Ask any chat tool one real factual question. Then follow up: 'Rate your confidence 1-10 and name what evidence would change your answer.' You're done when you've decided — based on its reply — whether this answer needs outside verification before you'd use it.
How-to · VER · VER.1
Ask what would change its answer
State your confidence and what evidence would change this answer' exposes weak reasoning better than accepting fluent output.
Worked example
A fluent answer and a reliable answer look identical on screen. One probe separates them: 'How confident are you, and what evidence would change this answer?' A model that can't name its own failure conditions is telling you to verify harder.