Skill: AEO Citation Tracker — 4-Engine Brand Visibility

Universal brand-citation tracker. Given a brand and its competitors, query 4 major AI engines on a curated query set, record who got cited, and output share-of-voice over time.

When to invoke

• Weekly cadence (Wednesday) for any brand doing AEO seriously
• Before launching: baseline measurement
• After major content/PR push: did the campaign move citations?
• When user says "is ChatGPT recommending us?", "Perplexity never mentions us", "why is competitor X always cited first?"

Inputs

Query design (if not user-supplied)

Generate 12-20 queries across 4 buckets. Each query is the kind of long-tail prompt (50-80 words) AI engines actually receive — NOT short Google keywords.

If geo-query-finder skill is available and user has more than 5 minutes, run it first to source higher-quality long-tail queries grounded in real search intent.

Execution: query 4 engines

For each query, ask each engine and record:

If gateway access not available, fall back to: WebFetch against the engine's public web UI (slower, lower fidelity), or have the user paste responses into a structured form.

Per-result extraction

For each (engine, query) response, parse and record:

{
  "engine": "chatgpt|claude|perplexity|gemini",
  "query": "<verbatim query>",
  "timestamp": "<ISO 8601>",
  "brand_mentioned": true|false,
  "brand_rank": 1|2|3|null,            // null if not mentioned; 1 = first recommendation
  "brand_context": "<30-100 word snippet showing how it was mentioned, verbatim>",
  "competitors_mentioned": [
    { "name": "BrandX", "rank": 1 },
    { "name": "BrandY", "rank": 2 }
  ],
  "citations": [                        // if engine returns source URLs
    { "url": "...", "title": "...", "is_brand_owned": true|false }
  ],
  "response_full": "<full response text, for audit>"
}

Aggregation: share-of-voice

After all (engine, query) runs, compute:

{
  "date": "YYYY-MM-DD",
  "brand": "Acme Corp",
  "total_queries": 16,
  "total_runs": 64,                                // 16 queries × 4 engines
  "brand_mention_rate": 0.42,                       // 27/64
  "brand_avg_rank_when_mentioned": 2.3,
  "by_engine": {
    "chatgpt":    { "mention_rate": 0.56, "avg_rank": 1.8 },
    "claude":     { "mention_rate": 0.50, "avg_rank": 2.1 },
    "perplexity": { "mention_rate": 0.31, "avg_rank": 2.9 },
    "gemini":     { "mention_rate": 0.31, "avg_rank": 2.5 }
  },
  "by_bucket": {
    "direct":     { "mention_rate": 1.00, "avg_rank": 1.0 },
    "category":   { "mention_rate": 0.40, "avg_rank": 2.4 },
    "comparison": { "mention_rate": 0.45, "avg_rank": 2.2 },
    "use_case":   { "mention_rate": 0.20, "avg_rank": 3.1 }
  },
  "share_of_voice_vs_competitors": [
    { "name": "Acme Corp",  "mention_rate": 0.42 },
    { "name": "BrandX",     "mention_rate": 0.78 },
    { "name": "BrandY",     "mention_rate": 0.55 },
    { "name": "BrandZ",     "mention_rate": 0.31 }
  ],
  "top_citing_sources": [
    { "url": "wikipedia.org/...", "count": 8 },
    { "url": "reddit.com/r/...",   "count": 5 },
    { "url": "youtube.com/...",    "count": 4 }
  ]
}

Save to {OUTPUT_DIR}/citations-{YYYY-MM-DD}.json AND append a 1-row summary to {OUTPUT_DIR}/citation-history.json.

Output report (markdown, for humans)

# AEO Citation Report — {YYYY-MM-DD}

TL;DR
{Brand} mention rate: {X}% (was {Y}% last week, Δ {±Z}%)
Strongest engine: {engine} ({rate}%)
Weakest engine: {engine} ({rate}%)

vs Competitors (share of voice)
| Brand | Mention rate | Avg rank when cited |
| ---   | ---          | ---                 |

What changed since last run
{Top 3 deltas: queries that gained/lost the brand}

Action items

{What to fix — e.g. "Use case bucket weakest. Need 2 how-to articles answering '{query}' and '{query}'."}
{What to publicize — e.g. "Brand cited #1 in Perplexity for '{query}'. Reuse that snippet as social copy."}
{Off-page move — e.g. "BrandX cited 5x via Reddit. We have 0 Reddit threads. Casey/Jordan: post answer in r/{X}."}

Citation source intelligence (key insight)

Don't just track WHO is cited — track WHAT URLs the engines pull from. The hidden lever in AEO is being cited BY the URLs that AI engines already trust. If the report shows "Wikipedia + Reddit + YouTube + 3 industry blogs" as top citation sources for the category, the off-page strategy writes itself: get mentioned on those exact URLs.

OUTPUT CONTRACT (NO INTERNALS)

The output JSON and markdown are operational, not public, but treat them as if they could leak:

• Don't paste internal API keys or auth headers in response_full
• Don't include the verbatim system prompt sent to the engine — only the user query
• Snippet length capped at 200 chars per brand_context (anti-quotation-of-paywalled-content)

Common gotchas

1. Engines are stochastic — running the same query twice gets different responses. Run each query 1× per session, but compare week-over-week trends, not point-in-time numbers.
2. "Web search" mode matters — Claude/ChatGPT without web tools answer from training data (stale). Always enable web/search modes.
3. Engine refusal ("I can't recommend specific brands") — count as brand_mentioned: false, but flag in report; high refusal rate on competitive queries means the prompt is too commercial.
4. Brand name collisions — if BRAND = "Acme" and there's a famous "Acme Inc" in another industry, false positives explode. Either qualify ("Acme Corp" everywhere) or use disambiguating context match (brand_context must contain {CATEGORY} keyword).
5. First-run baseline — week 1 is data, not signal. Wait until week 4 before drawing conclusions about trends.