A fallback response is what a chatbot says when it cannot confidently handle a user’s message. Rather than pretending to understand, it acknowledges the gap and gives the user a way to continue. section 12.2.1 section 12.1.4
An intent recognizer scores the intents a bot has been programmed to recognize. If every score falls below a chosen confidence threshold, the bot can treat the message as unrecognized and use its fallback response. In a conversation diagram, the special `__default__` label specifies the fallback intent for each bot message that requires a user reply. section 12.2.1 section 12.1.3
A useful fallback does more than say “I don’t understand.” It can:
- Offer clear choices so the user can try again. section 12.1.4
- Suggest another function the bot can perform. section 12.1.4
- Offer a connection to a human representative. section 12.1.4
Fallbacks matter for generative chatbots, too. If a generated claim is absent from the bot’s knowledge base, it may be off topic or incorrect; relying on a fallback is safer than presenting it as fact. Monitor when fallbacks occur, much as you would log application errors and warnings, to find gaps in the conversation design and mistakes in generated answers. section 12.4
