Why Does ChatGPT Give Wrong Answers? And How to Make It Stop

Why does ChatGPT just make things up?

Because it predicts plausible words, not verified facts. When it lacks the information, it confidently constructs a plausible answer. This is called hallucination and every AI model does it to some degree.

How often does it happen?

Varies by topic. Common, everyday topics with tons of training data are usually reliable. Obscure, very specific or recent topics produce more errors. Highly technical or numerical content needs extra checking.

How do I reduce wrong answers?

Give it sources or your own data to work from ('use only this document'), ask it to tell you what it's uncertain about, ask for the reasoning, and always double-check numbers and legal/financial facts independently.

Is there a way to make it show uncertainty?

Yes — prompt it: 'For each claim, mark [C] confident, [U] uncertain.' It still won't be perfect, but you'll know where to look. This verification habit is taught as a core discipline in all our courses.