Running Your First Agent

Key Concepts

Consumer Agents vs. Brand Agents: Consumer agents (ChatGPT, Perplexity, Claude, Gemini) research, evaluate, and recommend on behalf of the consumer. Brand agents are the marketing workflows you build to ensure those consumer agents find you, understand your differentiation, and surface your content. Both sides meet on the open web.

Agent: An automated workflow inside Profound. A sequence of stages (called nodes) that accomplish a specific job: create a content brief, generate a brand health report, find linking opportunities. Each node has inputs and outputs, and data flows between them through named variables.

Node: A single step in an agent. Each node does one thing - pulls data, scrapes a page, reasons over information, sends a message. The chain of nodes does the whole job.

Templates: Pre-built agents for common jobs. A template is a hypothesis about what steps matter for a given task. Running one tests that hypothesis against your data. Modifying it tests whether a different approach works better. Templates are starting points, not finished products.

Quick Reference

Five Node Types

• Data nodes - Pull your Profound metrics (visibility scores, citation share, top-cited pages) into the agent as inputs
• Research nodes - Go external: scrape a webpage, search Google, pull People Also Ask questions, run deep research queries
• LLM nodes - The thinking layer. Give them a prompt plus data from upstream nodes. They reason, synthesize, or generate. This node determines what the output actually looks like.
• Integration nodes - Deliver results somewhere useful: Slack message, WordPress publish, file upload to a knowledge base
• Logic nodes - Conditional branching. If visibility drops below a threshold, take one path; if stable, take another. Makes the agent adaptive.

How Data Flows Between Nodes

Every node output is labeled with a variable name. Downstream nodes reference upstream data by selecting that variable from a dropdown or typing / and the variable name. This is how context passes through the chain - the variable system is the connective tissue of the entire agent.

Running a Template

1. Go to Agents and open the template library
2. Select a template and click Customize Template - this opens a new agent in your workspace with all nodes prewired
3. The agent canvas shows every node as a block, with lines showing data flow between them
4. Enter the required inputs (varies by template, e.g., article URL and target prompt)
5. Click Run Test
6. Watch the agent execute node by node - each step lights up as it runs, and you can inspect the output of any step

The Iteration Loop

1. Run the template
2. Read the output
3. Identify where you'd intervene (which node produced something you'd change)
4. Modify that node (change the prompt, adjust inputs, add structured output fields)
5. Run again and compare

You don't need to run the full agent every time. Use the play button on individual nodes to test just that node with the same upstream data. Change the prompt, test the node, read the output, adjust. Once you're happy, run the full chain.

LLM Node Details

The LLM node is where you control what the agent does with gathered information. Key configuration:

• Inputs - Multiple upstream variables can be fed in as context (scraped pages, research results, scorecard data)
• Prompt instructions - Tell the LLM what to do with all that context
• Structured outputs - Enable JSON structure with defined fields (e.g., topic, target_audience, key_angles, competitive_gaps) for consistent, parsable output instead of freeform text
• Creativity slider - Balance between predictable and creative output

Adding Integrations

You can add integration nodes to any agent. For example, adding a Slack node requires selecting the workspace, channel, and message variable. The output from an upstream LLM node becomes the message content, delivering results directly to your team.