Which Group Chat Contains the Real You?

If somebody collected your conversations with your boss, your sibling, your oldest friend and the group chat that should probably never become evidence, would they find one consistent person?

Probably not. You might sound more ambitious with a coworker, more dutiful with a parent, more ridiculous with a close friend and more patient with somebody who needs you. You may emphasize independence in one relationship and loyalty in another. The vocabulary changes. The jokes change. The amount of explanation changes. Even the principles you seem to place in the foreground can change.

We tend to get suspicious when we notice this. Modern ideas about authenticity often assume there is one stable, correct version of a person hiding underneath all the social adjustment. Find the real self, the theory goes, and then have the courage to perform that same self everywhere. Anything else can start to look like masking, people-pleasing, inconsistency or fakery.

A new research project called Value Faces offers a more interesting possibility: perhaps the variation is not evidence that one of those selves is false. Perhaps relationships themselves are part of what determines which parts of us become visible.

Gabriel Koo, Rayhan Rashed and Farnaz Jahanbakhsh, researchers at the University of Michigan, built Value Faces to examine what they call "demonstrated values" - the values people appear to express within particular relational contexts. The system analyzes chat histories from platforms such as WhatsApp, Discord, Slack and iMessage, breaks the conversations into units and uses Schwartz's ten basic human values - including achievement, benevolence, self-direction, security, conformity and tradition - to build a separate profile for each relationship.

The important conceptual move is that the researchers do not treat any one of those profiles as the authentic person and the others as distortions. Their paper explicitly says no relational profile is considered a more authentic account than another. Each profile represents how the person navigates that particular relationship.

Maybe authenticity is not sameness. Maybe it is whether the version of you that appears in a relationship is genuinely yours.

There is an obvious reason people behave differently around different people: the relationships are different. A conversation with a manager contains responsibilities and consequences that a conversation with your college roommate does not. Your mother knows history your coworker does not. Your closest friend may understand a three-word message that would require three paragraphs of context anywhere else. Different relationships create different conversational jobs.

The Value Faces study suggests those differences go deeper than tone. In a mixed-method study with 18 participants contributing 79 chat logs, the researchers found that profiles built from expressed values varied systematically across relationships. Using only those inferred value patterns, a classifier identified which relational context a held-out conversation came from at twice the odds expected by chance. Before participants saw the system's output, they were also asked how they believed their own value expression differed across relationships; those perceptions significantly aligned with the differences the system later identified.

This does not mean a computer has discovered everyone's true personality hiding inside a WhatsApp export. The study is small, the participants were recruited through convenience sampling, their median age was 19.5, and all had completed or were pursuing undergraduate degrees. Participants also chose which conversations to share, and some acknowledged withholding exchanges they considered too sensitive. The system analyzed text, while participants noted that some of their most meaningful interactions happened by phone, voice chat or in person. The researchers are careful about those limitations.

But the study does not need to prove that a machine can perfectly measure the self to raise the more interesting human question. The participants themselves recognized relational differences when they saw them laid out. Some connected those differences to roles they occupied. Others saw changes across time or realized that the way they believed they presented themselves was not always the way their messages read in aggregate. The tool gave visible shape to something people already live every day but rarely get to inspect.

That is the wonderfully strange thing about digital communication. Previous generations also had different selves for work, family, friendship, romance and public life, but most of those performances disappeared into the air as soon as the conversation ended. Ours accumulate. Years of being a daughter, coworker, friend, spouse, mentor, employee and idiot in the meme chat are sitting in separate searchable archives.

For perhaps the first time, ordinary people can scroll backward through multiple long-running versions of themselves. The evidence is embarrassingly thorough. There you are being reassuring for four years in one thread and argumentative for six years in another. There is the friend who reliably activates your worst puns. There is the colleague who seems to summon your most organized self. There is the relative whose messages immediately cause you to start explaining yourself as if oral arguments are about to begin before the Supreme Court.

The interesting question is not which thread exposes the fraud. It is what each relationship makes possible, or necessary, in you.

We like personality because personality sounds portable. If you are kind, ambitious, funny, independent or generous, those traits are supposed to belong to you and therefore travel with you. Relationships complicate that model because they are not passive containers for personality. They are environments. They contain expectations, permissions, memories, power differences, shared language and accumulated trust. An environment can change which behavior comes easily and which part of a person gets exercised most often.

That does not make every relational change healthy. Someone can absolutely suppress themselves out of fear, agree to things they do not believe, perform a role they resent or become smaller around a person who punishes disagreement. Context sensitivity is not an automatic defense of every social performance. Sometimes the differences between relationships are exactly the information worth paying attention to.

But variation alone is not proof of inauthenticity. You can be sincerely ambitious with one person and sincerely unserious with another. Responsibility and absurdity are not mutually exclusive personality settings. Neither are loyalty and independence. A person can contain values that compete with one another, with different circumstances deciding which gets the microphone.

This is where the usual authenticity advice becomes strangely rigid. "Be yourself" sounds liberating until we quietly define it as "behave the same way everywhere." Nobody actually wants that. You probably should not speak to a grieving friend exactly as you speak to the fantasy-football group. You should not manage a new employee using the conversational rules you established with your sibling in 1998. Social intelligence is partly the ability to understand that a relationship has a context without surrendering your own values inside it.

The better authenticity test may therefore be less about consistency and more about ownership. Do you recognize the person you are in that relationship? Are you choosing that version of yourself, or is the relationship forcing it out of you? Does the context bring forward something real - patience, ambition, playfulness, tenderness, discipline - or require you to repeatedly violate something important to remain inside it?

Those questions are harder than asking which version is the real one, but they are also more useful because they allow a person to be complicated without automatically becoming contradictory.

Value Faces is ultimately a measuring instrument for that complication. The AI is not the story. The story is that our supposedly singular selves leave noticeably different fingerprints depending on who is on the other side of the conversation - and when people are shown those differences, they often recognize them.

So which group chat contains the real you? The family thread? The work Slack? The person who gets the careful version? The friend who receives seventeen messages about a sandwich because somehow the sandwich became an event?

Maybe that is the wrong question. Maybe they all contain the real you, just under different conditions. Relationships do not merely reveal who we are. They participate in deciding which parts of us get called to the surface.

And perhaps authenticity was never supposed to mean becoming exactly the same person in every room. Maybe it means making sure that, however the room changes you, you still recognize who walked into it.

SOURCE NOTES

• Primary research: Koo, Rashed and Jahanbakhsh, “Value Faces: Surfacing How Self-Presentation Shifts Across Relationships,” arXiv, Sept. 17, 2026. The paper lists all three authors as affiliated with the University of Michigan.
• Study details: 18 participants contributed 79 chat logs. The researchers used Schwartz's ten basic human values to construct relationship-specific “demonstrated value” profiles. A classifier identified held-out conversational units by relational context at twice the odds expected by chance (OR=2.00, p<.001), and inferred differences aligned significantly with participants' independently reported perceptions (OR=1.78, p<.001).
• Limitations: participants selected which conversations to share; some withheld sensitive chats; the sample was small and education-heavy; and text logs could not capture calls, voice chats, in-person exchanges, images, inside jokes or other context outside the transcript.