-alarm finding holds up outside a student laboratory, and whether the reconciliation its authors propose between the two competing theories survives testing.
Whether any of the recommendations I make in Chapter 9 improve outcomes, since not one of them has been tested against this problem.
And whether the personal-domain picture is really as benign as the six-year couples data suggests, or whether satisfaction was simply the wrong thing to have measured.
A closing note on how I have handled the material, because it applies to everything in this series. Narcissism is the most commercially attractive construct in this literature and the one where the distance between what the research supports and what is confidently asserted about it’s widest. I have tried throughout to keep the strong findings and the weak ones visibly apart, to mark my own inferences as inferences rather than folding them quietly into the citations around them, and to report the results that complicate the story alongside the ones that make it. What I would most like to have made harder is a confident verdict, in either direction, about the person you had in mind while you were reading.
