Cornell study shows user prompting habits lock in within five chats

Cornell researchers analyzed thousands of anonymized real-world ChatGPT chat logs to see how people phrase their requests over time. They found that while people ask chatbots to handle an ever-expanding range of tasks, their phrasing choices lock into personal routines within five chat sessions. Users typically settle on a few favorite prompt templates rather than exploring new ways to communicate.

This behavior reveals a gap between how flexible AI models are and how rigidly humans end up using them. Because open text boxes accept any input, many assume users refine their style over time to get better responses. In reality, people quickly build individual habits, such as relying on consistent sentence structures or politeness levels, across coding, writing, and advice. Crucially, the researchers observed that major upgrades to AI models served as the primary trigger for people to break out of their routine phrasing.

It remains unclear whether these fixed habits help users get better results or if they simply represent comfortable paths of least resistance, as the study did not measure answer quality or user satisfaction. If people naturally fall into repetitive communication patterns, AI platforms might eventually need to proactively nudge users toward better prompting techniques, rather than expecting individuals to discover broader capabilities on their own.