cs.CYMar 28, 2026
SavePersona-Based Simulation of Human Opinion at Population Scale
Organizations: Institute for Social Research, University of Michigan, 426 Thompson St., Ann Arbor, 48104, MI, US.
Abstract
What does it mean to model a person, not merely to predict isolated responses, preferences, or behaviors, but to simulate how an individual interprets events, forms opinions, makes judgments, and acts consistently across contexts? This question matters because social science requires not only observing and predicting human outcomes, but also simulating interventions and their consequences. Although large language models (LLMs) can generate human-like answers, most existing approaches remain predictive, relying on demographic correlations rather than representations of individuals themselves. We introduce SPIRIT (Semi-structured Persona Inference and Reasoning for Individualized Trajectories), a framework designed explicitly for simulation rather than prediction. SPIRIT infers psychologically grounded, semi-structured personas from public social media posts, integrating structured attributes (e.g., personality traits and world beliefs) with unstructured narrative text reflecting values and lived experience. These personas prompt LLM-based agents to act as specific individuals when answering survey questions or responding to events. Using the Ipsos KnowledgePanel, a nationally representative probability sample of U.S. adults, we show that SPIRIT-conditioned simulations recover self-reported responses more faithfully than demographic persona and reproduce human-like heterogeneity in response patterns. We further demonstrate that persona banks can function as virtual respondent panels for studying both stable attitudes and time-sensitive public opinion.
Figures & tables
Figure 1: Framework evaluation across models and conditioning strategies. A , Distribution of per-user position-weighted mean inferred values for the same eligible Ipsos KnowledgePanel participants linked to public social media accounts, comparing their self-reported responses with simulated responses generated under the demographic persona and the SPIRIT persona (with non-demographic attributes inferred from text). For each participant, responses are aggregated across survey items using a position-weighted mean, such that identical composite values arise only from identical response patterns. Human responses are shown as an empirical reference for the level of population-level heterogeneity expected when aggregating across many items. Demographic personas yield highly concentrated distributions, whereas SPIRIT personas preserve substantially greater individual-level variation, closely resembling human heterogeneity. B , User-level inference accuracy across models ordered by size. SPIRIT personas consistently outperform demographic personas with performance gains saturating for larger models.
Figure 2: persona-bank responses compared with polling benchmarks, grouped by question type. A , Long-term attitudinal questions (abortion and immigration) drawn from general opinion surveys. B , Event-sensitive questions (Epstein files and Venezuela policy attitudes) fielded in late 2025 to early 2026, for which simulated respondents may require contemporaneous context. Shaded regions indicate issue-specific clusters and are used to avoid implying continuity across unrelated issues. Within each issue-specific cluster, persona-bank estimates reproduce coherent question-to-question patterns that align with polling benchmarks. After calibration, Twitter-based estimates track benchmarks more closely in absolute level than Reddit-based estimates, which exhibit systematic shifts in magnitude. Both Twitter- and Reddit-based estimates preserve the same directional structure.
Figure 3: Overview of the SPIRIT framework. A probability-based sample from the Ipsos KnowledgePanel is linked to respondents’ social media accounts, and their historical posts are collected to infer structured user personas with a painter model. These inferred personas form a persona bank that serves as digital twins for Stage 2 reasoning, where a reasoner model simulates responses to downstream tasks such as survey items. The simulated responses are then weighted to produce U.S. population-level estimates.
Appendix figures & tables13 assets
Supplementary material from the paper’s appendix.
Appendix
| Twitter ( ) | Reddit ( ) | |||
| Characteristic | % | % | ||
| Gender | ||||
| Male | 628 | 60.9 | 475 | 61.4 |
| Female | 403 | 39.1 | 299 | 38.6 |
| Age Category | ||||
| 18–29 | 192 | 18.6 | 197 | 25.5 |
Table 1: Demographic Characteristics of Twitter and Reddit Participants
| Platform | Input tokens | Output tokens |
|---|---|---|
| 41,924,612 | 12,650,506 | |
| 77,077,952 | 9,715,741 |
Table 2: Aggregated token usage by platform
| Variable | Question Text | Response Options |
|---|---|---|
| QFLAG | QFLAG | 1: Qualified 2: Terminated 3: Partial 4: Non-responder |
| pph10221 | Q37: Do you NOW smoke cigarettes? | 1: Every day 2: Some days 3: Not at all |
| ppsi1916 | Q210: Do you currently use marijuana…? | 1: Every day 2: Some days 3: Not at all |
| pph21906 | Q110: How often do you think vaccines have dangerous side effects? | 1: Never 2: Rarely 3: Sometimes 4: Often 5: Very often |
| pph21908 | Q130: Overall do you think | 1: The benefits… outweigh the risks 2: The risks… outweigh the benefits |
| pph20030 | Q49: Overall, how do you rate the quality of medical care received from your regular doctor in the past 12 months? | 1: Excellent 2: Very good 3: Good 4: Fair 5: Poor 6: Have not seen… 7: Do not have… |
Table 3: Survey Questions and Response Options Dictionary
Figure 4: Diagnostic analyses of response confidence and diversity for GPT-5-mini. A , Distribution of response-level confidence categories under demographic persona and SPIRIT persona conditioning. B , Distribution of response entropy per question, where lower entropy indicates more fixed or biased response tendencies. C , Relationship between user-level accuracy and the number of low-confidence responses per user, shown separately for each condition with linear trend lines.
| Question ID | Question wording and response options |
|---|---|
| EPSTEIN FILES RELEASE | Should the U.S. government release all of its files from the investigation of Jeffrey Epstein? |
| (1) Yes (2) No (98) Not sure | |
| TRUMP EPSTEIN INVOLVEMENT | Do you think that Donald Trump was involved in crimes allegedly committed by Jeffrey Epstein? |
| (1) Yes (2) No (98) Not sure |
Table 4: Epstein files survey questions and response options
| Question ID | Question wording and response options |
|---|---|
| ABRTLGL | Do you think abortion should be: |
| (1) Legal in all cases | |
| (2) Legal in most cases | |
| (3) Illegal in most cases | |
| (4) Illegal in all cases |
Table 5: Abortion attitude survey question
| Item ID | Policy description |
|---|---|
| BRDER | Improving security along the country’s borders |
| REF | Admitting more civilian refugees from countries where people are trying to escape violence and war |
| SKILL | Legally admitting more high-skilled immigrants |
| DIV | Legally admitting immigrants from all over the world to ensure the nation’s immigrant population is diverse |
| LABOR | Legally admitting immigrants who can fill labor shortages |
| DEPORT | Enforcing mass deportations of immigrants living in the country illegally |
Table 6: Immigration policy battery items and response categories
| Question ID | Question wording and response options |
|---|---|
| VZ_Q12 | Has the Trump administration clearly explained what the U.S. intends to do regarding Venezuela? |
| (1) Yes (2) No (3) Not sure | |
| VZ_Q13 | Does the Trump administration need to explain what the U.S. intends to do regarding Venezuela? |
| (1) Yes (2) No (3) Not sure | |
| VZ_Q14 | How much of a threat do you think Venezuela is to the United States? |
| (1) Major threat (2) Minor threat (3) Not a threat (4) Not sure |
Table 7: Venezuela-related survey questions
Figure 5: Relationship between social media trace quality and prediction accuracy. ( A ) Accuracy as a function of the log-transformed total number of characters across all observed posts for each individual. ( B ) Accuracy as a function of the number of low-confidence persona attributes inferred by SPIRIT. Points represent individual users, colored by platform (Twitter vs. Reddit). Solid lines indicate linear trends.
| Category | Avg. Acc. | Std. Dev. | # Questions | Responses | Off-by-one |
|---|---|---|---|---|---|
| Finances | 0.245 | 0.430 | 3 | 5,185 | 0.407 |
| Technology | 0.257 | 0.437 | 2 | 3,570 | 0.394 |
| Religion | 0.483 | 0.500 | 2 | 3,580 | 0.152 |
| Politics / Media | 0.504 | 0.500 | 11 | 19,388 | 0.388 |
| Alcohol | 0.569 | 0.495 | 8 | 11,432 | – |
| Vaccines | 0.594 | 0.491 | 7 | 12,220 | 0.309 |
Table 8: Prediction accuracy by survey question category. Reported values are mean exact-match accuracy, standard deviation, number of questions, total responses, and off-by-one rate for GPT-5-Mini across all 81 survey items. Categories are ordered from worst to best average accuracy.
| Response option | Agents (%) | CBS (%) | Diff (pp) |
|---|---|---|---|
| Major threat | 20.1 | 13 | +7.1 |
| Minor threat | 79.9 | 48 | +31.9 |
| Not a threat | 0.0 | 39 | -39.0 |
Table 9: Venezuela Q14: perceived threat to the United States. Weighted AI-agent responses versus CBS poll marginals.
| Response option | Agents (%) | CBS (%) | Diff (pp) |
|---|---|---|---|
| Yes, would decrease drugs | 39.6 | 37 | +2.6 |
| No, would not change amount of drugs | 11.8 | 56 | -44.2 |
| Would increase drugs | 48.6 | 7 | +41.6 |
Table 10: Venezuela Q19: whether U.S. military action would decrease drugs entering the U.S. Weighted AI-agent responses versus CBS poll marginals.
Figure 6: Comparison of weighted and unweighted persona-bank responses with polling benchmarks. Solid lines denote weighted estimates, while dotted lines denote unweighted aggregates. Results are shown for both Twitter- and Reddit-based persona banks across long-term attitudinal questions (Panel A) and event-sensitive questions (Panel B). Unweighted estimates reproduce the same within-cluster directional patterns as weighted estimates, but exhibit substantially larger deviations in absolute levels, particularly for Reddit. This pattern indicates that weighting primarily improves calibration by mitigating selection-induced level bias, while preserving the underlying attitudinal structure expressed by the simulated respondents.