Researchers examined how 26 participants used ChatGPT during a complex web development task. They identified nine failure types, grouped into incorrect or incomplete responses, cognitive overload and context loss.
Participants tried to address these problems by adding structure to their requests, clarifying prompts and debugging the system’s answers. Still, 17 participants eventually abandoned ChatGPT after persistent failures. The analysis found that unhelpful responses increased the likelihood of abandonment by a factor of 11, while each additional prompt reduced the probability of abandonment by 17%.



