AEO Freshness Decay Tracker

Source: github.com/psyduckler/aeo-skills Part of: AEO Skills Suite

Monitor when your content exits the recurring retrieval set — and catch decay before you disappear.

Background

Being cited once doesn't matter. Influence compounds through repeated inclusion in the recurring retrieval set — the group of sources Gemini consistently draws from when answering a prompt. But that set isn't static. As content ages, competitors publish fresher alternatives, and the model's search queries evolve, your pages can slip out of the candidate set entirely.

In a search-first system like Gemini, retrievability depends on alignment with the queries the model generates, comprehensive topic coverage, and freshness signals. When any of these degrade, your citation rate decays. This skill tracks that decay over time so you can catch it early and refresh before you lose your position in the recurring retrieval set.

The "long long tail" of Gemini's search behavior means the model generates varied, specific queries — and as those queries evolve, previously-cited content may no longer match. Tracking decay reveals when this happens.

Defaults

• Model: gemini-3-flash-preview — the same model powering Google AI Overviews
• Samples: 20 runs per prompt — captures probabilistic citation patterns
• Data format: Append-only JSON — never overwrites historical data

Requirements

• Gemini API key (free from aistudio.google.com) — set as GEMINI_API_KEY env var
• Python 3.9+
• No pip dependencies (stdlib only)

Usage

Scan — Record current citation rates

# Scan prompts and save to data file
GEMINI_API_KEY="$GEMINI_API_KEY" python3 scripts/track.py scan \
    --domain acme.com \
    --prompts "best content optimization tools" "SEO content software" \
    --data-file freshness.json

# With prompts from a file
GEMINI_API_KEY="$GEMINI_API_KEY" python3 scripts/track.py scan \
    --domain acme.com \
    --prompts-file prompts.txt \
    --data-file freshness.json

Report — Analyze decay patterns

# Text report
python3 scripts/track.py report --data-file freshness.json

# JSON report
python3 scripts/track.py report --data-file freshness.json --output json

Run from the skill directory. Resolve scripts/track.py relative to this SKILL.md.

Options

Scan Options

Report Options

Output

Scan Output

Quick summary after each scan:

Freshness Decay Tracker — Scan
Domain: acme.com | Prompts: 2 | Model: gemini-3-flash-preview

Scan complete: 2 prompts × 20 runs
Saved to freshness.json (scan #3)

Quick Results (acme.com):
  "best content optimization tools":
    Mention: 85% | Citation: 80%
  "SEO content software":
    Mention: 60% | Citation: 55%

Report Output

Freshness Decay Tracker — Report
Data: freshness.json (5 scans)
Period: 2025-01-15 → 2025-03-15
Domain(s): acme.com
======================================================================

REFRESH URGENCY SUMMARY:
  🔴 HIGH:   1 prompts — content losing citations fast
  🟡 MEDIUM: 1 prompts — gradual decline
  Declining: 2 of 4 prompts

PROMPT-BY-PROMPT ANALYSIS:
======================================================================

🔴 "best content optimization tools"
  Urgency: HIGH | Trend: rapid_decline
  Citation rate dropped from 80% to 35% over 8 weeks.
  At this rate, citations may reach 0 in ~6 weeks.
  Timeline:
    2025-01-15: citation 80% | mention 85%
    2025-02-01: citation 70% | mention 80%
    2025-02-15: citation 55% | mention 65%
    2025-03-01: citation 40% | mention 50%
    2025-03-15: citation 35% | mention 45%
  Cited URLs:
    https://acme.com/blog/content-optimization-tools
      Counts across scans: 14 → 12 → 9 → 6 → 5

🟢 "SEO content software"
  Urgency: LOW | Trend: stable
  Citation rate stable at 55% (was 60%).
  ...

PAGES MOST IN NEED OF REFRESH:
======================================================================
  🔴 https://acme.com/blog/content-optimization-tools
    Prompt: "best content optimization tools" | Change: -45%

Data File Format

Append-only JSON. Each scan adds an entry, never modifying previous data:

{
  "scans": [
    {
      "timestamp": "2025-01-15T14:30:00+00:00",
      "domain": "acme.com",
      "model": "gemini-3-flash-preview",
      "runs_per_prompt": 20,
      "results": {
        "best content optimization tools": {
          "successful_runs": 20,
          "mention_rate": 85,
          "mention_count": 17,
          "citation_rate": 80,
          "citation_count": 16,
          "cited_urls": {
            "https://acme.com/blog/content-optimization-tools": 14
          }
        }
      }
    }
  ]
}

Refresh Urgency Levels

How It Works

Scan

1. Runs each prompt 20 times against Gemini with Google Search grounding
2. Records mention rate (brand in text), citation rate (domain in sources), and which URLs were cited
3. Appends scan data with timestamp to the JSON file — never overwrites

Report

1. Reads historical scan data
2. For each prompt, builds a time series of citation rates
3. Detects trends: stable, growing, declining, rapid decline
4. Estimates weeks until zero citations (for declining prompts)
5. Ranks prompts by refresh urgency
6. Lists specific URLs that need refreshing

Tips

• Run scans weekly for active prompts — this gives the report enough data points to detect meaningful trends
• Set up a cron job or heartbeat task to automate weekly scans
• Focus refresh efforts on HIGH urgency prompts first — those pages are actively losing their position in the candidate set
• After refreshing content, continue scanning to verify citations recover (expect 2–4 weeks for re-indexing)
• Pair with aeo-source-authority-profiler to understand what the new top-cited pages look like (your blueprint may have shifted)
• Pair with aeo-content-free to refresh decaying content using the latest competitive intelligence
• Use the same prompts file across scans for consistent tracking

References

Notes

• Gemini API key stored in macOS Keychain under google-api-key
• Data file is created automatically on first scan
• Append-only: historical data is never deleted or modified
• Prompts file supports comments (#) and blank lines
• Retries API calls up to 5 times with exponential backoff
• Need at least 2 scans to detect any trends; 4+ scans for reliable decay detection