Participants often misidentified responses, assuming ChatGPT's were human, leading to higher ratings for AI therapy.