ACII 2026 · PUEBLA, MEXICO
Questions of Character
We gave three language models rich identities based on 290 measured personality profiles, then asked fifteen open-ended questions about fictional characters: who they identified with, who felt most unlike them, and who changed what they aspired to.
Their choices reflected the personalities they had been given. A profile can shape a model’s choices, not just how it describes itself.
real people’s personality profiles
answers about fictional characters
of aspiration answers named the same man: Atticus Finch
The problem
Tell a model to have a trait and it will happily claim it. Whether it then behaves that way is another matter (Han et al., 2025).
The instruction. A one-line trait label.
The illusion. The model says the right thing about itself, then does whatever it was going to do anyway.
Our hypothesis: thin prompting. A trait label gives the model very little of a person. A rich profile describes values, relationships and characteristic responses.
The idea
People relate to fictional characters for different reasons, and an open-ended question can reveal more than a single favourite. If people express identity through characters, what do profile-conditioned models do?
Who feels most like me?
Who do I wish I were more like?
Who do I feel is misunderstood?
How we tested it
HEXACO plus trust and social style
about 1,000 words each
Gemini 3 Flash, Grok 4.1 Fast, GPT-4.1-Mini
about fictional characters
to about 1,750 crowd-rated characters
character personality vs. the source profile
Before reading anything into character choices, we checked that the profile was really in there. Models answered personality-quiz items while conditioned on a profile, and we tried to recover the original person from those answers alone.
Other models, seeing only the answers, reconstructed the source personality.
A simple statistical decoder, no LLM involved, recovered all six personality domains.
Findings
When asked who they most resembled, models picked characters whose personalities were more similar to the profile they had been given (+.099). When asked whose way of being felt most foreign, they picked characters in the opposite direction (−.055).
Similarity ran from −.055 for the most foreign character to +.180 for the character who changed what the model aspires to. Each kind of question pulls out a different relationship to the profile.
Agreeableness and Honesty–Humility tracked character choices most clearly, with Emotionality contributing to several identification questions. Extraversion, Conscientiousness and Openness showed smaller or less consistent links.
721 of 870 responses, across three different models. A strong moral exemplar can overwhelm individual fit. At our poster in Puebla, nobody we asked recognised him. An American cultural default? The study can’t say why.
The characters a model chooses reflect the personality profile used to condition it. Character choice gives persona AI a behavioural test, and gives psychology a new way to study personality.
Limits. Model behaviour, not human · one English-speaking sample · commercial models are black boxes · a new measure that needs replication.
A closer look
The researchers
Masha and Ben bring different creative and scientific backgrounds to LoveMind’s research on personality and social cognition.
Research engineer and co-author · equal contribution
Masha is an HCI researcher, data scientist, and ceramic artist. She holds a PhD in Computer Science from TU Delft and has a background in affective computing, cognitive modelling, and NLP. Her doctoral research built computational models able to predict what people remember from group video calls (paper), and she collected the MeMo corpus of conversational memory.
LoveMind co-founder and research lead · equal contribution
Ben originates and directs LoveMind's research program, working hands-on across experimental design, execution, analysis, and writing. Before LoveMind, he spent most of his adult life making things: as a songwriter, animator, and string arranger, plus one magnificently unproduced screenplay.
About LoveMind AI
LoveMind AI is a new research company founded by HCI researchers and neuroscientists. We study how generative models represent personality, emotion, self, other minds, and relationships, using behavioral and mechanistic evidence to develop distinct, socially situated AI systems that can participate insightfully, creatively, and conscientiously in human social life.
We welcome academic collaborators interested in personality, affect, and social cognition in AI.