AI
Grounded AI assistants: the difference between helpful and embarrassing
An assistant that invents prices or promises delivery dates is a liability. Grounding, limits and disclosure make it an asset.
By Liaqat Ali Khan · · 3 min read
Website chatbots fail in a predictable way: they answer confidently about things they were never told. The fix is not a cleverer model — it is restricting what the assistant may talk about and being transparent with the visitor.
Our checklist
- Answers are grounded in approved, published content only
- Prices and availability come from the catalogue, never the model
- Clear 'you are talking to an AI' disclosure
- Rate limits and usage caps per visitor
- A human contact path is always one click away
- Provider timeouts degrade to a helpful message, not a spinner