Why the tone tells you nothing.
THE PROBLEM
AI output reads with uniform confidence whether it is accurate or invented.
A person hedges when unsure. These systems generally do not, unless asked to.
WHY
The training data is mostly confident prose. The model reproduces that register.
WHAT THIS MEANS PRACTICALLY
You cannot judge reliability from how the answer sounds.
ASKING FOR UNCERTAINTY
Prompting explicitly helps: "state clearly where you are uncertain" or "flag anything you may be wrong about".
The result is imperfect and better than nothing.
WHAT TO TRUST MORE
Answers about material you provided Answers about widely documented subjects Reasoning you can follow and check
WHAT TO TRUST LESS
Specific figures Citations and references Recent events Anything obscure Anything where being wrong would matter
THE HABIT WORTH BUILDING
Ask yourself what the cost is if this particular statement is wrong.
Verify in proportion to that cost.
