Detailed Explanation
LLM hallucinations occur when a generative model fills knowledge gaps with plausible-sounding but fictitious details. In customer service, this can cause the AI to promise unauthorized refunds or fabricate pricing. Grounded-answer architecture combines strict system prompts, factual verification filters, and confidence thresholds to minimise ungrounded statements.
How It Works in CustomerGPT
CustomerGPT restricts answers to your approved documentation, which sharply reduces false promises. When the documentation has no answer, it says so and hands off to a person.
Related Glossary Terms
Frequently Asked Questions
Will CustomerGPT ever give away unauthorized discount codes?
It is built not to. CustomerGPT is instructed to refuse promo codes, discounts, or policies that do not explicitly exist in your indexed documentation, and to hand off to a person instead.
Deploy Grounded RAG AI for Your Support
Get started in under 2 minutes with our 14-day free trial.
Start Free Trial →