Here's the dirty secret of AI user research: users say they want AI, but they don't trust it.

Every survey says the same thing. "Would you like AI to help you with [task]?" 80% say yes. Then you ship it and 80% of them turn it off within a week.

Why? Because there's a massive gap between "that sounds cool" and "I trust this enough to use it for real work."

What users SAY in surveys:     "Yes, I'd love AI to help with that!"

What users DO in production:   Ignore the suggestion
                               Edit the AI output so heavily it's a rewrite
                               Turn the feature off
                               Complain to support that "the AI is wrong"

Traditional user research — surveys, interviews about hypothetical features — is almost useless for AI products. You need different methods.