A developer is designing trigger phrases for a 'Check Order Status' topic in Copilot Studio. Which set of trigger phrases best reflects recommended practice for maximizing topic recognition accuracy?

Copilot Studio uses a natural language understanding (NLU) model to match user utterances to topic trigger phrases. The model generalizes from the examples provided — the quality and diversity of trigger phrases directly determines how well the topic activates for real user input.

Option B is correct. The recommended practice is to provide 5–10 diverse trigger phrases that cover different ways users might express the same intent. 'Where is my order', 'track my package', 'I need order status', 'has my shipment been dispatched' — each captures a distinct linguistic pattern. The NLU model uses these as training signals to recognize the intent across even more variations it has never seen.

Option A is incorrect. A single trigger phrase gives the NLU model almost no generalization signal. If a user phrases their question even slightly differently ('where is my package'), the topic may not fire and the agent falls to the Fallback topic.

Option C is incorrect as a practice recommendation. Copying phrases from external sources without understanding your own users' language patterns may not reflect how your users actually speak. Trigger phrase design should be based on real user utterances from your target audience.

Option D is incorrect. Long, highly specific phrases are rarely how users actually type in a chat interface. Users tend to use short, natural expressions. Very long trigger phrases also reduce the NLU's ability to generalize to shorter or differently-worded inputs.