Can ChatGPT Read Who You Are?

ChatGPT can infer personality traits from text with competitive accuracy compared to human raters, suggesting large language models have genuine capability in psychological assessment—but this same capability raises serious privacy concerns since personality can be extracted from seemingly innocuous writing samples.

The system exhibits a consistent "positivity bias," rating people more favorably across all personality dimensions than they rate themselves, which could lead to systematically inaccurate personality profiles if deployed in real applications like hiring, mental health screening, or educational placement.

How you phrase questions to ChatGPT significantly affects its personality assessments, indicating that results are not objective measurements but rather dependent on prompt design—a critical finding for anyone considering using AI for consequential personality-based decisions.

The research emphasizes that personality inference from text raises major ethical concerns around privacy, consent, and autonomy that go beyond technical accuracy; people may not realize their writing is being analyzed for psychological traits, and they haven't agreed to such assessment.

This work demonstrates the importance of responsible AI development in psychology-adjacent applications, suggesting that even when AI performs well technically, deployment requires careful consideration of how the technology might be misused or cause harm to individuals.