Baxnet Ideas · Founder note
The Mentor in One Chat Is the Mentee in Another
TL;DR: We do not have one messaging voice. Each relationship calls forward a different version of us before we choose any words.
At 9:12, a former colleague asks whether they should accept a new job.
You ask what they actually want, point out the risk they have missed, and tell them they are probably more ready than they think. They came to you for advice, so you become the calm one.
At 9:40, you open another chat and ask an old manager almost the same question about your own work. The confidence disappears. You explain too much, revise the message twice, and end with, Would value your honest view.
Later, a friend sends a photograph from a disastrous holiday. No advice is expected. Your job is to find the one detail that will make them laugh.
None of these versions is false. None is the whole person either.
We do not carry one messaging voice from chat to chat. Each relationship calls forward a different version of us before we have chosen any words.
The role exists before the sentence
People often talk about a personal writing style as if it is something we own independently: my tone, my voice, how I text.
There are certainly habits that travel with us. Some people use full stops. Some send six short messages where one would do. Some apparently cannot answer a simple question without locating the correct GIF.
But a chat history also records adaptation.
The person who treats us as a mentor keeps asking for our judgement. We learn to answer carefully because our confidence carries extra weight. With our own mentor, we may expose uncertainty that would never appear in the first conversation. A friend who has laughed at our jokes for fifteen years gives us permission to be funny. Another friend may know us as the person who does not make light of things.
The role is produced between people. It is partly who we are and partly who the other person has learned to expect.
Research into communication accommodation gives this everyday observation some useful support. Two studies of instant messaging found that people tended to converge on the length and duration of their contributions, with differences between friends and strangers and between social and task-based conversations. A separate set of experiments found that relative social power affected linguistic accommodation in instant messages, and that the interpersonal effect depended on the norms of the situation.
These studies do not map the private roles inside long relationships. They show something narrower: even in typed conversation, language responds to who is present, what people are doing, and the position they occupy relative to one another.
A familiar tone can still make the wrong move
This creates a harder problem for AI writing tools than reproducing somebody’s punctuation.
A system may learn that I use more emojis with one person, write longer replies to another, and become unusually formal with a third. Given enough examples, it may produce something that resembles the surface of each conversation.
The more consequential decision comes earlier.
Am I meant to reassure this person or challenge them? Do they want an answer from me, or do they want room to think aloud? Is humour an expression of closeness here, or a way of avoiding the subject? Can I admit that I do not know? How direct am I allowed to be?
A drafted message can sound uncannily like us while getting one of these decisions wrong. It can offer advice in a relationship where we usually ask questions. It can manufacture warmth where mutual bluntness is a form of trust. It can make the funny friend perform on an evening when the decent response would be to stop performing.
That last example matters because a role is not a contract. Sometimes the most human message is a departure from the pattern.
The mentee disagrees with the mentor. The dependable one admits they cannot help. The person who normally fixes the problem asks to be looked after. A system trained to continue a relationship’s established style may be very good at returning everybody to their usual positions just when one of them is trying to move.
The judgement is part of the communication
I am uneasy with the idea of outsourcing personal messages because writing is not merely the conversion of a finished thought into acceptable prose.
The hesitation, the decision to soften something, the abandoned joke, and the sentence we finally choose to stand behind are part of how we take responsibility for speaking to another person. Deciding which version of ourselves the moment requires is part of that work too.
Could a system infer some of these roles from enough conversation history? Probably. It could identify patterns and make increasingly convincing predictions. That would not make the prediction neutral, or make a familiar pattern the correct way to continue the relationship.
This is also a useful limit for products that analyse chats. They may be able to show that someone often advises in one relationship and seeks advice in another. As I argued in The Organiser Is Not a Personality Type, a role found in one room should remain attached to that room rather than becoming a permanent description of the person.
Before a message is written, there is usually a smaller decision: who am I to this person today?
We make it so quickly that we barely notice. I am not convinced it is merely overhead that should be automated away.
Further reading: Communication Accommodation in Instant Messaging, Linguistic Style Accommodation Shapes Impression Formation and Rapport in Computer-Mediated Communication, and When AI Helps Write the Message, What Does the Conversation Still Tell Us?.