An assistant needs explicit behaviour for missing information, restricted access and unsupported tasks. Fallback is part of the product experience, not merely an error message.
Distinguish the reason
A fictional technical-documentation assistant may face a question outside its corpus, a private document or an unauthorised action. Reusing the same “I do not know” response misses the different next steps.
| Cause | Product response |
|---|---|
| Missing source | Explain the limit and request useful context |
| Conflicting sources | Escalate the contradiction to an owner |
| Restricted access | Preserve the boundary without exposing content |
| Provider unavailable | Offer ordinary search or a later retry |
Evaluate useful non-answers
Include cases with no valid answer in the evaluation set. Check that the system does not invent a procedure simply to fill the response area. For partial answers, separate supported information from unresolved details. A decorative citation must not imply support for an unrelated conclusion.
Should the interface display confidence percentages?
Only when their meaning and calibration are understood. A number can imply unjustified precision. A concrete explanation such as missing documentation, an outdated version or an absent field may support a better decision. Provide a real next step rather than a refusal that strands the user.
When the user's language differs from the source
If the answer translates a source, keep the source identifiable and make material ambiguity visible. For a specialised term without a reliable equivalent, preserve the original wording or request clarification. Fluency should not conceal uncertainty about the underlying meaning.
From this guide to your project
AI ↗
We start with a task and a quality criterion. AI belongs where it meaningfully improves the service.
Tell us about your ideaExamples are educational and do not describe client results. Content prepared with AI assistance; see our editorial policy. Our method ↗