Add a prompt node to an agent flow

The prompt node makes a single AI call from inside an agent flow. Provide an instruction, optionally pass dynamic content from earlier steps, and the model returns text or structured output that downstream actions consume. Use the prompt node when the step needs the model to transform or generate text, without tool orchestration or knowledge retrieval.

By using the prompt node, you can:

• Send a prompt instruction to a model and use the response in later steps.
• Insert dynamic content from earlier steps directly into the prompt template.
• Select the model that powers the call.
• Choose how the response is shaped, including text or structured output.

Add a prompt node

1. In Copilot Studio, go to Flows and open an existing agent flow, or create a new one.

   • New agent flow: You land on the designer to configure a trigger.
   • Existing agent flow: Open the agent flow and go to the Build tab.

2. Select Add a new action in the step where you want to add the prompt node.

3. In the AI capabilities section of the action picker, select Run a prompt. A Run a prompt node is added to the canvas with a prompt creation window open.

4. Create your prompt, whether starting with an empty prompt, picking a template, or describing the prompt you want to create with natural language. Craft instructions, add knowledge resources, select the language model you want to use, and choose the output shape that best fits your scenario.

5. When you're done, select Save to add the prompt node to your flow.

The sections that follow go into more detail on how to create a prompt in the prompt node, how to choose the output shape, and how to use the prompt response in your agent flow. If you're new to building with prompts in Copilot Studio, we recommend you read through these sections to get the most out of the prompt node.

Write the prompt instruction

In the Instructions field, write the prompt instruction the node runs every time the agent flow reaches this step. Use the dynamic content picker to insert tokens from earlier steps so the template fills in with real run-time data.

For example, in an agent flow that triggers when a Forms response is submitted, the instructions might be Rewrite the following customer feedback into a one-line summary: followed by the Comments token from the form.

Be specific about the task and the format of the response. If you need a list, ask for a list. If you need a one-paragraph summary, say so.

To pick the model that powers the call, use the model dropdown in the upper right of the Instructions box. Select a more capable model when the task needs careful interpretation or longer responses. Select a faster model when the task is simple and runs at high volume.

Choose the output shape

Use the Output dropdown to control the shape of what the prompt node returns. The shape determines how downstream agent flow steps consume the result.

Use the prompt response in your agent flow

When the prompt node finishes, its result is exposed as dynamic content for any downstream step.

1. Select the next action where you want to use the result (for example, Send an email or Update a row).

2. Use / and select Insert dynamic content to open the dynamic content picker on the field you want to fill.

3. Select the output from the earlier prompt node step.

Automation scenarios

The prompt node is at its best as one focused step in a longer agent flow. Earlier steps gather the inputs, the prompt node generates or transforms text, and later steps push the result into the systems people use.

Translate inbound customer messages

An agent flow triggers when an email arrives in a global support inbox. An earlier step extracts the email body; the prompt node sends Translate the following customer message into English. Preserve product names and reference numbers exactly: followed by the Body token from the trigger. Later steps route the translated message to the right regional team and store the original alongside the translation in Dataverse.

Extract structured fields from free-form support tickets

An agent flow triggers when a support ticket is created in Dynamics 365 with a free-form description. The prompt node, with Output set to JSON, sends Extract the following fields from this customer description: urgency (low|medium|high), category (billing|technical|account), and primary intent (one short phrase). followed by the Description token. Later steps route the ticket based on category, set the priority based on urgency, and prefill the Primary intent field on the case record.

Generate a clean lead summary from raw call notes

An agent flow triggers when a sales rep saves call notes to a Dynamics 365 lead. The prompt node sends Rewrite the following raw call notes into a three-bullet executive summary, focusing on the customer's stated problem, the timeline they mentioned, and any next steps: followed by the Notes token from the trigger. Later steps update the lead's Description field with the summary and post it as a comment on the related opportunity.

Frequently asked questions

When should I use a prompt node vs. an agent node?

Both nodes call AI from an agent flow, but they're built for different jobs.

Use the prompt node when you just need the model to transform or generate text, like rewriting, summarizing, translating, or extracting fields. Use the agent node when the step needs reasoning, tool orchestration, or grounded knowledge.

The prompt node is faster and simpler than the agent node. There's no agent to configure, no tools to attach, and no knowledge sources to wire up. Just a prompt instruction and a model.

Does the prompt node have access to my data?

The prompt node only sees what's in the Instructions field, including any dynamic content tokens you insert. It doesn't browse files, search SharePoint, or read mail on its own. If the model needs to ground its response in external data, fetch the data with an earlier step and pass it into the template, or use an agent node instead.

Related content

• Add an agent node to a workflow
• Add a Microsoft 365 Copilot node to a workflow
• Edit and manage your workflow in the designer
• Use prompts to make your agent or agent flow perform specific tasks
• Agent flows and workflows overview