Behavioral questions
Amazon
How do you verify whether AI-generated work is correct?
BehavioralLast seen Nov 9, 2026 · Software engineer
AI usage
Also asked as
- When do you trust an LLM and when do you not?
- How do you ensure the output of AI is reliable?
What they're looking for
This tests Dive Deep style thinking and good judgment about AI. The interviewer wants a clear, repeatable method for checking output, a sense of when to trust a model and when not to, and examples of catching errors. A strong answer treats AI output as a draft that needs evidence.
What works
- Describe your checks in order: read it, run it, test it, compare against a source of truth.
- Explain how the risk of the task changes how much you check, for example a throwaway script versus production code.
- Give one real example where verification caught a wrong or made up result.
- Say when you do not trust a model at all, such as exact numbers, recent facts or security code.
What hurts
- Saying you just run it and see if it works.
- Claiming you always trust it, or never trust it, with no middle ground.
- Forgetting that tests written by the same AI can share the same mistake.
- Giving no example, only general rules.
Sample answer
Think of your own story first, then compare.