A study asked generative AI models, including ChatGPT, Claude and LLaMa, to read ordinary language that people produced. The researchers used short daily video diaries and longer recordings from more than 160 participants gathered in both real-life and lab settings.
The AI systems answered standard personality questions for each person, and older text-analysis methods did not perform nearly as well. The study found that AI personality scores closely matched self-ratings and in many cases matched those self-ratings better than ratings from friends or family.
The AI ratings also predicted real aspects of life: daily emotions, stress levels, social behaviour, and whether a person had a mental health diagnosis or had sought treatment. The authors and other academics said the findings show language carries strong clues about psychological traits, but important questions about comparisons across groups and methods remain.
Difficult words
- generative — that creates new text or other content
- participant — a person who takes part in a studyparticipants
- self-rating — a person's own score of their traitsself-ratings
- predict — to say what will happen laterpredicted
- diagnosis — a named medical or mental health condition
- trait — a characteristic of a person's behaviour or mindtraits
- text-analysis — methods that study written or spoken words
Tip: hover, focus or tap highlighted words in the article to see quick definitions while you read or listen.
Discussion questions
- Have you ever written or said something that showed your personality? Give a short example.
- Would you feel comfortable if an AI used your messages to predict your emotions? Why or why not?
- What benefits and problems could come from using AI to predict mental health?
Related articles
New method could link quantum computers over long distances
Researchers at the University of Chicago describe a theoretical method to connect quantum computers over about 2,000 km by improving how long atoms keep quantum states. The team used a different crystal growth process and will test links in the lab.
Study: Low numbers of women in science academies
An international analysis finds women made up about 19% of academy members in 2025, with little change in recent years. The report highlights leadership gaps, higher reports of harassment, and calls for stronger gender equality measures.
Social media can give early warning of displacement
Researchers find that analysing social media posts can give early warning of population movements and help humanitarian agencies respond faster. The study in EPJ Data Science tested methods across three case studies using nearly 2 million posts on X.