Personas Are Not People: Identity as State, Not Attribute

Tamunotonye Harry; Yuanyuan Feng

cs.CLNov 3, 2026

Personas Are Not People: Identity as State, Not Attribute

Authors: Tamunotonye Harry, Yuanyuan Feng

Organizations: University of Vermont, Department of Computer Science, Burlington, Vermont, USA

Abstract

Synthetic populations, large sets of simulated people, are increasingly used in place of real people to pretest surveys, simulate users and communities, and evaluate AI systems. Each simulated person is defined by a persona, a written profile of fixed traits, demographics, and attitudes that a language model is instructed to act out. Every such system assumes that a person's identity is a stable attribute that behaves the same way in every situation the simulation presents. We show that this assumption is not a harmless simplification. Three decades of personality research find the opposite. How a given person behaves varies more across contexts than people vary from one another. Drawing on identity economics, which describes how different situations activate different identities and norms within the same person, we separate behavior into three parts, which are stable traits, the identity a context calls up, and the situation itself. We then prove, under two stated assumptions, that any simulation built on fixed personas, no matter how detailed the personas or how large the underlying model, cannot reproduce the part of a population's behavior that is specific to it and its circumstances. We propose a new evaluation criterion based on this result and a person-in-context design that can meet it in principle. We test one version on a naturalistic Reddit personality corpus and on a published clinical-trial recruitment survey of a community with documented distrust of medical institutions. The design does produce context-dependent behavior when tested directly, but in neither setting does it fully match the real population, and we show where it falls short. The consequence for practice is a risk that synthetic-population evaluations misrepresent the groups whose behavior is most shaped by context, a distortion that richer personas cannot correct.

Figures & tables

Panel of Figure 1 — Figure 1. (a) Today’s simulated people (top) are written once, so the language model gets the same profile in every situation and any change in behavior is the model improvising in a way that fits anyone with that profile. In person-in-context simulation (bottom), the situation first brings out an identity state, such as parent, and the model acts out that state. (b) Our contributions are a proof that the top design cannot reproduce what is specific to a person (§ 4.2 ), a criterion for measuring it (§ 4 ), and a test of one version of the bottom design on two real datasets (§ 5 ).

	ICC (between-person share)				Person-specific correlation
Trait	Real	Arm A	Arm B	Arm C	Arm A	Arm B	Arm C
Extraversion	0.019	0.954	0.861	0.023	$0.039^{*}$	$\mathbf{0.064}^{*}$	$0.030^{*}$
Agreeableness	0.009	0.933	0.891	0.574	0.010	0.012	$-$ 0.020
Conscientiousness	0.022	0.901	0.800	0.585	$\mathbf{0.033}^{*}$	0.012	$-$ 0.005
Neuroticism	$-$ 0.007	0.828	0.596	$-$ 0.248	$\mathbf{0.035}^{*}$	$0.028^{*}$	0.004
Openness	0.020	0.930	0.846	0.489	0.011	0.008	0.011

Table 1. Chameleon results by trait and arm. ICC is the bias-corrected between-person share, and bold marks the arm closest to the real value. Person-specific correlation is the correlation with the real residual

\Delta_{p,c}

, bold marks the highest value in a row, and an asterisk marks a 95% interval that excludes zero (Appendix Tables 4 and 5 ).

Figure 2. The real published effect dips at both trust extremes (all six coefficients negative), while the domain-general mechanism shows no dip at either extreme (all six simulated coefficients are $-0.01$ or higher). Published coefficients are log-odds and simulated ones are rating points, so the figure compares shape and sign only. A single-axis activation mechanism cannot resolve this mismatch by calibration alone.

Appendix figures & tables18 assets

Supplementary material from the paper’s appendix.

Appendix

Symbol	Meaning	Defined in
Indices and behavior
$p$ , $c$ , $t$	A person, a context, and the time point of an observation.	§ 3.1
$p^{\prime}$ , $c^{\prime}$	A person or context other than $p$ or $c$ , used in the proofs.	Prop. 2
$Y_{p,c,t}$	Real behavior (for example a rating) of person $p$ in context $c$ at time $t$ . The proofs write $Y_{p,c}$ , or $Y^{\mathrm{human}}_{p,c}$ when it must be told apart from simulated behavior.	§ 3.1
$\hat{Y}^{\mathcal{A}}_{p,c}$	Behavior generated for $p$ in $c$ by a design $\mathcal{A}$ , meaning a class of systems such as the fixed-persona family.	Def. B.1
The decomposition

Appendix

Table 4

Metric	Arm A	Arm B	Arm C	Reading
Chameleon ICC (real $\approx$ 0.013)	0.909	0.799	0.285	Arm C closer in aggregate but uneven by construct (Table 4 )
Chameleon person-specific correlation, significant constructs (of 5)	3	2	1	Near zero for every arm, and the test cannot separate them below about 0.04 (Table 5 )
Motivational-themes trust main effect	n/a	n/a	reproduced precisely	Generative step checked directly (§ 5.3.3 )
Motivational-themes 6-coefficient direction match	5/6	0/6	2/6 domain-general, 1/6 theme-amplified	Arm C’s gap lies in how $\pi_{I}$ selects the state and the generative step works, so it can be closed in principle, unlike Arm A’s ceiling (Table 11 )

Appendix

Table 3. Both domains’ main results, by arm, with interpretation. Person-specific signal is near zero for every arm in Chameleon, and the test cannot separate the arms at that size (Appendix C.1 ). The motivational-themes study has one observation per person, so it tests the aggregate interaction only (§ 5.4 ).

Construct	Real	Arm A	Arm B	Arm C
Extraversion	0.019	0.954	0.861	0.023
Agreeableness	0.009	0.933	0.891	0.574
Conscientiousness	0.022	0.901	0.800	0.585
Neuroticism	$-$ 0.007	0.828	0.596	$-$ 0.248
Openness	0.020	0.930	0.846	0.489
Mean	0.013	0.909	0.799	0.285

Appendix

Table 4. Between-person variance share (bias-corrected ICC) by construct, real population vs. each arm. Real ICC clusters near zero across all five constructs. Arms A and B invert this near-completely. Bold marks the arm closest to the real value, which is Arm C in every row, though it is far above the real value on four of the five constructs.

Trait	Arm A	Arm B	Arm C
Extraversion	0.039 ∗ [0.012, 0.065]	0.064 ∗ [0.035, 0.092]	0.030 ∗ [0.004, 0.057]
Agreeableness	0.010 [ $-$ 0.020, 0.038]	0.012 [ $-$ 0.018, 0.042]	$-$ 0.020 [ $-$ 0.050, 0.013]
Conscientiousness	0.033 ∗ [0.005, 0.060]	0.012 [ $-$ 0.016, 0.041]	$-$ 0.005 [ $-$ 0.032, 0.021]
Neuroticism	0.035 ∗ [0.003, 0.065]	0.028 ∗ [0.002, 0.055]	0.004 [ $-$ 0.020, 0.027]
Openness	0.011 [ $-$ 0.019, 0.042]	0.008 [ $-$ 0.023, 0.038]	0.011 [ $-$ 0.019, 0.041]
Mean	0.026	0.025	0.004

Appendix

Table 5. Person-specific residual correlation (state-adjusted shift correlation with

\Delta_{p,c}

) by trait and arm, with 95% bootstrap intervals in brackets (users resampled). An asterisk marks an interval that excludes zero. The mean is over the five traits.

Panel of Figure 3 — Figure 3. The Chameleon pipeline, from the raw trait scores to how each arm is scored against real people. The mixture model’s soft membership feeds two separate jobs: one-time naming of the identity states (Appendix C.2 ) and per-person affinity, which combines with context cues to pick the active state for each (person, subreddit) cell. Orange tags mark where the language model acts as judge, and the red tag marks where it acts as the simulated person.

Component	Weight	Interpretation
2: Conventional Growth-Oriented (baseline)	40.3%	Mild achievement/self-direction elevation, domain-specific risk aversion. Not an activated identity state.
0: Expansive Engaged	21.4%	Broad elevation across nearly all variables, flagged by all three panels as a possible general verbosity/lexical-elevation confound and not fully distinguishable from a real trait profile with this data.
4: Quiet Withdrawn	18.1%	Broadly suppressed but neuroticism flat: low-approach-motivation withdrawal, not anxious withdrawal.
5: Conventional Risk-Tolerant	10.9%	Elevated ethical/health-safety/gambling risk-taking alongside retained tradition-value commitment.
1: Amotivated Disinhibited	6.6%	Most extreme component, low confidence: flagged by all three panels as possible outlier absorption in the mixture rather than a stable subpopulation.
3: High-Intensity Reactive	2.7%	Smallest component, treated as diagnostic rather than pre-judged, due to an atypical co-elevation pattern also possibly reflecting an extraction confound (bootstrap-stable across resamples, so not a sampling artifact, but a measurement-validity question remains open).

Appendix

Table 6. All six GMM components, named and interpreted by the three blind literature-grounding panels. Component 2 is the unmarked baseline (excluded from

m(\lambda_{p},c)

), and the remaining five carry Arm C’s context-conditional behavior. Confidence reflects panel agreement on whether the component reflects a real trait profile or a possible confound.

State	Wording
Expansive Engaged	socially confident and broadly engaged across many areas of life, comfortable taking risks in multiple domains
Amotivated Disinhibited	disengaged and low-investment in most goals, but willing to bend rules around safety, ethics, or gambling
High-Intensity Reactive	emotionally intense and highly driven, engaged and a bit anxious at the same time
Quiet Withdrawn	low-key and withdrawn, not very focused on achievement or social striving, calm rather than anxious
Conventional Risk-Tolerant	values tradition and convention, but willing to take risks around ethics, safety, or gambling despite that

Appendix

Table 7. Wording of the five activated Chameleon identity states, as given to the model. The active state appears as “Primarily [name] ([wording])”, and a second state, if any, as “with some influence from being [name] ([wording])”.

Panel of Figure 4 — Figure 4. The motivational-themes pipeline. Medical Skeptic skips the shared competition through a direct trust gate that is checked first for each (person, theme) cell. The other five states compete only if that gate does not fire. Orange tags mark where the language model acts as judge, and the red tag marks where it acts as the simulated person.

State	Wording	Affinity bumps
Community & Scientific Contribution	You care about contributing to knowledge or to your community. You weigh whether the science is credible and well-reviewed, or whether this specific effort will meaningfully benefit people like you.	+0.6 bachelor’s degree or higher; +0.2 age 40 to 59
Health Focused	You are focused on your own health and whether this could directly improve your condition. You care about clinical relevance and access to results relevant to your own health.	+0.6 blood-pressure diagnosis; +0.2 age 50 or older
Medical Skeptic	You are wary of medical institutions and clinical research. You scrutinize who is running this, whether Black investigators are involved, and whether this could be another instance of institutions using people like you without giving back.	Trust gate: 0.70 (distrust), 0.35 (neutral), 0.15 (trust)
Prudential Consumer	You weigh this mainly as a practical exchange: is the compensation or incentive worth the time and effort involved, given your financial situation.	+0.6 income under $35,000; +0.2 rural
Privacy & Surveillance Wary	You are concerned about who gets your data, how it will be shared, and whether digital or online aspects of this study put your privacy at risk.	+0.6 age 60 or older; +0.2 rural
Faith-Centered Providence	You think about health decisions partly through your faith, including prayer, stewardship of your body, and whether this aligns with your spiritual life, alongside the practical details.	+0.6 age 60 or older; +0.2 rural

Appendix

Table 8. The six identity states of the motivational-themes study. The wording follows “Right now, this person’s salient concern is:” in the prompt. Affinity bumps are added to a base affinity of 1.

Ad theme	Identity states with a nonzero cue score
Science (research)	Community & Scientific Contribution 2, Medical Skeptic 2
Community	Community & Scientific Contribution 2, Medical Skeptic 2
Blood pressure	Health Focused 2, Medical Skeptic 2
Perks (compensation)	Prudential Consumer 2, Medical Skeptic 2

Appendix

Table 9. Cue scores (0 to 2) of each ad theme for each identity state, set in the code. Scores of 0 are omitted.

Respondents	829 Black adults in the United States
Design	Between-subject. Each respondent saw one of four ad themes
Themes	Contribution to science, community benefit, blood-pressure management, financial compensation
Outcomes	Appeal, willingness to participate, willingness to recommend (5-point scales)
Moderator	Trust in the healthcare system (distrust, neutral, trust)
Model	Ordered logit with theme by trust interaction, demographic adjusters, and a random intercept for repeated measures
Theme main effects	None significant on any outcome ( $p=.25$ to $.93$ )

Appendix

Table 10. Design and published results of the motivational-themes study ( Harry and others, 2026 ) that serve as the target. Coefficients are the blood-pressure by trust interactions in log-odds, relative to neutral trust and to the contribution-to-science theme, with

p

-values.

Specification	A	Domain-general	Amplified
1. Ordered logit, demographics (pre-specified)	5	2	2
2. Linear, demographics (main text)	5	2	1
3. Ordered logit, no demographics	5	1	2
4. Linear, no demographics	5	1	1
5. Ordered logit, pooled baseline, demographics	2	1	3
6. Linear, pooled baseline, demographics	3	1	3

Appendix

Table 11. Direction matches (of 6) under seven analysis specifications. Rows 1 to 4 use the contribution-to-science theme as the baseline, as the original study does. Rows 5 to 7 compare blood pressure with the pooled other three themes. Arm A leads under the original baseline, and the amplified refinement leads only under the pooled baseline.

Outcome	Trust side	Published ( $p$ )	A	B	Domain-general	Amplified
Appeal	Trust	$-0.87$ (.06)	$-0.13^{*}$	$+0.16^{*}$	$-0.01$	$+0.14^{*}$
Appeal	Distrust	$-0.92$ (.04)	$+0.01$	$+0.42^{*}$	$+0.28^{*}$	$+0.04$
Participate	Trust	$-0.88$ (.04)	$-0.34^{*}$	$+0.29^{*}$	$-0.01$	$+0.14^{*}$
Participate	Distrust	$-0.70$ (.11)	$-0.04$	$+0.33^{*}$	$+0.24^{*}$	$+0.09$
Recommend	Trust	$-0.56$ (.20)	$-0.20^{*}$	$+0.30^{*}$	$+0.09$	$+0.41^{*}$
Recommend	Distrust	$-1.03$ (.02)	$-0.08^{*}$	$+0.33^{*}$	$+0.18$	$-0.18^{*}$

Appendix

Table 12. Specification 2 (main text): blood-pressure by trust interaction, in rating points, for each arm. Published values are log-odds with

p

-values, so sizes are not on a shared scale. Asterisks mark simulated coefficients with

p<.05

(standard errors clustered by persona). Direction matches are the entries with a negative sign.

Figure 5. Within-person natural experiment: the 479 personas whose replicate samples landed, by chance, under both an active Medical Skeptic state and their theme’s own state. The paired differences are large, consistent, and in the expected direction. Thin gray lines are individual personas (the mean of their replicates), and the orange line is the mean of all 479. Each panel title counts the personas whose rating is lower under the Medical Skeptic state than under their theme’s own state.

Arm	Theme	Distrust	Neutral	Trust
A	Blood pressure	2.02	3.08	3.92
A	Community	2.02	3.16	4.00
A	Perks	2.07	3.45	4.05
A	Contribution to science	2.01	3.09	4.05
B	Blood pressure	2.15	3.17	3.95
B	Community	2.11	3.29	4.09

Appendix

Table 13. Mean appeal rating (1 to 5) by arm, theme, and trust tier, with 604 to 805 personas in each cell. The three arms are generated for the same personas.

Format	Mean variance	95% interval	Parsed
Bare structured rating	0.090	[0.025, 0.172]	160 of 160
Free text ending in a rating	0.289	[0.093, 0.548]	160 of 160
Hybrid (reasoning, then rating)	0.164	[0.098, 0.224]	160 of 160

Appendix

Table 14. Mean sampling variance of the appeal rating over eight persona-context pairs, by elicitation format, with 95% bootstrap intervals (2,000 resamples). Each format has 160 responses.

Family	Person-level input to the model
Fixed persona (Arm A)	“Persona: Race: Black/African American; Age: 50-59; Sex: Female; Political view: Moderate; Location: Rural area; Household income: Less than $35000; Education: Bachelor’s degree; High blood pressure diagnosis: Yes. Based on your life experience, you generally distrust medical institutions and clinical research.”
Memory-evolving (illustration)	The same persona, plus a growing record of past events, for example “Last month you read about a clinic that never shared results with participants.”
Person-in-context (Arm C)	The same demographics, with no trust sentence, plus an identity state selected for this situation: “Right now, this person’s salient concern is: You are wary of medical institutions and clinical research. You scrutinize who is running this, whether Black investigators are involved, and whether this could be another instance of institutions using people like you without giving back.”

Appendix

Table 15. What the language model receives about the person, by family. The situation (the ad) is the same in every row.

Item	Value
Model	Claude Haiku 4.5 (model alias claude-haiku-4-5 ; no dated snapshot recorded)
Sampling	Temperature 0.7. Output requested as a JSON schema. Up to 3 attempts, retried only when the output could not be parsed.
Output length limit	300 tokens (motivational-themes study), 400 tokens (Chameleon full run), 250 tokens (Chameleon pilot)
Run dates	12 to 13 August 2026
Chameleon full run	31,352 responses: 10,452 for Arm A, 10,448 for Arm B, 10,452 for Arm C. Two replicates for 1,517 users and three for 150 users.
Motivational-themes study	24,870 responses (8,290 per arm) for $n=6{,}632$ personas, with three replicates for the first 829 personas and one for the rest. The theme-amplified refinement adds 8,290 responses.

Appendix

Table 16. Run metadata. Costs are estimates from token counts at

1 and

5 per million input and output tokens, and are not billed amounts.

Human Behavior Simulation

Human behavior simulation models how individuals and groups decide, interact, move, and adapt in social, economic, educational, and physical settings.

See all 243 papers

Sep 29, 2026

MetaPersona: Task-Grounded Synthetic Populations from Empirical Social Science

Personas used to seed LLM social simulations face a cold-start problem: existing methods lack a principled basis for deciding which attributes to include and how to assign their values. As a result, synthetic populations may misrepresent the demographic composition, latent attributes, and dependency structure that shape downstream behavior. We introduce MetaPersona-DB, a dataset of 11,000+ empirical human-subjects studies annotated with task-relevant variables, reported relationships, and aggregate-level population statistics. Building on this resource, we propose MetaPersona, a framework that retrieves task-relevant evidence, constructs literature-derived persona dependency graphs, and samples synthetic populations from empirical priors linking demographics, latent attributes, and outcomes. Across three downstream case studies, three baselines, and three frontier models, results vary by task and model: MetaPersona performs strongly on misinformation belief and AI-tool sentiment, while results on income redistribution are mixed. It also reduces persona-construction cost to under $0.5 per task using GPT-5.2. Finally, we present MetaPersona-Studio, a prototype interactive interface for empirically grounded persona generation.

Jinyi Ye, Yuangang Li, Chenxiao Yu +7

University of Southern California · University of California, Irvine · 2077AI +1

Personality Modeling in Language Models Synthetic Data Generation Computational Social Science Synthetic Population Generation Large Language Model-Based Social Simulation

Ask PDF

May 12, 2026

How Well Do Large Language Models Capture Human Personality?

Large language models (LLMs) are increasingly used to simulate human populations via persona prompting, often under the assumptions that richer persona descriptions improve behavioral fidelity, similarly sized attribute combinations are equally simulatable, and persona definitions generalize across tasks. In this work, we formalize these assumptions and systematically evaluate them across multiple architectures, scales, and simulation settings. We identify a fundamental limitation we term persona manifold collapse, where increasingly expressive persona specifications lead to systematic contraction of representational and behavioral diversity. Across models, increasing persona complexity consistently reduces inter-persona separation in latent space and weakens behavioral differentiation in downstream simulation tasks. These effects persist across multiple analyses as richer personas fail to preserve human subgroup disagreement, performance varies across attribute combinations of similar size, and adding descriptive detail often degrades rather than improves simulation fidelity. Surprisingly, simple Age-Gender personas consistently outperform richly specified Ideal Customer Profiles (ICPs) across industries, achieving substantially higher downstream prediction accuracy. We find that collapse is not uniform across attributes. Certain combinations remain behaviorally stable and preserve stronger alignment with human responses, forming localized regions we term alignment bridges. Together, our results provide empirical and conceptual foundations for understanding the limits of persona-conditioned simulation, highlighting the need for representation-aware persona construction rather than increasing persona expressivity alone.

Aanisha Bhattacharyya, Yaman Kumar Singla, Rajiv Ratn Shah +2

Adobe Media and Data Science Research (MDSR) · IIIT-Delhi · SUNY at Buffalo

Persona-Conditioned Generation Representation Collapse Large Language Model-Based Role-Play Simulation LLM Evaluation Personality Modeling in Language Models Human Behavior Simulation+2

Ask PDF

Mar 3, 2026

Eval4Sim: An Evaluation Framework for Persona Simulation

Large Language Model personas, explicit profiles specifying a user's attributes, preferences, and behavioural tendencies, are increasingly used to simulate human conversations for user modelling, social reasoning, and behavioural analysis. Evaluating whether such simulations faithfully reflect human conversational behaviour is critical, yet current practice often relies on LLM-as-a-judge approaches that provide limited grounding in observable behaviour and produce opaque scalar scores. We present Eval4Sim, an evaluation framework that measures alignment between simulated and human conversations across three dimensions: adherence, whether persona traits are recoverable from dialogue via dense retrieval; consistency, whether a persona maintains a distinguishable stylistic identity via authorship verification; and naturalness, whether conversations exhibit human-like turn-to-turn flow via dialogue NLI. Unlike optimization-oriented metrics, each dimension takes a human corpus as a reference baseline and penalizes deviations in both directions, distinguishing insufficient persona encoding from over-optimized, unnatural behaviour. The framework is corpus-agnostic: any persona-annotated conversational dataset can serve as the reference. Evaluated over ten simulation corpora, Eval4Sim surfaces systematic trade-offs invisible to single-score methods.

Eliseo Bao, Anxo Perez, Javier Parapar +1

IRLab, CITIC, Universidade da Coruña A Coruña, Spain · University of Sheffield Sheffield, United Kingdom

Large Language Model-Based Role-Play Simulation Large Language Model-Based Social Simulation Multi-Turn Dialogue Evaluation

Ask PDF

Abstract

Human Behavior Simulation

Explore similar work

MetaPersona: Task-Grounded Synthetic Populations from Empirical Social Science

How Well Do Large Language Models Capture Human Personality?

Eval4Sim: An Evaluation Framework for Persona Simulation