Handling AI Responses Outside the Knowledge Base

Last updated: March 23, 2026

When a caller asks your Maple AI something it doesn't have an answer for, the system is designed to handle the situation gracefully — never guessing, never hanging up. This article covers how the AI behaves when it encounters questions outside its configured knowledge, what safeguards are in place to prevent inaccurate responses, and how you can fine-tune fallback behavior to reduce unnecessary call transfers.

What happens when a customer asks something the AI can't answer?

When a caller asks a question that isn't covered in your knowledge base or FAQ configuration, the AI will offer to transfer the call to your designated handoff number (typically your restaurant's main phone line). Your staff member picks up and handles the request directly.

The AI will never hang up on a caller or leave them without a path forward. Common scenarios that trigger a transfer include:

  • Catering requests the AI isn't configured to handle

  • Reservation modifications (e.g., changing a time or party size on an existing booking)

  • Complaints or complex special requests that require human judgment

  • Business policy questions that haven't been added to the knowledge base (e.g., corkage policy, BYOB rules)

Will the AI make up answers if it doesn't know something?

No. Maple's AI is designed to only use the information you've provided in your knowledge base. It does not research or pull information from external sources.

When you add a question-and-answer pair to your knowledge base, that specific answer becomes the only context the AI has for that topic. For example, if you add:

  • Question: "Do you have parking?"

  • Answer: "No"

The AI will respond with "No" and will not supplement it with directions to nearby parking garages or other externally sourced information.

That said, earlier versions of the AI occasionally generated inaccurate information — such as inventing menu items that don't exist or fabricating policies (like a pet-friendly patio policy that was never configured). These issues have been resolved with Maple's next-generation AI model, which has dramatically reduced hallucination and significantly improved accuracy for handling specials, menu items, and general business information. This updated model is enabled on all accounts at no extra charge.

How can I reduce the number of calls that get transferred to my staff?

The most effective way to reduce unnecessary transfers is to add more information to your knowledge base. Every question the AI can confidently answer is one fewer call your staff needs to handle.

Focus on adding answers for:

  • Business policies — corkage fees, BYOB rules, dress code, pet policies, accessibility

  • Frequently asked questions — parking, Wi-Fi, private dining, dietary accommodations (e.g., gluten-free options)

  • Menu availability details — which items are available during which service hours (e.g., "Desserts are only available during dinner service, 4–9 PM, Monday through Saturday")

  • Daily specials — if your specials rotate by day, include the schedule so the AI can accurately describe what's available

The more complete your knowledge base, the fewer edge cases the AI encounters.

Can callers always reach a human if they want to?

Yes. Maple's system is designed so that callers can transfer to a human at any point during the call — not just when the AI can't answer a question. Depending on your configuration, callers may hear options at the start of the call such as:

  1. Talk to the AI to place an order or ask questions

  2. Press a number to transfer to a staff member at any time

  3. Press a number to receive a booking link via SMS

These options remain available throughout the entire call, so a customer is never locked into the AI interaction.

What types of requests will always require a human?

Some tasks are outside the AI's current capabilities and will always result in a transfer:

  • Modifying existing reservations (time changes, party size adjustments, cancellations)

  • Handling complaints or sensitive customer service issues

  • Complex catering inquiries requiring custom quotes or coordination

  • Requests that require accessing third-party systems the AI isn't integrated with

For these scenarios, the AI will politely let the caller know it can't assist directly and offer to connect them with your team.

How do I make sure the AI only answers based on what I've configured?

This is the default behavior. Whatever you add to your knowledge base is the only context the AI uses to respond. It will not independently research answers or pull from outside sources.

To verify your configuration is working as expected, you can place test calls to your AI line and ask questions you've added to the knowledge base. Confirm the AI responds with your exact configured answers and doesn't embellish or add unsourced details.

What if the AI gives an incorrect answer?

If you notice the AI providing inaccurate information — such as wrong pricing, non-existent menu items, or fabricated policies — take these steps:

  1. Review your knowledge base to ensure the correct information is entered and up to date

  2. Check for outdated entries that may conflict with current offerings (e.g., old specials or discontinued items)

  3. Contact Maple support if the issue persists — the team can review call logs and adjust your AI configuration

Maple's next-generation AI model has resolved the vast majority of hallucination issues, but keeping your knowledge base accurate and current is the best way to ensure reliable responses.