3.2.3. Prompt Actions in Copilot Studio
Prompt actions allow agents to use AI-powered operations as discrete steps within a topic or flow. Instead of writing complex logic to analyze, summarize, classify, or generate content, you define a prompt that the AI model executes — and the result becomes part of the agent's response or processing pipeline.
Prompt actions bridge the gap between simple topic responses and full custom model development. They let you leverage LLM capabilities for specific tasks without building separate AI pipelines.
Common Prompt Action Use Cases:
| Use Case | Prompt Action Example | Why Not Just a Topic? |
|---|---|---|
| Summarization | "Summarize this support ticket in 2 sentences" | Topic handles routing; prompt action handles the AI task |
| Classification | "Classify this email as: complaint, inquiry, or feedback" | Requires LLM reasoning, not keyword matching |
| Extraction | "Extract the invoice number, date, and total from this text" | Structured data extraction from unstructured input |
| Generation | "Draft a professional response to this customer complaint" | Creative generation with specific constraints |
| Translation | "Translate this response to Spanish, maintaining professional tone" | Language transformation with context preservation |
Designing Effective Prompt Actions:
- Be specific about output format. "Classify this email" is ambiguous. "Respond with exactly one of: COMPLAINT, INQUIRY, FEEDBACK" is precise and parseable.
- Include examples (few-shot). For classification or extraction tasks, 2-3 examples dramatically improve consistency.
- Set constraints. "Use only information from the provided text. Do not add external knowledge" prevents hallucination in extraction tasks.
- Chain prompt actions when tasks are complex. First extract data, then classify, then generate a response based on the classification. Each step has a focused prompt.
Prompt Actions vs. AI Builder Prompts:
Copilot Studio prompt actions and AI Builder prompts serve similar purposes but differ in scope:
- Prompt actions operate within a Copilot Studio agent's conversation flow
- AI Builder prompts can be used across Power Platform (Power Apps, Power Automate) and are managed as reusable components
For solutions that need the same AI operation in multiple places (inside an agent AND in a Power App), design it as an AI Builder prompt and reference it from both locations.
Exam Trap: Don't confuse prompt actions with system prompts. System prompts define the agent's overall behavior and persona. Prompt actions are discrete AI operations executed as steps within a topic or flow. A single agent has one system prompt but can have many prompt actions.
Reflection Question: A customer service agent receives support tickets in multiple languages. It needs to: detect the language, translate to English, classify the issue, and route to the appropriate team. Design this as a sequence of prompt actions.
