Asking ChatGPT what it is unsure about can reveal weak assumptions

TechRadar’s Eric Hal Schwartz describes a simple follow-up question that can expose weaknesses in ChatGPT’s answers: “What are you unsure about in your answer?”

In a travel-planning test, ChatGPT produced a family itinerary that appeared practical but had assumed the family’s energy levels, child’s nap schedule, preferred downtime and accommodation location. When asked about its uncertainty, ChatGPT identified those assumptions and questioned the itinerary’s pacing. Schwartz then added clearer constraints, including one significant activity each day and flexible afternoons, and ChatGPT rebuilt the plan.

A similar test involved an older laptop that was struggling. ChatGPT initially suggested that replacement might be the more sensible long-term option. When asked what it was unsure about, the chatbot acknowledged that it did not know the repair cost, the failing component, the laptop’s model, specifications or symptoms. It said a relatively cheap battery or storage upgrade could instead extend the computer’s useful life.

The question can also lead to follow-ups about which uncertainty could change a recommendation or what missing information ChatGPT needs. Schwartz notes that ChatGPT’s own description of its confidence can be wrong and is not a substitute for checking reliable sources when facts matter. Still, he says the prompt can make hidden assumptions easier to examine in everyday conversations.