A disclaimer is the first message in a chat interface: a short, plain-language notice that aligns user expectations with what an assistant can actually do. It states the task domain, accessible data, limits, and input practices before a failed query turns sound retrieval into an apparent product failure. section 6.0 section 6.1
A production-ready disclaimer should make four points explicit:
- Name the domain the assistant serves, such as company product information. section 6.1
- Identify the data sources it can access and information it cannot see. section 6.1
- State technical limits that affect results, such as retrieving only a small set of records per question. section 6.1
- Model effective queries by asking for full sentences and contrasting them with context-free identifiers. section 6.1
- Warn that responses can hallucinate and should be double-checked before reliance. section 6.1
Product-information assistant
“I search only the Products database,” sets a useful boundary: the assistant may answer product-detail questions but cannot rank items by popularity when popularity data is stored elsewhere. A request such as “What is product 12345?” supplies more context than “12345” alone. section 6.1 section 6.17
Place the notice directly in the chat UI rather than leaving it as a policy elsewhere. Keep it concise enough to read before a first query, then reinforce it constructively when feedback reveals an out-of-scope request or an underspecified prompt. Treat recurring complaints as signals to distinguish missing data access from retrieval or prompt problems. section 6.1 section 6.17 section 6.18
