How do you design an AI-powered feature whose outputs are sometimes wrong, so users trust it the right amount?

What the interviewer wants to hear

• Match the use case to the risk of errors
• Set expectations and label AI content
• Show sources, confidence or reasoning to verify
• Easy editing, undo and human control
• Feedback loops and measuring trust and accuracy

Model answer

I start with risk: what happens when the AI is wrong? For low-risk tasks like suggesting a title, errors are cheap. For contracts or medical notes, the design must make verification easy. I set expectations clearly, label AI content and explain what it can and cannot do. Where possible I show sources or the input behind each claim so users can check quickly. The user stays in control: suggestions can be edited, accepted or undone, and high-stakes actions need confirmation. I design error and empty states for when the model fails or refuses. I add lightweight feedback like thumbs or corrections, which helps both users and the model team. Finally, I measure not just usage but whether users catch errors and whether their trust matches actual accuracy.

More hard UI/UX Designer questions

• Sales, marketing and support each want different changes to the same screen, all marked urgent. How do you decide what to design?
• You led a redesign and after launch the conversion rate dropped by 8%. What do you do?
• Tell me how you would handle a situation where a business goal seems to conflict with what users need.
• Our product has grown to 10 teams and the interface feels inconsistent. How would you fix this?
• Leadership thinks user research slows the team down. How would you convince them to invest in it?
• Leadership asks you to make cancelling a subscription harder to reduce churn. How do you respond?
• Your company runs three brands on one product platform, each with light and dark themes. How would you structure the design tokens?
• How do you design for a complex enterprise tool used all day by expert users, such as a logistics dispatch system?
• You must make an important design decision, but research is inconclusive and there is no time for more. How do you decide?