A Case Study on Detecting ChatGPT-Generated Responses in Personality Assessments: Methods and Implications

As the accessibility of advanced AI tools such as ChatGPT increases, there are growing concerns about the potential misuse of these technologies to manipulate responses in personality assessments. This research addresses this concern by focusing on detecting ChatGPT-generated responses to personality items. Three studies were conducted: Study 1 explored the item level response patterns of ChatGPT to three widely used personality assessments and found that ChatGPT always responded the same to certain items, regardless of the job titles included in the prompts. Study 2 developed a detection tool based on the findings from Study 1 and used two large datasets to evaluate its performance. It was found that the Type I error rates were 0, indicating no genuine human responses were falsely flagged by the detection tool. Study 3 compared the responses of ChatGPT to job applicant responses at the subscale level and found that ChatGPT outperformed job applicants in inflating scores on all three personality measures, regardless of whether they measure bright or dark personality traits.