Topics and Triggers

Overview

An agent that can only handle one type of question is not very useful. Real-world agents need to handle dozens or even hundreds of different requests, from password resets to leave policy questions to facility issue reports. The challenge is making sure each request lands in the right place.

Topics and triggers solve this problem. Topics are the self-contained conversation units that handle specific requests, and triggers are the routing system that matches what a user says to the right topic. Getting these right is the foundation of a capable, reliable agent.

This video covers how to create topics, write effective triggers, build conversation flows, and connect topics together through handoffs.

What You’ll Learn

• Topics: What they are and how to structure them for your agent
• Trigger phrases: How to write triggers that accurately match user intent
• Conversation flow: How to build branching conversations with conditions and variables
• Handoffs: How to redirect between topics and pass context seamlessly

Script

Hook: Structure your agent’s conversations

Your agent needs to handle dozens of different questions. A user might ask about office hours, another about resetting a password, and another about requesting new equipment. How does your agent know which conversation to start?

Topics and triggers are the answer. Topics are the individual conversations your agent can have, and triggers are the mechanism that detects what a user is asking and routes them to the correct topic. Together, they form the routing system that makes your agent intelligent.

In the next ten minutes, you will learn how to create, organize, and connect topics so your agent handles any request that comes its way.

Understanding topics

A topic in Copilot Studio is a self-contained conversation unit. Think of it as one complete interaction path. A user asks a question, the agent recognizes what they need, and the topic guides the conversation from start to finish.

Every topic has three core components. First, trigger phrases define how users activate the topic. These are the words and sentences that tell Copilot Studio this is the topic the user wants. Second, conversation nodes define the flow of the interaction, including what the agent says, what questions it asks, and what logic it follows. Third, variables store data collected during the conversation, like a user’s name, their issue category, or a date they selected.

Copilot Studio includes several system topics that come prebuilt with every agent. The Greeting topic fires when a user first opens the conversation. The Escalate topic handles requests to speak with a human. The Fallback topic catches messages the agent does not understand. And the End of Conversation topic wraps things up cleanly. You can customize all of these, but they provide a solid foundation out of the box.

Custom topics are where your agent’s real value lives. For a government agency help desk agent, you might have custom topics for badge access requests, leave request information, IT support questions, facility maintenance reports, and travel authorization inquiries. Each topic handles one specific type of request.

Creating topics

To create a new topic, navigate to the Topics section in the left panel and click “Add a topic,” then select “From blank.” Start with a clear, descriptive name. Use consistent naming conventions across your agent. For example, “Leave Request - Information” and “Leave Request - Submit” are clearer than “Leave Stuff” or “Topic 7.”

Next, add trigger phrases. This is the most important step in topic creation. Trigger phrases tell Copilot Studio what user messages should activate this topic. Add five to ten variations that cover how real users would phrase their request.

For a “Leave Request Information” topic, you might add: “How do I request leave?” and “What is the leave policy?” and “I need to take time off” and “How many leave days do I have?” and “PTO request” and “Annual leave” and “How do I submit a leave request?” Include both formal language like “What is the procedure for requesting annual leave?” and informal language like “How do I take a day off?” Add abbreviations and acronyms your agency commonly uses.

Now build the conversation flow. Add a message node for the agent’s initial response. Add a question node if you need to collect information from the user, such as “What type of leave are you asking about?” with options for annual leave, sick leave, and compensatory time. Add condition nodes to branch the conversation based on the user’s answer. Each branch can provide different information relevant to that leave type.

For government environments, think about the specific terminology your agency uses. Federal agencies have specific leave categories, acronyms, and procedures that your trigger phrases and responses should reflect.

Configuring triggers

Understanding how triggers work under the hood helps you write better ones. When a user sends a message, Copilot Studio’s natural language understanding engine compares that message against the trigger phrases of every topic in your agent. It calculates a confidence score for each potential match and selects the topic with the highest score, as long as that score exceeds the minimum threshold.

This means your trigger phrases do not need to be exact matches. The NLU engine understands intent. If your trigger phrase is “How do I reset my password?” and a user types “I need to change my password,” the engine recognizes the similar intent and triggers the correct topic.

To write effective trigger phrases, cover the full range of how users express their intent. Use both complete sentences and short fragments. “How do I reset my password?” and “password reset” and “forgot password” and “can’t log in” all express the same intent in different ways.

Watch out for trigger conflicts. If your “Password Reset” topic has the trigger “I need help” and your “General Support” topic also has “I need help,” Copilot Studio may not consistently pick the right one. Make your triggers specific enough that each topic’s phrases are distinct. Test frequently to catch these conflicts early.

For advanced scenarios, topics can also be triggered by events rather than user messages. A topic can fire when the conversation starts, when a user has been inactive for a period, or when another topic explicitly redirects to it. These event-based triggers are useful for proactive agent behaviors like greeting messages or timeout warnings.

A government-specific tip: include agency acronyms and jargon in your trigger phrases. If your users say “CAC issue” instead of “Common Access Card problem,” make sure “CAC issue” is a trigger phrase. Listen to how your actual users talk and mirror that language.

Managing topic flow

Once a topic is triggered, the conversation flow determines the user’s experience. Simple topics might have a single message response. Complex topics might collect multiple pieces of information, branch based on conditions, and call external services.

For linear conversations, the flow is straightforward. The agent asks a question, the user answers, and the agent provides a response. For branching conversations, you use condition nodes to create different paths. If a user selects “annual leave,” show information about annual leave. If they select “sick leave,” show different information.

Question nodes are your primary tool for gathering information. You can ask open-ended questions where the user types freely, or multiple-choice questions with predefined options. Store each answer in a variable so you can reference it later in the conversation or pass it to another topic.

When designing flows, follow these best practices. Keep each topic focused on one task. If a topic is getting complicated with deeply nested branches, consider splitting it into multiple topics connected by handoffs. Always provide an exit path so users are never stuck. And limit conversation depth to avoid making users answer too many questions before getting help.

End topics cleanly. Confirm that the user’s question was answered with a message like “Did that answer your question?” Offer to help with something else. If the user says yes, redirect to a general menu topic. If they say no, end the conversation gracefully.

Here is a government scenario. You are building a topic for reporting a facility issue. The topic asks the user to select a category: plumbing, electrical, HVAC, or other. Based on the category, it asks for a description and the building location. It confirms the details back to the user and provides a reference number. This multi-step flow uses question nodes, condition branches, and message confirmations to create a complete, professional interaction.

Topic handoffs

Sometimes one topic needs to transfer control to another. This is called a topic handoff, and it is a powerful pattern for building modular, maintainable agents.

To redirect from one topic to another, add a “Redirect to another topic” action node in your conversation flow. Select the target topic, and the conversation seamlessly continues in the new topic. The user does not see a disruption. They simply experience a continuous conversation.

You can pass variables between topics during a handoff. If the first topic collected the user’s name, you can pass that variable to the next topic so it does not have to ask again. This makes conversations feel natural and avoids repetitive questions.

Common handoff patterns in government agents include a general inquiry topic that asks “What do you need help with?” and routes to specialized topics for HR, IT, or facilities based on the user’s selection. Any topic can include a handoff to the Escalate system topic when the user requests a human agent. And FAQ topics can redirect to detailed procedure topics when the user needs step-by-step instructions instead of a quick answer.

Keep handoffs seamless. Avoid messages like “Transferring you to another topic.” The user should not know or care about your topic structure. They should just experience a smooth, helpful conversation.

Close: Topics as building blocks

Let us recap. Topics are the building blocks of your agent. Each one handles a specific type of conversation. Triggers are the routing system that matches user messages to the right topic. And conversation flows define the experience within each topic, from simple answers to multi-step interactions with branching logic.

Well-structured topics make your agent easier to maintain, easier to test, and easier to scale. When you need to handle a new type of request, you add a new topic. When an existing response needs updating, you edit one topic without affecting the rest.

Here are your next steps. Review your existing topics to make sure trigger phrases cover common variations. Add condition branches for scenarios where users need different information. Test topic handoffs end to end to verify that variables pass correctly and the user experience is seamless.

A well-organized set of topics is the difference between an agent users love and one they abandon. Invest the time in getting them right.

Sources & References

• Create and edit topics in Copilot Studio — Official documentation for creating topics, adding trigger phrases, and building conversation flows
• Configure topic triggers — Detailed guide on trigger configuration, NLU matching, and trigger phrase best practices
• Microsoft Copilot Studio documentation — Comprehensive documentation hub for all Copilot Studio capabilities