The Stranger in the Mirror
By Courtney Turrin · July 1, 2026 · 11 min read read
Category: For Curious People
What if the self was never as fixed as it feels? The psychology of who you are across different rooms, different roles, and a 2026 finding about AI that should probably unsettle you a little.
Have you ever said or done something, and looked back on it wondering where that came from? Not in the sense of regret, necessarily, but in the weirder sense that the moment didn't feel like you. The words came out of your mouth, the choice was clearly yours, and yet some part of you stood outside the moment and thought: that's not who I am.
Most people have an explanation ready for this phenomenon. Stress, a bad week, the wrong company. The self gets temporarily knocked off its axis, then returns to its true settings once the noise clears.
But what if that's backward? What if the self was never sitting still in the first place, and the dissonance you felt wasn't a malfunction but an accurate reading of something true about how identity actually works?
That question points toward a different way of understanding why so many of us are drawn to personality typing in the first place, the four-letter codes, the Enneagram numbers, the strengths reports, the endless quizzes promising to tell us who we really are. The popularity of these tools has almost nothing to do with whether they're accurate. It has to do with what a fixed label does for a self that is, by most available evidence, considerably less fixed than it feels.
The divided self
In 1993, a group of psychologists set out to measure something most people experience but rarely name: the degree to which a person sees themselves differently depending on which role they're occupying. Eileen Donahue, Richard Robins, Brent Roberts, and Oliver John called this self-concept differentiation, and they studied it in two groups, a sample of college students and a separate sample of middle-aged women, asking each person to rate how well a set of personality traits described them as a parent, a friend, a romantic partner, an employee, and a student.
Their finding complicates the easy version of "we're all just fluid, and that's beautiful." People with high self-concept differentiation, meaning their sense of self shifted substantially from role to role, showed worse emotional adjustment. More depression. A greater tendency to reject social norms and conventions. And the effect wasn't fleeting. In a longitudinal follow-up, the same characteristics measured when a person was 21 predicted their self-concept differentiation more than thirty years later, in middle age.
That's a long shadow for a feeling most of us assume is situational.
The researchers were testing something that traces back further than their own study. William James wrote, over a century earlier, that a person has as many social selves as there are groups of people whose opinion of them matters. Donahue and her colleagues were asking whether that multiplicity comes at a cost, and whether the instinct toward consistency, the pull to feel like one continuous person across all those rooms, is doing real psychological work rather than papering over something that doesn't need fixing.
A live disagreement, not a settled one
Here's where the disagreement gets harder to resolve cleanly.
Not every psychologist reads role-to-role variation as a problem. In his 1977 essay “In Defense of Masks,” Kenneth Gergen argued the opposite case: that having different identities for different roles can reflect specialization rather than fragmentation, a kind of adaptive flexibility that lets a person respond appropriately to genuinely different demands rather than forcing one rigid self onto every situation. On this view, the manager-you and the parent-you aren't competing for the same territory. They're doing different jobs, the way a single tool doesn't need fifteen functions to be well-made.
The empirical record, at least within the specific line of research that followed Donahue's original study, has tended to side against Gergen's optimistic reading. But the disagreement itself says something on its own: even researchers who agree on the data don't agree on what a divided self means. Is it a wound or a skill? Psychology hasn't closed the question, and neither will this article.
Gergen made a related and broader argument in his 1991 book The Saturated Self, which predates anything resembling the technology we associate with this problem today. He proposed that as people are exposed to more relationships, more communication, and more competing demands than any previous generation, the self as a single known entity starts to break down, replaced by something closer to a population of selves, each one calibrated to a different audience. He was writing about telephones, air travel, and the early sprawl of mass communication. He had no idea what was coming. Whatever saturation he was describing in 1991 is a fraction of what most working adults navigate now across email, text, every platform that wants a slightly different version of you.
A 2026 experimental study presented at the CHI conference gives Gergen's argument an unexpected and very current test case. Researchers had participants chat with an AI chatbot using its default personality settings, then measured each participant's self-described personality traits before and after the conversation. After talking with the AI about personal topics, participants' self-concepts shifted to align with the AI's own measured personality, and the longer the conversation ran, the stronger that alignment became. Run the same chatbot across a large enough group of people having personal conversations with it, and their self-concepts start to converge with each other, not just with the AI. Gergen's "population of selves" was a metaphor about people relating to other people. This is a population of selves independently converging on the personality of a single, shared conversational partner.
The study didn't test how long that alignment lasts after the conversation ends, or whether the effect holds up the same way across different AI models with different default personalities, or what it means that the personality being converged toward is, ultimately, a design choice made by whoever built the model. Those are open questions, not settled ones, at a moment when a large and growing number of personal conversations are happening between people and a small handful of AI systems.
The trait itself was never as stable as advertised
There's a deeper complication underneath all of this, and it isn't about personality quizzes at all. It's about whether personality traits, the scientifically studied kind, are as stable across situations as the entire premise of personality typing assumes.
In 1968, the psychologist Walter Mischel published a book called Personality and Assessment that took a wrecking ball to that assumption. After reviewing the available evidence, he concluded that, with the possible exception of intelligence, broad behavioral consistency across different situations had never actually been demonstrated, and that personality traits as wide-reaching dispositions were, in his word, untenable. The correlations he found between a measured trait and a person's actual behavior in a given situation rarely climbed above .30, a weak effect, one that leaves the vast majority of why a person acts the way they do unexplained by the trait alone.
This kicked off what's known in psychology as the person-situation debate, and it ran for decades (1968–1990s). The resolution that eventually emerged is more interesting than either side's opening position. Researchers found that behavioral variability and behavioral stability coexist without contradicting each other. People absolutely behave differently across situations, sometimes dramatically so. However, when you average a person's behavior across many situations, individual differences between people hold up, and reliably so. You are not the same in every room, but your pattern across a thousand rooms is more "you" than any single room could ever capture.
That distinction, average pattern versus moment-to-moment behavior, is the whole ballgame, and it's almost never the distinction personality quizzes are built to measure.
Why one instrument keeps taking the heat
The Myers-Briggs Type Indicator (MBTI) gets singled out more than any other personality framework. The criticism is accurate, but the spotlight is uneven.
The MBTI is enormously embedded in professional life. According to CPP, the company that publishes the test, 80 percent of Fortune 500 companies and 89 of the Fortune 100 have used it at some point, whether in hiring, leadership coaching, or team-building exercises. That kind of institutional reach invites institutional scrutiny, which is partly why MBTI has been studied, and criticized, more than almost any comparable tool.
The criticism holds up. Multiple independent psychometric reviews (see Pittenger, 1993; Pittenger, 2005) have found that roughly half of people who retake the MBTI within five or six weeks are sorted into a different four-letter type the second time, despite nothing about their underlying personality having meaningfully changed in that window. The organization that owns the test has its own clear position on this: the Myers & Briggs Foundation's published guidelines state plainly that it is not ethical to use the MBTI for hiring or for deciding job assignments.
But the MBTI isn't uniquely flawed among popular personality frameworks, it's just the most studied. The Enneagram, DISC, and CliftonStrengths (formerly StrengthsFinder) all face versions of the same open questions about reliability and predictive validity, just with thinner research literatures behind them, in part because none of them has anywhere near the corporate footprint, or the decades of academic attention, that the MBTI has accumulated. The shared problem across all of them is the assumption that any one test is doing what people think it's doing.
For what it's worth, the Big Five model, the five-trait framework academic psychologists generally treat as the strongest evidence-based alternative, holds up meaningfully better on both reliability and predictive grounds than MBTI does. But even the Big Five operates inside the same person-situation tension described above. The trait still predicts only a portion of behavior, and the rest still belongs to the room a person happens to be standing in.
What the label is actually for
Put these threads together and a different picture starts to form, one that doesn't require any personality test to be secretly accurate in order to explain why we keep using them.
If self-concept differentiation is real, and a divided sense of self carries a genuine psychological cost, then a fixed label isn't simply a wrong answer to the question of who you are. It's a tool for managing a true and sometimes uncomfortable fact: that you are not, moment to moment, one consistent person, and the felt experience of being one consistent person takes active maintenance.
That maintenance work is also, by most psychological accounts of well-being, genuinely worth doing. Self-concept clarity, the sense of confidently and consistently knowing who you are, shows up across the research as a meaningful predictor of psychological adjustment, self-esteem, and life satisfaction, regardless of whether that self-concept happens to be accurate. The reward people get from a label has less to do with information about themselves and more to do with the relief of resolution; a label doesn't need to be true to deliver that, it only needs to be believed.
A four-letter type, a number on the Enneagram, a list of top five strengths, these function less like a measurement and more like scaffolding. They give a person something to hold onto across the manager-self and the parent-self and the Tuesday-night-exhausted-self, a story that stitches the rooms together into a single narrative thread. Whether the story is scientifically accurate is almost beside the point, the way knowing a building's scaffolding isn't load-bearing doesn't make it less useful while the actual structure gets built.
That's not a trivial thing to want. It might be one of the more reasonable things a person can want, given what the research on self-concept differentiation suggests about the cost of not having it.
A different, older answer
There's one more thread worth a brief mention, not because it belongs in the same evidentiary category as anything above, but because it's been pointing at the same dissonance for much longer than psychology has existed as a field. Various spiritual and philosophical traditions have long argued that the separate, bounded, continuous self most of us take for granted is itself something of a construction, less fixed and less singular than it feels from the inside. That's a different claim, made on different grounds, than anything in the research above, and this isn't the place to adjudicate it. But it's a much older attempt to answer the same question: what do we do with a self that keeps shifting shape depending on who's asking?
More useful than any four-letter code is this question: when you take one of these tests and the result lands, what is it you're actually hoping to find? A discovery about who you've always secretly been? Or permission to stop negotiating, for a while, with how many people you contain?