AEO Analytics (Free)

Source: github.com/psyduckler/aeo-skills Part of: AEO Skills Suite — Prompt Research → Content → Analytics

Track whether AI assistants mention and cite your brand — and how that changes over time.

Requirements

• Primary: Gemini API key (free from aistudio.google.com) — enables grounding with source data
• Fallback: web_search only — weaker signal but zero API keys needed
• web_fetch — optional, for deeper analysis of cited pages

Input

• Domain (required) — the brand's website (e.g., tabiji.ai)
• Brand names (required) — names to search for in responses (e.g., ["tabiji", "tabiji.ai"])
• Prompts (required for first scan) — list of target prompts to track. Can come from aeo-prompt-research-free output.
• Data file path (optional) — where to store scan history. Default: aeo-analytics/<domain>.json

Commands

The skill supports three commands:

scan — Run a new visibility scan

Execute all tracked prompts against the AI model and record results.

report — Generate a visibility report

Analyze accumulated scan data and produce a formatted report.

add-prompts / remove-prompts — Manage tracked prompts

Add or remove prompts from the tracking list.

Scan Workflow

Step 1: Load or Initialize Data

Check if a data file exists for this domain. If yes, load it. If no, create a new one. See references/data-schema.md for the full JSON schema.

Step 2: Run Prompts

For each tracked prompt:

Method A — Gemini API with grounding (preferred): See references/gemini-grounding.md for API details.

1. Send prompt to Gemini API with googleSearch tool enabled

2. From the response, extract:

   • Response text — the AI's answer
   • Grounding chunks — the web sources cited (URLs + titles)
   • Web search queries — what the AI searched for

3. Analyze the response:

   • Mentioned? — Search response text for brand names (case-insensitive, word-boundary match)
   • Mention excerpt — Extract the sentence(s) containing the brand name
   • Cited? — Check if brand's domain appears in any grounding chunk URI
   • Cited URLs — List the specific brand URLs cited
   • Sentiment — Classify the mention context as positive/neutral/negative
   • Competitors — Extract other brand names and domains from response + citations

Method B — Web search fallback (if no Gemini API key):

1. web_search the exact prompt text
2. Check if brand's domain appears in search results
3. Record as "web-proxy" method (less direct than grounding)

Step 3: Save Results

Append the scan results to the data file. Never overwrite previous scans — history is the whole point.