As AI chatbots become widely adopted for functional, social, and emotional use, concerns around how their design impacts user interaction and wellbeing are growing. While early research highlights potential benefits from chatbots — such as reduced loneliness and increased perceived support — it also raises concerns, including possible emotional dependence, social isolation, risks to privacy, and financial harm. These risks not only emerge from the underlying models powering chatbots, but are also deeply influenced by the design choices embedded in chatbot interfaces and interactions.
In this report, we examine AI chatbots through the lens of dark patterns — deceptive or manipulative design choices that may undermine user autonomy or well-being. Drawing on prior work in human-computer interaction and deceptive design disciplines, we investigate how established dark patterns documented in other digital technology contexts may translate to AI chatbots –– particularly those positioned for social, emotional, or relational use.
Using a deductive, multi-stage literature review methodology, we aim to build the conceptual foundation for dark patterns in AI chatbots by identifying which dark patterns are possible, applicable, and relevant in the chatbot context. Our research synthesized hundreds of existing dark patterns, filtered them for relevance, and analyzed them. The result is a comprehensive taxonomy of 37 dark patterns applicable to AI chatbots, referring to both general-purpose systems (e.g., ChatGPT, Gemini, Claude) and “companion” platforms (e.g., Replika, Character.AI).
Click here to read the full report on the CDT website.