This research reveals how cognitive biases restrict users from fully utilizing LLM-enabled chat search, especially for complex tasks.

• Users tend to interact with LLMs in expected or familiar ways, limiting exploration
• Decision-making across public safety, health, sustainability, and AI ethics domains is affected
• Pre-existing mental models constrain how people leverage new AI capabilities
• Users need better guidance to break free from fixed interaction patterns

For education, this highlights the importance of teaching adaptive interaction strategies with AI tools and designing systems that encourage flexible usage patterns beyond users' initial expectations.

Trapped by Expectations: Functional Fixedness in LLM-Enabled Chat Search