# HUJI Study Uses GPT-4 to Build Personality Surveys > Source: > Published: 2026-08-22 07:44:45+00:00 # HUJI Study Uses GPT-4 to Build Personality Surveys Hebrew University researchers reported in iScience on August 6 that GPT-4 generated personality questionnaires from DSM-5 and astrology texts, then forecast population-level response patterns before 600 participants completed them. The DSM-based questionnaire showed stronger internal consistency than the astrology-based version, a reminder that generated survey items still require human validation. Researchers at the Hebrew University of Jerusalem reported a method for using GPT-4 to create and evaluate personality questionnaires from source texts. The peer-reviewed iScience study, published August 6, used excerpts from the DSM-5 personality-disorders section and, as a deliberately weak control, an astrology text. The team generated one questionnaire from each source and administered both alongside the Big Five Inventory to 600 adults. The DSM-based questionnaire showed high internal consistency within personality clusters and tracked patterns seen in the Big Five measure. The astrology-based questionnaire showed weaker internal consistency, although individual items still carried some signal for outcomes such as anxiety, depression and well-being. ### What the model predicted Before the participant data were collected, the researchers asked the model to estimate mean responses and relationships among questionnaire items. The study reports that those forecasts aligned with the eventual population-level response patterns for both generated questionnaires. That result is about aggregate psychometric structure, not a finding that a language model can diagnose an individual or replace clinical assessment. The authors also caution that the observed results may not generalize across languages and cultures, given the training-data context of the models. ### A faster way to prototype is not a finished instrument The practical contribution is a workflow for producing candidate survey items and checking whether their expected structure looks plausible before committing to a larger human study. It does not remove the need for construct validation, representative sampling, cultural testing or careful review of the source material used to generate items. LDS interpretation: this is most useful as a research-prototyping method. Teams using LLMs in psychometrics should preserve the boundary between generating a draft instrument and establishing that it is reliable, fair and appropriate for a specific population or decision. ## Key Points - 1The iScience study used GPT-4 to generate personality questionnaires from DSM-5 and astrology source texts, then administered them with the Big Five Inventory to 600 adults. - 2The DSM-based questionnaire showed stronger internal consistency than the astrology-based version, while the model forecast aggregate response patterns before data collection. - 3The researchers caution that results may not generalize across languages and cultures; generated items still need human validation before consequential use. ## Scoring Rationale The peer-reviewed study offers a concrete LLM-assisted psychometrics workflow and a bounded evaluation result. Its relevance lies in research-method design and validation discipline, not in an immediately deployable diagnostic product. ## Sources Primary source and supporting public references used for this report. Practice interview problems based on real data 1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)