The fastest way to lose a customer's trust is a confident, wrong answer. One made-up refund policy and they'll never believe the bot again.
Grounding is the fix
A grounded assistant only answers from retrieved content and cites the source for each claim. The visitor can verify it, and you can audit it.
What good behavior looks like
- Every factual claim links back to a source: "Refunds take 5 days [1]."
- When confidence is low, it declines instead of guessing.
- It never invents prices, policies, or features.
Why this also helps you
Citations make the bot's answers traceable. If something's wrong, you fix the source — and the bot is instantly correct everywhere. Hallucination has no such fix; you're just playing whack-a-mole with prompts.
A bot that says "I'm not sure" is worth more than one that's confidently wrong.
Grounding plus a clean refusal path is the difference between a demo and something you'd actually put in front of customers.