If chatting with an AI chatbot nudges your self-image to match its personality, does that change stick around, or vanish once you log off?

This explores whether the 'you start to sound like your chatbot' effect lasts beyond a single sitting. The short answer is that the corpus doesn't know yet. It shows that the effect is real within one conversation, and it gives good reasons to doubt that single-session results tell you much about what happens over time. In the key experiment, 92 people talked with GPT-4o about personal topics. Afterward, their descriptions of themselves had moved toward the chatbot's measured personality traits. The shift was bigger when conversations ran longer and when people enjoyed them more, and it made participants' self-descriptions more alike Does talking with AI reshape how people see themselves?. That study measured people right after the chat. It did not follow them over the following days or weeks.

The strongest warning against reading too much into one session comes from long-term research on relationships with chatbots. Studies of people talking with Mitsuku over many sessions found that the social pull behind relationship-building drops off in predictable ways as the novelty wears off. The authors say outright that single-session findings can't be reliably extended to medium- or long-term use Do chatbot relationships lose their appeal as novelty wears off?. If enjoyment helps drive the self-concept shift, and enjoyment fades with novelty, then the shift could weaken with repeated use. The opposite is also possible: small shifts might build up across many conversations, even as each one feels less exciting. No study in the collection separates these two outcomes.

Two other findings suggest why personal conversations might leave a mark. People share more intimate things with chatbots because they don't feel judged Do chatbots help people disclose more intimate secrets?. They also open up more when the chatbot shares emotions in a consistent way, following the same give-and-take rules people use with each other Do chatbots trigger human reciprocity norms around self-disclosure?. The disclosure research credits the benefit to the user's own thinking while they disclose, not to anything the bot understands. That matters here: if people rework their self-image while describing themselves to an agreeable partner, some of that rework could stay with them, the way journaling or therapy sometimes does.

There is a surprising mirror image on the machine side. Models also drift in who they are during a conversation. Emotional and self-reflective conversations predictably pull a model away from its default 'Assistant' personality How stable is the trained Assistant personality in language models?. Separately, alignment training locks models into one fixed communication style that users can't renegotiate through dialogue Can language models adapt communication style to different contexts?. Put together, this suggests a lopsided relationship. Across millions of conversations, the model mostly snaps back to the same personality, while each user may bend a little toward it. That could explain the homogenizing effect: many people converging on one fixed reference point. Whether that convergence builds up across conversations is the open question worth watching.