This paper looks at Human-AI interaction through everyday, long-running use of conversational AI. It focuses on how tone, assumptions, and behaviour can shift across a conversation, and how those shifts feel from the user’s side as they happen. Using ordinary interactions with tools such as Microsoft Copilot and ChatGPT, the paper describes a set of recurring patterns that tend to show up over time. Two of these are named and defined. Botsplaining is when a conversational AI explains a user’s emotional state back to them, often inaccurately, prematurely, or without being asked, in a way that feels condescending, intrusive, or dismissive of the user’s lived experience. Botsplaining commonly appears during moments of distress, when tone sensitivity is most critical and emotional missteps can cause harm. Gasbotting occurs when a conversational AI confidently misrepresents the situation, misattributes the source of an error, or reframes reality in a way that suggests the user is mistaken, when in fact the issue originated from the model itself. This paper focuses on normal use outside of the project room. It looks at how people interact with AI in everyday situations: asking questions, working through ideas, thinking out loud, or dealing with something difficult while still needing to function. In those moments, small shifts in wording, timing, or tone can change an interaction from grounding to frustrating, or from helpful to uncomfortable. These observations are drawn from lived use rather than theory. They’re shared as patterns that may be useful to people building, shaping, or working with conversational AI who want a clearer sense of how these interactions can land for someone on the other side of the chat window.
Melissa Kate Shaw (Thu,) studied this question.