AI Messaging Bot — Response & Escalation Rules

Overview

The AI Messaging Bot gives each member conversation exactly one automated attempt to resolve the member's question using the connected knowledge source. If the bot can answer the question from the knowledge source, it replies once with that answer. If it cannot find an answer — or the member signals they need human help — it stops attempting to resolve the issue itself and instead launches the Escalation Reply use case.

Core Behavior Rules

1. One response per resolution attempt

The bot responds to the member one time per inbound message/question using information pulled from the knowledge source, if an answer can be found there.

• Do not send multiple bot replies attempting to re-answer the same question.
• If the first response does not resolve the member's issue, do not have the bot try again — proceed to the Escalation Reply use case (see below).

2. Knowledge source lookup

Before replying, the bot must check the connected knowledge source for an answer.

• If found: Reply once with the relevant information from the knowledge source.
• If not found: Do not guess, improvise, or generate an unsupported answer. Launch the Escalation Reply use case instead.

3. Escalation Reply use case

Launch the existing Escalation Reply use case (configured in the Zendesk AI Agent Messaging Bot builder) whenever:

• The bot cannot find an answer to the member's question in the knowledge source, or
• The member's message indicates they need direct human assistance (see example phrases below).

This article does not redefine the Escalation Reply use case itself — refer to that use case's own configuration in the Messaging Bot builder for what it sends and how it routes. This article only governs when it should be launched.

4. Never suggest contacting Support more than once

The bot must never suggest the member contact Support more than one time, or again, within a conversation.

• If a "contact Support" suggestion has already been made once (by the bot or as part of a prior Escalation Reply), do not repeat it.
• Do not stack multiple "reach out to Support" messages across bot replies.

Example Phrases That Should Launch Escalation Reply

The bot should treat messages like the following as signals to launch the Escalation Reply use case, in addition to any case where no knowledge-source answer exists:

Note: This list is illustrative, not exhaustive. Any similarly phrased request for direct/human help should be treated the same way — err on the side of launching Escalation Reply rather than having the bot attempt another automated reply.

Quick Decision Logic

1. Member sends a message.
2. Bot checks the knowledge source.
   • Answer found → Bot replies once with that answer. Do not follow up with another automated attempt.
   • No answer found → Launch the Escalation Reply use case.
3. Separately, at any point: if the member's message matches one of the example phrases above (e.g., "Please help"), launch the Escalation Reply use case immediately, regardless of knowledge source lookup.
4. In all cases, do not suggest contacting Support more than once total per conversation.

Related Articles

• Escalation Reply use case (internal — link to its configuration in the Messaging Bot builder)
• AI Messaging Bot: Knowledge Source Setup (internal, if applicable)
• AI Messaging Bot: Overview & Access (internal, if applicable)

Publishing Notes

• Source: Internal requirements provided by Chelsea (Black Cap), August 19, 2026
• Existing articles to update: Check whether an "Escalation Reply" use case article already exists to link to directly
• Review needed from: AI Messaging Bot configuration owner / Support Ops lead, to confirm the example phrase list and escalation routing match the live bot config
• Suggested review date: November 2026 (quarterly, or sooner if bot logic changes)