Pith. sign in

REVIEW 6 major objections 5 minor 1 cited by

Can LLMs Infer Personality from Real World Conversations?

T0 review · 6 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Current LLMs cannot reliably infer validated Big Five personality traits from naturalistic interviews, despite stable outputs.

desk verdict A credible negative result on LLM personality inference from real interviews, held back by missing psychometric details that need to be nailed down before the strong claim is accepted. read the letter →

arxiv 2507.14355 v1 pith:HJ4ISBB4 submitted 2025-07-18 cs.CL

Jianfeng Zhu , Ruoming Jin , Karin G. Coifman This is my paper · ORCID
classification cs.CL
keywords largelanguagemodelspersonalityassessmentBigFiveInventoryBFI-10constructvalidityzero-shotpromptingchain-of-thoughtpsychometrics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tests whether off-the-shelf large language models (LLMs) can infer a person's Big Five personality traits from natural conversation. It introduces a benchmark of 555 semi-structured interviews about emotional and social experiences, each paired with BFI-10 self-report scores collected in the same session, and evaluates three LLMs on item-level and trait-level prediction. All models produced highly stable outputs across repeated runs, but their scores barely matched the self-reports: the strongest correlation was $r = 0.27$, chance-adjusted agreement was near zero, and predictions clustered toward moderate or high trait levels. Chain-of-thought prompting and longer transcripts improved distributional alignment without improving trait-level accuracy. A sympathetic reader should care because this is direct evidence that current LLMs are not psychometrically valid tools for scalable personality assessment, despite their fluency.

What carries the argument

The load-bearing machinery is the evaluation protocol rather than any single model. The benchmark pairs 555 semi-structured interview transcripts, autobiographical reflections on recent emotional and social experiences, with same-session BFI-10 self-report scores, the 10-item Big Five Inventory scored on a 1–5 Likert scale. Three LLMs (GPT-4.1 Mini, Meta-LLaMA, DeepSeek) are prompted zero-shot for BFI-10 item scores and with zero-shot and chain-of-thought prompts for Big Five trait scores. Outcomes are measured psychometrically: Pearson correlation for criterion validity, MAE and RMSE for error, discretized Low/Moderate/High bins with exact-match, off-by-one, and Cohen's $\kappa$ for categorical agreement, and intraclass correlation coefficients across repeated runs for test-retest reliability. The protocol's design lets reliability and validity be separated, which is what makes the dissociation visible.

What would settle it

A preregistered replication using the same transcripts but an observer-rated or informant-rated Big Five assessment as the criterion would settle the claim: if model predictions correlate above roughly $r = 0.5$ with observer ratings while remaining near zero against self-report, the paper's weak-validity conclusion would have to be revised as a criterion problem rather than a model failure.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is a dissociation: LLMs produce highly stable personality scores from naturalistic interview transcripts, yet those scores are nearly uncorrelated with the BFI-10 self-reports collected from the same participants in the same session. The best item-level correlation was $r = 0.27$ (GPT-4.1 Mini on “tends to be lazy”), the best trait-level correlation was $r = 0.25$ (conscientiousness), and Cohen's $\kappa$ was below 0.10 across conditions, meaning agreement beyond chance was minimal. Predictions were systematically biased toward moderate or high trait levels, and neither chain-of-thought prompting nor longer input context improved trait-level accuracy. The paper concludes that current LLMs lack the construct validity needed for real-world personality assessment.

Load-bearing premise

The conclusion depends on treating the same-session BFI-10 self-report scores as a valid and reliable gold standard for personality; if those self-reports are noisy, biased by mood or self-presentation, or otherwise unrepresentative of the person's actual traits, the weak correlations and near-zero $\kappa$ could reflect a flawed criterion rather than a genuine LLM limitation.

Editorial extensions

If this is right

  • Current LLM-based personality scores should not be used in clinical, hiring, or personalized-AI decisions without validity evidence.
  • High test-retest reliability is not enough: models can be consistently wrong, and chance-adjusted agreement can still be near zero.
  • Chain-of-thought prompting is unlikely to close the validity gap for latent traits, since it changed distribution shapes but not trait-level accuracy.
  • Longer interview contexts improve distributional alignment modestly but increase prediction variance, so context length is a trade-off, not a fix.
  • Small average errors do not mean good inference; DeepSeek's low MAE and RMSE coexisted with near-zero or negative correlations.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the combination of stable outputs and weak criterion alignment suggests the models are measuring a stable property of the language itself, such as expressed affect or narrative style, rather than the self-reported Big Five construct.
  • Editorial inference: a direct test of the criterion-bias explanation would be to collect a second self-report at a different session and compare model errors against self-report test-retest instability; if errors track instability, part of the validity gap is measurement noise in the gold standard.
  • Editorial inference: the near-zero or negative correlations on sociability and energy items may reflect the adversity-focused interview content; an extension would compare models on interviews with neutral versus trait-relevant topics to separate construct signal from topic confounds.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

6 major / 5 minor

Summary. This paper presents a benchmark for evaluating whether LLMs can infer Big Five personality traits from naturalistic conversation. The dataset consists of 555 semi-structured interviews with BFI-10 self-reports collected in the same session; after preprocessing, 518 participants were analyzed. Three instruction-tuned LLMs (GPT-4.1 Mini, Meta-LLaMA-3.3-70B-Instruct-Turbo, DeepSeek-R1-Distill-70B) were tested with zero-shot and chain-of-thought prompts at BFI-10 item level and Big Five trait level, with additional input-length ablation and test-retest reliability analysis. The main findings are weak correlations with self-reported scores (max r = 0.27), low Cohen's kappa (< 0.10), high ICCs for one model, and a tendency to overpredict moderate-to-high trait levels; the authors conclude that current LLMs show high output stability but limited construct validity.

Significance. If the central claims survive the methodological checks below, this would be a valuable contribution: it provides a real-world interview benchmark rather than synthetic or social-media text, uses a psychometric criterion, and demonstrates that high retest reliability does not imply criterion validity. The multi-model and multi-prompt design is appropriately cautious. However, the paper's key negative result is only interpretable if the BFI-10 trait scores are computed correctly and the prompt conditions are cleanly separated; the manuscript currently does not establish either point. The absence of confidence intervals and a human-rater ceiling also limits the strength of the conclusions, though these are secondary.

major comments (6)
  1. [Dataset and Preprocessing / Evaluation Metrics] The manuscript never specifies how the five BFI-10 trait scores were aggregated from the ten items, and it reports no reliability estimate for the BFI-10 in this pooled sample. Standard BFI-10 scoring requires averaging the two items per trait after reverse-scoring five items (BFI_1, BFI_3, BFI_4, BFI_5, BFI_7); if reverse-scoring was omitted, the trait-level ground truth used in Figure 5 and the correlation/error analyses is invalid. Even with correct scoring, the two-item scales have limited reliability, and without alpha, retest, or informant reliability the observed weak correlations are not corrected for attenuation. Please report the exact scoring formula and sample reliability, and rerun or re-interpret the trait-level analyses accordingly.
  2. [Table 6 and Table 7] The zero-shot Big Five prompt in Table 6 explicitly instructs the model to reason step by step and to justify scores, which is the same chain-of-thought format as the CoT prompt in Table 7. Consequently, the zero-shot versus CoT comparison for Big Five traits in Figures 5-8 does not isolate the effect of chain-of-thought prompting, and the RQ2 conclusion that CoT offers limited benefit is not supported by the reported design. Please either use a direct score-only prompt for the zero-shot condition or relabel the conditions and revise the claims accordingly.
  3. [Table 2] Table 2 lists BFI_5 with the description 'Is full of energy' and interprets it in terms of exuberant expression, whereas the BFI-10 prompt in Table 5 and the Results text identify BFI_5 as 'has few artistic interests.' This discrepancy affects the item-level results and their interpretation: the reader cannot determine which item was actually presented to the models. Please correct the item labels and verify that all item-level analyses use the official BFI-10 wording.
  4. [Reliability Analysis] The test-retest reliability analysis was conducted on a single model (listed as both GPT-4.1 Mini and GPT_4o Mini) using a 50-participant subsample, yet the Results and Discussion state that 'all models' demonstrated high test-retest reliability. This generalization is unsupported; please compute ICCs for the other models or restrict the reliability claim to the model actually tested. In addition, report confidence intervals for the ICCs given the small subsample.
  5. [Evaluation Metrics] The categorical alignment analysis uses different bin boundaries for ground-truth (Low 1-2, Moderate 3, High 4-5) and LLM-predicted (Low 1.0-2.4, Moderate 2.5-3.4, High 3.5-5.0) scores. These asymmetric, hand-chosen thresholds directly determine the exact-match and kappa values in Figures 4, 8, and 12; for example, a predicted score of 2.5 is treated as Moderate while a ground-truth score of 2 is Low. Please justify the cutpoints with a sensitivity analysis or apply identical thresholds to both distributions.
  6. [Results vs. Discussion] The claimed direction of the prediction bias is stated inconsistently. Results (Figures 6-7) say models 'tended to overpredict Moderate and High scores, underestimating Low scores,' while the Discussion says models 'consistently underrepresented moderate trait ratings, disproportionately favoring extreme values.' These are opposite statements about the Moderate bin. Please reconcile the text with the actual binned distributions and with the abstract's 'moderate-to-high' characterization.
minor comments (5)
  1. [Throughout] The model name is inconsistent ('GPT-4.1 Mini' vs. 'GPT_4o Mini'); please use one name and confirm the exact version used in each analysis.
  2. [Tables] Two tables are labeled 'Table 2' (BFI-10 item performance and best-performing models by trait); renumber the tables and update in-text references.
  3. [Statistical reporting] No confidence intervals are reported for the Pearson correlations, kappa statistics, or ICCs; given n=518 and a 50-participant reliability subsample, confidence intervals are needed to assess precision.
  4. [Ablation Analysis] The chi-square tests in the Ablation section are not described (test name, assumptions, effect sizes) and no multiple-comparison correction is mentioned; please specify the test and report effect sizes.
  5. [Data availability] Since the paper introduces a 'novel benchmark,' include a data availability statement; if transcripts cannot be released due to consent, state this explicitly.

Circularity Check

0 steps flagged · score 2.0 of 10

No circularity: the benchmark is externally grounded in BFI-10 self-reports, LLM outputs are zero-shot with no fitted parameters, and the only self-citations are non-load-bearing.

full rationale

This paper is an empirical benchmark rather than a derivation: GPT-4.1 Mini, Meta-LLaMA, and DeepSeek produce scores from interview transcripts using zero-shot or chain-of-thought prompting, and those scores are compared against BFI-10 self-reports collected in the same session. No parameter is fitted to the target outcomes, no equation defines the LLM outputs in terms of the ground-truth scores, and no prediction is constructed from the criterion it claims to predict. The ground-truth BFI-10 scores are external to the model outputs, so the observed weak correlations (max r = 0.27) and low kappa values are free to be low. The only self-citations are Refs [27] and [28], cited in the general introductory sentence about researchers exploring LLMs for psychological assessment; these motivate the topic but do not carry the load-bearing psychometric argument, which rests on the externally validated Rammstedt and John BFI-10 measure and on the authors’ own new interview data. The asymmetric binning of self-report scores (Low 1–2, Moderate 3, High 4–5) versus LLM scores (Low 1.0–2.4, Moderate 2.5–3.4, High 3.5–5.0) is a methodological choice that could affect distributional-alignment claims, but it is not a circular reduction. The limitations passage acknowledges that the BFI-10 is brief and may restrict expression of nuanced traits, but it does not address same-session self-report reliability or the missing BFI-10 scoring formula; those omissions are genuine validity risks to the negative conclusion, because a noisy or mis-scored criterion could understate LLM performance, but they are measurement-validity concerns rather than circularity. The central claim is therefore self-contained against an external benchmark, with no step that reduces by construction to its own inputs.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The study introduces no fitted parameters in the mathematical sense. The hand-chosen binning thresholds and truncation lengths are the main researcher degrees of freedom that affect reported metrics. The central evaluation rests on two domain assumptions: self-reported BFI-10 is a valid criterion, and the scripted interviews provide sufficient real-world linguistic signal.

free parameters (2)
  • LLM continuous-score binning thresholds = Low: 1.0-2.4, Moderate: 2.5-3.4, High: 3.5-5.0
    Hand-chosen thresholds used to discretize continuous LLM predictions for comparison with ground-truth bins (Low: 1-2, Moderate: 3, High: 4-5). This asymmetry may inflate or deflate agreement metrics.
  • Transcript truncation lengths = 100 words, 1000 words, full transcript (mean ~2955 words)
    Hand-chosen input lengths for the ablation study; not fitted but affect reported performance.
assumptions (3)
  • domain assumption BFI-10 self-report scores are valid ground truth for personality
    The paper uses self-report BFI-10 as the criterion for construct validity. Self-report personality measures can be biased by response style and transient mood.
  • domain assumption Semi-structured interview transcripts contain sufficient linguistic signal for personality inference
    The study assumes that personality-relevant cues appear in the text. If transcripts are too short or scripted, the negative results may reflect data limitations rather than LLM limits.
  • domain assumption LLM numeric outputs on a 1-5 Likert scale are directly comparable to human Likert responses
    The evaluation compares predicted scores to human scores without calibrating response styles; LLMs may have different distributional tendencies.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Can LLMs Infer Personality from Real World Conversations?." pith.science (2026). https://pith.science/paper/HJ4ISBB4

@misc{pith2026250714355,
  author       = {Pith},
  title        = {Pith review of: Can LLMs Infer Personality from Real World Conversations?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HJ4ISBB4}},
  note         = {Machine review of arXiv:2507.14355}
}
abstract

Large Language Models (LLMs) such as OpenAI's GPT-4 and Meta's LLaMA offer a promising approach for scalable personality assessment from open-ended language. However, inferring personality traits remains challenging, and earlier work often relied on synthetic data or social media text lacking psychometric validity. We introduce a real-world benchmark of 555 semi-structured interviews with BFI-10 self-report scores for evaluating LLM-based personality inference. Three state-of-the-art LLMs (GPT-4.1 Mini, Meta-LLaMA, and DeepSeek) were tested using zero-shot prompting for BFI-10 item prediction and both zero-shot and chain-of-thought prompting for Big Five trait inference. All models showed high test-retest reliability, but construct validity was limited: correlations with ground-truth scores were weak (max Pearson's $r = 0.27$), interrater agreement was low (Cohen's $\kappa < 0.10$), and predictions were biased toward moderate or high trait levels. Chain-of-thought prompting and longer input context modestly improved distributional alignment, but not trait-level accuracy. These results underscore limitations in current LLM-based personality inference and highlight the need for evidence-based development for psychological applications.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. BlossomPsy: A User-Centric AI System for Adaptive and Engaging MBTI Personality Assessments

    cs.HC 2026-07 conditional novelty 5.0 of 10

    BlossomPsy combines multi-turn LLM dialogue, photo-based questions, a multi-head classifier, and a modified UCB bandit algorithm to deliver MBTI assessments with higher user engagement and preliminary consistency with...

Reference graph

Works this paper leans on

45 extracted references · 29 canonical work pages · cited by 1 Pith paper

  1. [1]

    Accessed July 11, 2025

    APA Dictionary of Psychology. Accessed July 11, 2025. https://dictionary.apa.org/

    work page 2025

  2. [2]

    WIDIGER T

    A. WIDIGER T. Personality and psychopathology. World Psychiatry. 2011;10(2):103-106. doi:10.1002/j.2051-5545.2011.tb00024.x

    work page arXiv 2011

  3. [3]

    Understanding Patient Personality in Medical Care: Five-Factor Model

    Redelmeier DA, Najeeb U, Etchells EE. Understanding Patient Personality in Medical Care: Five-Factor Model. J Gen Intern Med. 2021;36(7):2111-2114. doi:10.1007/s11606-021- 06598-8

    work page doi:10.1007/s11606-021- 2021

  4. [4]

    The Association Between Psychopathic Personality Traits and Criminal Justice Outcomes: Results From a Nationally Representative Sample of Males and Females

    Beaver KM, Boutwell BB, Barnes JC, Vaughn MG, DeLisi M. The Association Between Psychopathic Personality Traits and Criminal Justice Outcomes: Results From a Nationally Representative Sample of Males and Females

    work page

  5. [5]

    Personality dimensions and criminal arrest

    Samuels J, Bienvenu OJ, Cullen B, Costa PT, Eaton WW, Nestadt G. Personality dimensions and criminal arrest. Comprehensive Psychiatry. 2004;45(4):275-280. doi:10.1016/j.comppsych.2004.03.013

    work page doi:10.1016/j.comppsych.2004.03.013 2004

  6. [6]

    A Survey on Personalized Affective Computing in Human- Machine Interaction

    Li J, Waleed A, Salam H. A Survey on Personalized Affective Computing in Human- Machine Interaction. Published online April 1, 2023. doi:10.48550/arXiv.2304.00377

    doi:10.48550/arxiv.2304.00377 2023

  7. [7]

    IMPACT OF VISUAL MODALITIES IN MULTIMODAL PERSONALITY AND AFFECTIVE COMPUTING

    Ryumina EV, Karpov AA. IMPACT OF VISUAL MODALITIES IN MULTIMODAL PERSONALITY AND AFFECTIVE COMPUTING. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 2023;XLVIII-2-W3- 2023:217-224. doi:10.5194/isprs-archives-XLVIII-2-W3-2023-217-2023

    work page doi:10.5194/isprs-archives-xlviii-2-w3-2023-217-2023 2023

  8. [8]

    Attitudes towards AI: measurement and associations with personality

    Stein JP, Messingschlager T, Gnambs T, Hutmacher F, Appel M. Attitudes towards AI: measurement and associations with personality. Sci Rep. 2024;14(1):2909. doi:10.1038/s41598-024-53335-2

    doi:10.1038/s41598-024-53335-2 2024

Show all 45 references
  1. [9]

    The Roles of Personality Traits, AI Anxiety, and Demographic Factors in Attitudes toward Artificial Intelligence

    Kaya F, Aydin F, Schepman A, Rodway P, Yetişensoy O, Demir Kaya M. The Roles of Personality Traits, AI Anxiety, and Demographic Factors in Attitudes toward Artificial Intelligence. International Journal of Human–Computer Interaction. 2024;40(2):497-514. doi:10.1080/10447318.20...

    2024

  2. [10]

    Generative AI and Its Impact on Personalized Intelligent Tutoring Systems

    Maity S, Deroy A. Generative AI and Its Impact on Personalized Intelligent Tutoring Systems. Published online October 14, 2024. doi:10.48550/arXiv.2410.10650

    2024 doi arXiv

  3. [11]

    ResearchGate

    (PDF) A five-factor theory of personality. ResearchGate. July 10, 2025. Accessed July 11,

    2025

  4. [12]

    Toward a new generation of personality theories

  5. [13]

    Personality structure: Emergence of the five-factor model

    Digman JM. Personality structure: Emergence of the five-factor model. Annual review of psychology. 1990;41(1):417-440

    1990

  6. [14]

    Five-factor model of personality

    Soto CJ, Jackson JJ. Five-factor model of personality. Journal of Research in Personality. 2013;42:1285-1302

    2013

  7. [15]

    Personality Detection and Analysis using Twitter Data

    Datta A, Chakraborty S, Mukherjee A. Personality Detection and Analysis using Twitter Data. Published online September 11, 2023. doi:10.48550/arXiv.2309.05497

    2023 doi arXiv

  8. [16]

    Personality Prediction Based on Users’ Tweets

    Dagha R, Dhaord R, Lakhani V, Hirlekar V. Personality Prediction Based on Users’ Tweets. In: 2022 2nd International Conference on Intelligent Technologies (CONIT). ; 2022:1-5. doi:10.1109/CONIT55038.2022.9847670

    2022

  9. [17]

    Automatic Extraction of Personality from Text: Challenges and Opportunities

    Akrami N, Fernquist J, Isbister T, Kaati L, Pelzer B. Automatic Extraction of Personality from Text: Challenges and Opportunities. In: 2019 IEEE International Conference on Big Data (Big Data). IEEE; 2019:3156-3164. doi:10.1109/bigdata47090.2019.9005467

    2019

  10. [18]

    The Accuracy of Self-Reported Intuitive and Analytical Ability

    Sobyra JA. The Accuracy of Self-Reported Intuitive and Analytical Ability

  11. [19]

    In Support of Personality Assessment in Organizational Settings

    Ones DS, Dilchert S, Viswesvaran C, Judge TA. In Support of Personality Assessment in Organizational Settings. Personnel Psychology. 2007;60(4):995-1027. doi:10.1111/j.1744- 6570.2007.00099.x

    2007

  12. [20]

    A contemplated revision of the NEO Five-Factor Inventory

    McCrae RR, Costa PT. A contemplated revision of the NEO Five-Factor Inventory. Personality and Individual Differences. 2004;36(3):587-596. doi:10.1016/S0191- 8869(03)00118-1

    2004 doi

  13. [21]

    Sparks of Artificial General Intelligence: Early experiments with GPT-4

    Bubeck S, Chandrasekaran V, Eldan R, et al. Sparks of Artificial General Intelligence: Early experiments with GPT-4. Published online April 13, 2023. doi:10.48550/arXiv.2303.12712

    2023 doi arXiv

  14. [22]

    Zero-Shot Strike: Testing the generalisation capabilities of out-of-the-box LLM models for depression detection

    Ohse J, Hadžić B, Mohammed P, et al. Zero-Shot Strike: Testing the generalisation capabilities of out-of-the-box LLM models for depression detection. Computer Speech & Language. 2024;88:101663. doi:10.1016/j.csl.2024.101663

    2024

  15. [23]

    Detecting PTSD in Clinical Interviews: A Comparative Analysis of NLP Methods and Large Language Models

    Chen F, Ben-Zeev D, Sparks G, Kadakia A, Cohen T. Detecting PTSD in Clinical Interviews: A Comparative Analysis of NLP Methods and Large Language Models. Published online April 1, 2025. doi:10.48550/arXiv.2504.01216

    2025 doi

  16. [24]

    Evaluating the ability of large language models to emulate personality

    Wang Y, Zhao J, Ones DS, He L, Xu X. Evaluating the ability of large language models to emulate personality. Sci Rep. 2025;15(1):519. doi:10.1038/s41598-024-84109-5

    2025 doi

  17. [25]

    Classifying Unstructured Text in Electronic Health Records for Mental Health Prediction Models: Large Language Model Evaluation Study

    Cardamone NC, Olfson M, Schmutte T, et al. Classifying Unstructured Text in Electronic Health Records for Mental Health Prediction Models: Large Language Model Evaluation Study. JMIR Medical Informatics. 2025;13(1):e65454. doi:10.2196/65454 9

    2025 doi

  18. [26]

    Do Large Language Models (LLMs) Really Understand Personality? A Test of Embeddings vs

    Maharjan J. Do Large Language Models (LLMs) Really Understand Personality? A Test of Embeddings vs. Zero-Shot (Preprint). Published online April 1, 2025. doi:10.2196/preprints.75347

    2025 doi

  19. [27]

    Leveraging Large Language Models to Analyze Emotional and Contextual Drivers of Teen Substance Use in Online Discussions

    Zhu J, Jin R, Jiang H, Wang Y, Zhang X, Coifman KG. Leveraging Large Language Models to Analyze Emotional and Contextual Drivers of Teen Substance Use in Online Discussions. Published online January 23, 2025. doi:10.48550/arXiv.2501.14037

    2025 doi

  20. [28]

    Large-Scale Deep Learning–Enabled Infodemiological Analysis of Substance Use Patterns on Social Media: Insights From the COVID-19 Pandemic

    Maharjan J, Zhu J, King J, Phan N, Kenne D, Jin R. Large-Scale Deep Learning–Enabled Infodemiological Analysis of Substance Use Patterns on Social Media: Insights From the COVID-19 Pandemic. JMIR Infodemiology. 2025;5(1):e59076. doi:10.2196/59076

    2025 doi

  21. [29]

    Accessed April 2, 2025

    Consistency of the Structural Properties of the BFI-10 Across 16 Samples From Eight Large-Scale Surveys in Germany. Accessed April 2, 2025. https://econtent.hogrefe.com/doi/epdf/10.1027/1015-5759/a000765

    2025 doi

  22. [30]

    Husain W, Haddad AJ, Husain MA, et al. Reliability generalization meta-analysis of the internal consistency of the Big Five Inventory (BFI) by comparing BFI (44 items) and BFI-2 (60 items) versions controlling for age, sex, language factors. BMC Psychology. 2025;13(1):20. doi:...

    2025 doi

  23. [31]

    Measuring personality in one minute or less: A 10-item short version of the Big Five Inventory in English and German

    Rammstedt B, John OP. Measuring personality in one minute or less: A 10-item short version of the Big Five Inventory in English and German. Journal of Research in Personality. 2007;41(1):203-212. doi:10.1016/j.jrp.2006.02.001

    2007 doi

  24. [32]

    psych: Procedures for Psychological, Psychometric, and Personality Research

    Revelle W. psych: Procedures for Psychological, Psychometric, and Personality Research. Published online May 6, 2007. doi:10.32614/cran.package.psych

    2007 doi

  25. [33]

    A study of the personality trait-focused digital mental health intervention

    Jeong S, Kim H, Lho SK, et al. A study of the personality trait-focused digital mental health intervention. European Psychiatry. 2024;67(S1):S554-S555. doi:10.1192/j.eurpsy.2024.1152

    2024

  26. [34]

    Personality traits and dimensions of mental health

    Kang W, Steffens F, Pineda S, Widuch K, Malvaso A. Personality traits and dimensions of mental health. Sci Rep. 2023;13(1):7091. doi:10.1038/s41598-023-33996-1

    2023 doi

  27. [35]

    Predicting the Big Five Personality Traits in Chinese Counselling Dialogues Using Large Language Models

    Yan Y, Ma L, Li A, Ma J, Lan Z. Predicting the Big Five Personality Traits in Chinese Counselling Dialogues Using Large Language Models. Published online June 25, 2024. doi:10.48550/arXiv.2406.17287

    2024 doi arXiv

  28. [36]

    Automatic personality assessment through social media language

    Park G, Schwartz HA, Eichstaedt JC, et al. Automatic personality assessment through social media language. Journal of Personality and Social Psychology. 2015;108(6):934-952. doi:10.1037/pspp0000020

    2015 doi

  29. [37]

    A caution against customized AI in healthcare

    Kostick-Quenet KM. A caution against customized AI in healthcare. npj Digit Med. 2025;8(1):13. doi:10.1038/s41746-024-01415-y

    2025 doi

  30. [38]

    GPT-4 Technical Report

    OpenAI, Achiam J, Adler S, et al. GPT-4 Technical Report. Published online March 4, 2024. doi:10.48550/arXiv.2303.08774

    2024 doi arXiv

  31. [39]

    The Llama 3 Herd of Models

    Grattafiori A, Dubey A, Jauhri A, et al. The Llama 3 Herd of Models. Published online November 23, 2024. doi:10.48550/arXiv.2407.21783

    2024 doi arXiv

  32. [40]

    Two of them always come to mind. The first one is the first baby we delivered—early morning call at 3 or 4 a.m.—it was a heroin case…

    DeepSeek-AI, Guo D, Yang D, et al. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. Published online January 22, 2025. doi:10.48550/arXiv.2501.12948 10 Figures and Tables Table 1. Example Interview Responses Participant ID: 001 Q1 (Daily Acti...

    2025 doi arXiv

  33. [44]

    \"\"{response}\

    Justify the trait scores based on these observations in 2-3 sentences. Provide a score for each trait, from 1.0 to 5.0 (inclusive), allowing for half-point values such as 3.5. Use the exact format below: Conscientiousness: X Agreeableness: X Neuroticism: X Openness: X Extraver...

  34. [45]

    Summarize the emotional tone and key themes (e.g., positive, negative, confident, anxious)

  35. [46]

    Identify any linguistic cues or behaviors that indicate the traits (e.g., responsibility for conscientiousness, empathy for agreeableness)

  36. [47]

    Provide a score (1.0-5.0) for each trait, allowing for half-step values like 3.5

    Justify the trait scores based on these observations in 2-3 sentences. Provide a score (1.0-5.0) for each trait, allowing for half-step values like 3.5. Strictly use this format: Conscientiousness: X Agreeableness: X Neuroticism: X Openness: X Extraversion: X Where X is a numb...

  37. [2025]

    https://www.researchgate.net/publication/284978581_A_five- factor_theory_of_personality

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.