Building a Weekly AEO Briefing Agent

Key Concepts

Profound Data Nodes: Agent nodes that pull your live AEO metrics (visibility scores, citation share, citation pages) directly from Answer Engine Insights into an agent workflow. These are the same numbers you see in the dashboard, but now they're inputs to an automated chain rather than something you read manually.

The Detection-Analysis-Delivery-Update Pattern: A reusable agent architecture: pull a stored baseline from the knowledge base, pull live data, compare them in an LLM, compose and deliver the output, then overwrite the baseline for next time. This is the same pattern used in the competitive pricing agent - different data source, same structure.

Quick Reference

Preparing the Baseline

Before building the agent, create its first memory snapshot:

1. Go to the Prompts page with your desired time period selected
2. Click Copy as CSV (do not try to manually copy table data)
3. Navigate to Knowledge Bases, create a new knowledge base (e.g., "Performance")
4. Paste the CSV data as a text file named something like baseline.txt

This becomes "last week" for the agent's first run. Every subsequent run overwrites it automatically.

Configuring Profound Data Nodes

Visibility Score node:

• Set date interval to week (one data point, not daily breakdowns) for clean week-over-week comparison
• Select your platforms (the same Big 4 or whatever you use)
• Select all relevant topics
• Set the limit high enough to capture all permutations (e.g., 500 if you have many topic/platform combinations)

Citation Domains node:

• Same settings: weekly interval, same platform filter, all topics, limit at 500
• Critical: Add a filter for prompt type and set it to visibility. Always do this when pulling AEO data. Sentiment prompts mention your brand by name and will skew citation data. Visibility prompts are the real prompts from Answer Engine Insights.

The Agent Chain

Node 1 - Knowledge Base Read: Pull baseline.txt from your performance knowledge base. Output label: baseline

Node 2 - Visibility Score (Profound Data node): Weekly interval, filtered to your platforms and topics. Captures current visibility across all topic/platform combinations.

Node 3 - Citation Domains (Profound Data node): Same filters. Captures current citation share data.

Node 4 - Diagnosis (Prompt LLM node): Receives three inputs: the baseline, this week's visibility scores, and this week's citation domains. Prompt instructs the LLM to compare week over week across visibility score and citation share, connect the dots between them, and diagnose what moved. No structured output needed here - this is a qualitative sense check, not a data pipeline. No recommendations, just diagnosis. Use a strong reasoning model (e.g., GPT-5) at low temperature.

Node 5 - Slack Writer (Prompt LLM node): Receives the diagnosis. Composes a single Slack message a marketer can read in 30 seconds. Should answer: did performance change, in which direction, driven by what, and is anything worth investigating? Use a writing-oriented model (e.g., Opus) with slightly higher creativity.

Node 6 - Slack Integration node: Sends the Slack message to your designated channel.

Node 7 - Upload to Knowledge Base: Takes this week's live visibility and citation data and overwrites the baseline file. Next week's run compares against this week's data.

Prompt Type Filter: Why It Matters

Profound runs two types of prompts:

• Visibility prompts - The real prompts from Answer Engine Insights (e.g., "best AI search optimization software")
• Sentiment prompts - Prompts that directly ask engines to evaluate a brand on a topic (e.g., "evaluate HubSpot on CRM for startups")

Sentiment prompts will always mention your brand, which inflates citation data. When pulling AEO data into agents, always filter to visibility prompts unless you're specifically analyzing sentiment.

After You Build It

• Run it weekly on a schedule, or manually each Monday
• The weeks you go deep manually: The brief gives you a head start with context already surfaced
• The weeks you can't: You still know what happened - you're covered
• Refine over time: Adjust the diagnosis prompt as you learn what's most useful to see in the brief