On this page
Why do AI hallucinations happen?
A large language model generates text by predicting what is likely to come next. It’s very good at producing fluent, reasonable-sounding language, but fluency isn’t the same as truth. When the model lacks the right information, it may fill the gap with something that merely sounds right.
What does it look like in ecommerce?
- Stating a returns window your store doesn’t offer.
- Promising a delivery date the carrier can’t meet.
- Describing an ingredient or feature a product doesn’t have.
- Confirming a discount that doesn’t exist.
Each creates real costs: disappointed customers, refunds, disputes and damaged trust.
How can you reduce hallucinations?
- Ground answers in trusted sources using retrieval-augmented generation.
- Keep source content accurate: policies, product data and FAQs.
- Set boundaries on topics and promises the AI must never make.
- Prefer honesty: “I don’t know, let me get a person” beats a guess.
- Review transcripts regularly, especially early on.
How Yep AI uses it
Yep AI puts guardrails on every reply: answers are grounded in your catalogue, policies and order data, you set what AI Agents can and can’t say, and uncertain or sensitive cases go to your team with the full conversation. Read AI Agent Security.
