Essays written by GPT and human subjects under the same conditions display the same emotional responses

We present a new method to generate a control group for psychometric studies based on text, which uses a ChatGPT (or other Large Language Model) under the same test conditions as the human subjects. We demonstrate the approach by comparing emotion levels in human-generated and synthetic essays, showing that the variations in emotion scores in the human data are also found in the synthetic data, therefore demonstrating that the signal can be explained by the prompt alone. We believe that the approach is general, and can lead to a new way to use large language models in psychometric research.