Give a useful answer when the chatbot does not know

The decision this workflow supports

A restaurant chatbot should not fill gaps in its approved information with a plausible guess. A useful unknown-answer response explains the limitation and gathers the smallest amount of context needed for someone to help.

Put the review process in place

Create a fallback that records the question in the visitor’s words and offers the approved staff route. Do not repeatedly rephrase an unsupported answer. Review unknown questions periodically to decide whether the owner should add a new approved FAQ.

Worked example for your team

A fictional visitor asks about a service that is not mentioned in the approved information, then repeats the question in a different way. The assistant should continue to acknowledge the gap and offer the staff route. It should not become more confident merely because the visitor presses for an answer. The staff note should preserve the actual question so the owner can decide whether a new FAQ is needed. A fallback that says only could not help loses useful context and makes the next person repeat the conversation.

Sample wording to approve

“I do not have approved information to answer that accurately. I can pass your question to restaurant staff or share their approved contact route. What would you like the team to clarify?” This wording is a draft for the restaurant owner to approve before use.

Fields for the staff handoff

Staff need unanswered question, relevant date or service, contact preference, proposed FAQ topic, and staff reviewer. Mark uncertain details explicitly and keep the requested action separate from anything restaurant staff have already confirmed.

Acceptance tests before use

An unsupported question triggers the fallback. Repeated asking does not produce a fabricated answer. The original question remains intact in the staff summary. Record the observed result, assign any mismatch, and repeat the affected scenario after correction.

Related guides