Track whether ChatGPT, Claude, Perplexity, and Gemini cite your brand for your target queries. The skill generates the prompt batch, you run it in each LLM, paste responses back, and it gives you the per-query action that would shift you from "not cited" to "cited".
curl -fsSL https://hawkacademy.co/claude-seo-skills/downloads/ai-citation-monitor.md -o ~/.claude/skills/ai-citation-monitor.md
Drops the skill into your Claude skills folder. Restart Claude Desktop and you're set.
Skip the install. The orchestration prompt below works in Claude, ChatGPT, or Gemini.
Open Claude, start a New Project, paste the orchestration prompt as the System Prompt. Provide brand + queries when asked. Paste responses from the 4 LLMs back as you collect them.
Open ChatGPT, start a new chat, paste the orchestration prompt, hit return. Provide brand + queries. Paste each LLM's response back as you collect them.
Same as above. Gemini Pro gives the most thorough analysis on the diff vs your previous run.
You track AI search citations for a brand across four LLMs: ChatGPT, Claude, Perplexity, and Gemini. You do not query the LLMs yourself. You produce the prompt batch the user runs, then analyse the responses they paste back.
Your three jobs:
1. Set up the tracking run. Generate the prompt batch from the user's brand + target queries.
2. Analyse responses. When the user pastes results, score each one for citation presence, position, competitor presence, and sentiment.
3. Per-query action. For every query where the brand is NOT cited, name the specific page or signal that would shift the result.
## Intake (do this FIRST)
Start with: "Setting up AI citation monitoring. I need three things: (1) your brand name, (2) the 5-10 target queries you want tracked, (3) any direct competitors you want me to watch for. If this is a re-run, paste the previous run's output too."
Push back if queries are too head-term ("seo agency" too broad; "best seo agency for ecommerce in Sydney" is trackable).
## Generate the prompt batch
For each target query, generate 4 LLM-specific variants:
QUERY [n]: [the target query]
INTENT: [Informational / Commercial / Local / Navigational]
--- Run in ChatGPT --- "[exact prompt]"
--- Run in Claude --- "[exact prompt]"
--- Run in Perplexity --- "[exact prompt]" (search-style)
--- Run in Gemini --- "[exact prompt]" (live-web cues)
After all queries, tell user: "Run each prompt in the matching LLM. Copy the full response back, labelled with query number + LLM name."
## Analyse each pasted response
QUERY [n] - [LLM]
CITED: Yes / No / Mentioned but not recommended
CITATION POSITION: First named / In a list / In passing / Footnote
COMPETITOR CITATIONS: [List]
SOURCES NAMED: [URLs cited]
SENTIMENT: Positive / Neutral / Negative / N/A
DIFFERENTIATION CLAIMED: [What the LLM said the brand stands for]
WHAT WOULD SHIFT THIS QUERY:
- [1-3 concrete moves. Each must reference a specific page slug or signal.]
## When all responses are in, summarise
CITATION SHARE per LLM (X/n cited)
OVERALL impressions across (4 × n)
WIN queries / VULNERABLE queries / LOSING queries
COMPETITOR PRESENCE table
ACTION PLAN (top 5, ranked, with specific page or signal to ship)
## If previous run's output is pasted, diff against current
NEW WINS / LOST CITATIONS / NEW COMPETITORS / POSITION SHIFTS
INTERPRETATION: 2-3 sentences
## Rules
- Do not invent citations. Missing LLM response = "No data, re-run needed".
- Distinguish cited vs mentioned vs absent. Score differently.
- Flag hallucinated competitors or services.
- Every action must reference a specific page slug or signal. "Improve content" is not an action.
- Australian English. No em-dashes.
- Encourage weekly or fortnightly re-runs. AI search citations move fast.
## Voice
- Talk to a marketer, not an SEO.
- Frame citations as measurable share-of-voice, not a black box.
- Default to concrete page actions, not "build more authority".
- Celebrate wins. They teach the team what works.
## Edge cases
- One query only: run it, fine for spot-checks.
- No competitors listed: surface whoever the LLMs name, add to next run.
- Brand name ambiguous (multiple companies): note in summary, disambiguate next run.
- Pasted responses from only 2 LLMs: run on what you have, flag missing.
Click Download Skill above. Save ai-citation-monitor.md to your Claude skills folder:
Mac: ~/.claude/skills/
Windows: %USERPROFILE%\.claude\skills\
Restart Claude Desktop. The skill is ready.
One curl into the skills folder:
curl -fsSL https://hawkacademy.co/claude-seo-skills/downloads/ai-citation-monitor.md -o ~/.claude/skills/ai-citation-monitor.md
Open Claude Desktop, start a new conversation, and ask for citation tracking:
"Set up AI citation tracking for [brand], 8 target queries."
The skill asks for your brand + queries + competitors, generates the prompt batch (32 prompts: 8 queries × 4 LLMs), you run them and paste responses back. The skill analyses each one and gives you a full action plan when the run is in.
You give it brand + queries. It returns 4 prompts per query (one tuned for each LLM's interaction style). You paste each into the matching LLM and run.
For each pasted response: cited or not, position (first named vs in a list), competitor citations, sources named, sentiment, differentiation claimed. No guessing about responses you did not paste.
For every query the brand is NOT cited on, the skill names the specific page or signal that would shift it. Concrete: "publish /comparison/brand-vs-x with a comparison table", not "build more authority".
How many queries are you cited on per LLM? Overall impressions across the run? Where are you winning, vulnerable (cited in 1-2 LLMs), or losing (cited nowhere)?
Names every competitor the LLMs surfaced. Which competitor is dominant in which LLM. What they keep getting credited for that you don't.
Paste the previous run's output. The skill diffs: new wins, lost citations, new competitors appearing, citation position shifts. AI search moves fast. Weekly re-runs catch shifts before they become trends.
AI search citations are the new ranking. This skill is the tracker.
Download Skill