ChatGPT may be doing more than writing emails, brainstorming ideas, and settling arguments about dinner. New research suggests the AI can also create personality tests—and predict how people will answer them.
The study, published August 6 in iScience, found that GPT-4 could turn almost any written material into a personality questionnaire. Researchers from the Hebrew University of Jerusalem tested the approach using two very different sources: the DSM-5, a widely used clinical manual for mental disorders, and an astrology textbook.
The unusual pairing was deliberate. The DSM-5 is built on decades of psychological research, while astrology is rooted largely in cultural beliefs rather than scientific evidence.
GPT-4 generated personality questions from both texts, and 600 people completed the tests alongside the established Big Five personality questionnaire.
The DSM-5-based test performed surprisingly well. Its questions produced personality patterns similar to those found in the Big Five test, suggesting AI can extract meaningful psychological information from complex text.
The astrology test was less consistent. Still, researchers found that both questionnaires could predict outcomes such as anxiety, depression, and well-being from participants’ answers.
But the biggest surprise came before anyone answered a question.
ChatGPT was able to predict how participants would respond to the AI-generated questionnaires. It accurately estimated average responses and even predicted relationships between different answers.
Researchers believe this could happen because large language models learn from enormous amounts of human-written language. While they were not specifically trained to understand personality, they may have absorbed patterns of human behavior simply by studying how people communicate.
There are important limits. The study focused largely on English and Western cultural contexts, so the results may not translate perfectly to other languages or cultures.
Still, the research suggests AI could become a powerful shortcut for personality research—turning existing texts into testable psychological tools and, perhaps more intriguingly, predicting what people will say before they say it.
