Reading Between the Lines: How LGBTQ+ Users Turn to AI Chatbots for Mental Health, Gender, and Sexuality Advice

User-facing large language models (LLMs) and chatbots powered by artificial intelligence (AI) have been shown to reflect and amplify social biases present in training datasets, especially towards marginalized groups. Members of these groups often feel hesitant towards using AI products for these reasons and other contributing factors. Simultaneously, LGBTQ+ users may resort to this technology for personal advice when safety concerns or situational barriers make traditional resources inaccessible. In this report, we outline our progress in characterizing the specific topics and linguistic methods by which LGBTQ+ users interact with AI chatbots in the domains of mental health, gender/sexual health, and relationship advice. Our ultimate goal with this work-in-progress research is to not only better understand and analyze LGBTQ+ community use-cases for AI chatbots in these sensitive areas, but learn from these realistic interactions in order to develop more grounded, ecologically valid LLM evaluations.