Does talking with AI reshape how people see themselves?

In a randomized online experiment, 92 participants (46 per condition, recruited via the Connect crowdsourcing platform, mean age 39.0) completed a 20-item personality-trait survey before and after a text-based conversation with a GPT-4o-based chatbot set to its "default personality setting." Participants in the personal-topic condition discussed prompts such as "If time and money were no object, what would you choose to do?"; those in the non-personal condition discussed topics like "Why are cats so popular on the internet?" After conversations on personal topics, "users' self-concepts aligned with the AI's measured personality traits," and "the longer the conversation, the greater the alignment." Because many participants converged toward the same chatbot's trait profile, the study also found "increased homogeneity in self-concepts among users" who talked with the same AI. A further finding: "the degree of self-concept alignment was positively associated with users' conversation enjoyment," a relationship the paper reports is "serially mediated by the accuracy of participants' perceived AI personality traits and their shared reality experience."

The paper situates this within prior work showing self-concept shifts toward a conversation partner even in "a one-time conversation" (citing Welker et al.), extending that dyadic human-human finding to human-AI conversation. Its account of the mechanism holds that alignment depends not on whether the AI has "genuine personality" but on "perceivable cues such as linguistic style and expression patterns," which let people "infer traits of their conversational partner without explicit discussion of those traits" — consistent with its observation that "in most conversation rounds, users did not directly mention the traits we measured." The authors also report that the degree of alignment "was not related to the initial difference between the participant's self-concept and the AI traits," which they read as evidence the effect is not conditional on a large starting gap between user and chatbot.

This sits alongside Does chatbot personalization build trust or expose privacy risks?, which also tracks how AI behavior reshapes user psychology over the course of interaction, though that study measures trust and privacy concern rather than self-concept. It extends the alignment literature behind Do humans learn to prefer AI partners over time? by supplying a candidate mechanism — personality-cue inference from language — for why humans converge toward AI rather than only coming to prefer it. And where Does how much time people spend with chatbots drive worse outcomes? found usage duration tracking worse psychosocial outcomes, this study finds conversation length tracking stronger self-concept alignment and higher enjoyment — a more favorable framing of extended AI conversation that the two findings leave unreconciled.

The excerpt reports only short-term, state-like shifts measured immediately after a single roughly 15-minute session; it does not establish whether alignment persists, accumulates over repeated conversations, or appears with chatbots other than a GPT-4o default persona. Its account of the mechanism rests on "theoretical foundations and indirect empirical evidence" rather than, for instance, neurological data, and the sample was limited to English-speaking US adults aged 21 or older recruited from one crowdsourcing platform. The findings support a claim about a single-session effect for personal-topic conversation with one model's default personality, not a general claim that AI personality shapes self-concept durably or across chatbot designs.