AI wants to give you an answer. When it doesn't have the facts, it can make them up, and it sounds just as sure as when it's right.

How do you prevent AI hallucinations?

Give it the facts and tell it what to do when it's unsure. Three things I do:

  1. Put the facts in files. Your pricing, your vendor list and your services go in a Claude Project, so Claude reads them instead of guessing.
  2. Set a confidence rule. My AI bookkeeper codes a transaction only when it's 90% sure or higher. Below that, it flags it for a person.
  3. Keep a human in the loop. Anything that goes to a customer stays a draft until I've read it.

Do long chats cause more hallucinations?

They can. When a chat runs long, the context window fills up and Claude starts to forget the early parts. Start a fresh chat with a summary when the answers get foggy.