LLMs Can Generate Psychologically Authentic Life Stories from Real Personality Profiles

LLMs Generate Life Stories That Encode Personality as Reliably as Humans

A groundbreaking study from cs.CL researchers demonstrates that large language models can generate first-person life story narratives from real psychometric profiles, with personality traits recoverable from the text at levels approaching human test-retest reliability.

The Study Design

Researchers conditioned LLMs on real psychometric profiles from 290 participants to generate first-person life story narratives, then tasked independent LLMs to recover personality scores from those narratives alone.

Key Results

What This Means

1. Personality is deeply encoded in natural language — LLMs can reproduce behavioral patterns that match real human psychology
2. Surface compliance is real — Scoring models achieve accuracy while counteracting alignment-induced defaults, suggesting genuine personality encoding rather than mimicry
3. Cross-model robustness — Results hold across different model architectures and providers
4. Emotional reactivity patterns in generated narratives replicate patterns found in real conversational data

Implications

• Clinical psychology — AI-generated narratives could supplement personality assessment tools
• Character development — Writers could use LLMs to create psychologically consistent characters
• AI safety — Understanding how LLMs encode personality is critical for responsible deployment
• Human understanding — The study reveals how personality manifests in natural language patterns