Place limited, provisional confidence in a personality report inferred from social media. Research finds that digital activity carries some signal about broad personality tendencies, yet agreement with questionnaires is modest and depends on the data and method. A report that sounds certain, or repeats the same result across runs, has not thereby shown that its claims describe you accurately. Use a specific claim as a low-stakes hypothesis to check against repeated behavior across settings, and ask what evidence supports the exact report and intended use before acting on it.
What does the recent evidence actually establish?
Suppose a report says your public posts suggest high openness or extraversion. Research indicates that social media can contain personality signal, but the key question is how far group-level findings support a description of one person. The recent direct study offers a cautious answer.
A study first published online on September 2, 2025, analyzed two years of Facebook posts from 1,214 Italian users. Data collection ran from March through June 2018; the posts covered the preceding 24 months. Researchers used two language models to infer Big Five traits and compared the results with participants’ scores on the Ten-Item Personality Inventory (TIPI), a brief self-report questionnaire. When inferences were combined across models and time points, correlations with self-reported traits ranged from .18 for neuroticism to .31 for openness. These are modest sample-level associations, not percentages of a person correctly understood or probabilities that a report sentence is true. The benchmark also has limits: TIPI has only two items per trait, and its internal consistency in this study varied from .32 to .70 across traits.
Earlier syntheses provide context, not a scorecard for today's reports. A 2018 meta-analysis identified 28 studies, although its quantitative analyses included 16 independent studies and 80 effect sizes; reported correlations across the Big Five ranged from .29 to .40. A 2024 systematic review and meta-analysis at the University of Bath synthesized 534 effect sizes from 42 computer-prediction studies and found moderate convergence with self-reports (ρ=.30). It also reported that prediction varied with data source. These reviews combine different samples, platforms, inputs, and methods, so their pooled associations cannot validate a particular commercial report or establish a simple trend over time. They show that digital traces can carry group-level personality signal; they do not tell you how accurately a specific report describes an individual.
The 2025 study involved a snowball sample that was 74.4% female and 73.6% aged 18–25, one platform, and a brief two-item-per-trait inventory. Its results cannot automatically transfer to a current report built from another platform, language, profile length, or trait framework. Fluent wording adds no evidence. Social activity may suggest broad tendencies in samples, while confidence in a specific person's account remains limited.
Sources: Inferring Personality From Social Media Activity Using Large Language Models: Cross-Model Agreement, Temporal Stability, and Convergent Validity With Self-Reports; Predicting the Big 5 Personality Traits From Digital Footprints on Social Media: A Meta-Analysis; Digital Data and Personality: A Systematic Review and Meta-Analysis of Human Perception and Computer Prediction
How should I compare a social-media report with a questionnaire?
Compare them on the same trait and against the same stated purpose. In the recent direct study, social-media inferences were compared with questionnaire self-reports. That is a relevant benchmark, but not a perfect view of a person: people may misunderstand an item, answer with a particular reference period in mind, or summarize themselves differently from how they behave in a narrow situation. A questionnaire is evidence about self-perception, not an unquestionable ground truth.
The methods also observe different things. A questionnaire asks someone to answer selected trait statements. Social media offers traces of behavior filtered through what the person chooses or needs to post, the audience, the platform’s format, and how often the account is used. A person who posts frequently for work may appear socially outgoing online without that pattern describing their preferences in private or in other settings. This is an interpretation of how the methods differ, not a result tested by the cited study for each participant.
One especially useful distinction is between consistency and validity. Consistency asks whether a method gives similar results across repetitions or time. Validity asks whether evidence supports the meaning and use assigned to those results. In the 2025 study, inferences averaged across repeated runs showed substantial repeatability, and estimates from different systems were correlated. Yet their association with self-reports remained modest. A result can therefore be repeatable without being a sufficiently accurate personal description. Agreement between systems is not an independent confirmation if both rely on similar visible cues or share similar blind spots.
This does not make self-report automatically superior. The 2018 review found that combining multiple footprint types often improved prediction in its included studies, while the 2024 Bath synthesis found variation by data source. Neither finding establishes a universal ranking of methods. A report should explain what data it used, what trait it estimates, how it was compared with a measure of that trait, who was studied, and what use the evidence supports. Without those details, a polished conclusion cannot be matched to an appropriate comparison. The Bath pooled estimate reflects many computer-prediction studies, so it cannot tell you how a particular platform or vendor performs.
Sources: Inferring Personality From Social Media Activity Using Large Language Models: Cross-Model Agreement, Temporal Stability, and Convergent Validity With Self-Reports; Predicting the Big 5 Personality Traits From Digital Footprints on Social Media: A Meta-Analysis; Digital Data and Personality: A Systematic Review and Meta-Analysis of Human Perception and Computer Prediction
What should I do with a claim in the report?
Check each statement separately. First, translate it into an observable tendency: what does the report say the person often does, and under what conditions? Then ask whether it describes a flexible pattern or turns that pattern into a fixed identity. Next, look for the report’s data window, scoring approach, comparison group, uncertainty, and evidence for the use you have in mind. If the report gives none of these, regard its confidence as a feature of its writing, not a measure of evidence.
Consider a hypothetical report that infers comfort with social interaction from frequent public posts. The posts might reflect a job requirement, a narrow audience, or an interest in sharing one kind of activity. They do not by themselves show how the person approaches unfamiliar groups, private conversations, or time alone. To evaluate the claim, compare it with several actual examples across settings and ask the person whether the pattern fits, where it does not, and what context changes it. One counterexample does not erase a tendency; repeated, relevant observations make the interpretation more useful.
A sensible confidence check has three levels. For a broad, low-stakes prompt such as ‘you may seek out varied ideas,’ the report can suggest a question for reflection if the person recognizes examples and exceptions. For a concrete statement about how someone handles feedback or plans work, seek repeated examples from more than one situation and consider situational demands. For a consequential judgment about another person, do not treat a social-media inference as adequate evidence of character or capability. None of the cited studies establishes that these reports diagnose a condition, predict an individual’s job performance, or select the right role.
Privacy is part of the decision too. Social posts can reveal information beyond what a person intended a report reader to infer. Do not upload another person’s account or treat public availability as permission for personality assessment. If you are evaluating a report about yourself, check what data were collected, who can access them, and whether you can remove them. The research base is not a substitute for consent or a reason to make hidden judgments.
For broad, revisable self-reflection, give the report limited confidence; do not use it for consequential judgments without evidence specific to that population and use. If one claim catches your attention, note examples that support it and one that complicates it, then decide whether it is a useful question to explore.
Questions readers ask
Does a stable social-media personality result mean it is accurate?
No. Stability means similar outputs recur across repetitions or time; validity asks whether evidence supports the interpretation for its intended use. The study first published in 2025 found repeatable inferences alongside modest correlations with TIPI self-reports, whose internal consistency also varied across traits. Consistency alone is not enough.
Can I use a social-media personality report to judge someone at work?
The cited research does not establish that these reports are suitable for hiring, promotion, or other employment judgments. Treat a claim as a tentative hypothesis at most, and do not use it as a score of capability or fit.
Sources and notes
- Inferring Personality From Social Media Activity Using Large Language Models: Cross-Model Agreement, Temporal Stability, and Convergent Validity With Self-Reports
Wiley record: first published September 2, 2025. Its abstract and methods report 1,214 Italian Facebook users, two years of posts, Gemini 1.5 Pro and GPT-4o compared with Ten-Item Personality Inventory self-reports; combined inference correlations with self-reports ranged .18–.31, and combined estimates showed temporal stability. The full record reports snowball recruitment, March–June 2018 data collection, and TIPI internal consistency of .32–.70.
- Predicting the Big 5 Personality Traits From Digital Footprints on Social Media: A Meta-Analysis
The opened paper reports 28 studies identified and 16 independent studies (80 effect sizes) included in meta-analyses; correlations ranged from .29 for agreeableness to .40 for extraversion. It reports significant heterogeneity and that combining footprint types improved prediction for some traits.
- Digital Data and Personality: A Systematic Review and Meta-Analysis of Human Perception and Computer Prediction
The opened University of Bath research record reports Study 2 synthesized 534 effect sizes across 42 computer-prediction studies, with moderate convergence with self-reports (ρ=.30), and reports data-source moderators.
Apply it to your work
Turn a work-pattern hunch into observations you can check
From this guide: If the report raises a question about how you plan, handle feedback, or collaborate, compare it with specific situations rather than accepting a label.
A social-media inference cannot show how a tendency appears across your work settings. The Work Pattern Report offers a private, low-stakes self-report across decisions, planning, feedback, conflict, collaboration, change, and learning. It supplies no norms or hiring score; use its prompts to name patterns you can compare with actual experience.