# How do you stop AI customer support from giving a confident answer based on stale information? — retained record

> Retained reddit record from reddit:smallbusiness.

Canonical: https://highsignal.app/data/records/9ce17787ba2fc57f

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:smallbusiness)  published 2026-09-27 09:56 ingested 2026-09-28 02:39

[original source ↗](https://www.reddit.com/r/smallbusiness/comments/1wrga2m/how_do_you_stop_ai_customer_support_from_giving_a/)

## retained text



```
I’ve been looking at how ecommerce teams are using AI for customer support, and one failure mode seems more interesting than the obvious hallucination problem.  
The answer can sound completely reasonable.  
The problem is that the business context behind it may be wrong.  
A return policy changed.  
A promotion expired.  
Delivery conditions changed.  
Product information lives in another system.  
The knowledge base was updated, but the retrieval layer is still pulling the old answer.  
So the AI isn’t necessarily “making something up.”  
It may be confidently answering from incomplete or stale context.  
That seems especially risky for things like returns, refunds, product compatibility, delivery promises or anything that can directly create a cost for the business.  
For people actually running AI support in production: how are you handling this?  
Do you verify policy/source freshness before the answer is sent?  
Use confidence thresholds?  
Force certain categories to a human?  
Or is there another pattern that has worked better?
```



document fetched 2026-09-28 00:31 · canonical https://reddit.com/r/smallbusiness/comments/1wrga2m/how_do_you_stop_ai_customer_support_from_giving_a
