An executive asks: "Why does the AI make things up, and how do I know it won't do it here?" Answer as you would to them.

Strong answer

"These models predict likely text, so when they don't have the right information they can produce something that sounds right but isn't. We design around that in three ways. First, the system only answers from your documents and shows the source for every answer, so your team can check it in one click. Second, when it can't find a source, it says so instead of guessing. Third, we test it on a set of real questions your experts answered, and we track how often it gets them right before and after launch. You'll get that number every week."

Weak-answer indicators

• Uses jargon like logits or temperature
• Claims it never happens
• No concrete safeguards