name: audit-llm-visibility
description: Checks whether an author's LinkedIn Article(s) are cited by LLM engines (ChatGPT, Perplexity, Google AI Overview, Claude.ai web). Runs a list of ICP queries, records citations, and computes share-of-voice. Run 4+ weeks after publishing an Article.
disable-model-invocation: true
argument-hint: [article-path] [--full | --queries-only]
allowed-tools: Read, Grep, Glob, Write, WebFetch, WebSearch
Audit LLM Visibility
This audit checks whether your content is being cited by LLM engines on queries relevant to your target audience. It's the GEO equivalent of Google Search Console for SEO.
When to invoke
- 4+ weeks after publishing an Article (earlier than that — indexation lag, nothing should have kicked in yet)
- Quarterly — full audit across all published Articles
- Before planning the next Articles series — to understand which patterns are working
When NOT to invoke
- Right after publication (too early to judge)
- For short posts (LLMs rarely cite short-form)
- Without a list of ICP queries (you need specific target queries, not generic ones)
Pre-requisites
Before running, you need:
- Article(s) published 4+ weeks ago
- Direct Pulse article URL saved in frontmatter
geo_target_queries written in frontmatter (minimum 5 queries)- Published on the personal profile (optionally — on the Company Page)
Pipeline (5 steps)
Step 1: Load context
- Read
Knowledge-Base/10-Articles-and-GEO/LLM-Visibility-Audit-Methodology.md — methodology
- Read the target article — find
direct_url, geo_target_queries, named_concept
- If running in
--full mode — read the competitor list from Knowledge-Base/03-Viral-Trends/Viral-Trends-Watchlist.md (for share-of-voice tracking)
Step 2: Prepare query list
Take geo_target_queries from the article frontmatter. Expand to minimum 20 queries covering 4 categories:
| Category | Count | Source |
|---|
| A. Branded ("[Company] reviews") | 5 | From frontmatter |
| B. Author-branded ("[Author] [topic]") | 3 | From frontmatter |
| C. Topic + named concept | 5 | From frontmatter (named_concept) |
| D. Generic ICP top-of-funnel | 5 | From frontmatter |
| E. Comparative ("[Your product] vs X") | 2 | From Viral-Trends-Watchlist |
If frontmatter doesn't have enough — ask the user.
Step 3: Run queries through LLM engines
For each query — perform web searches and record results:
- WebSearch: query → check whether your Article shows up in results
- WebFetch for AI Overview:
https://www.google.com/search?q=[query] → check AI Overview section
- Optional: if API access is available — Perplexity API, ChatGPT API with browsing
For each result, record:
- Does your brand appear?
- Which URL is cited (if any)?
- Verbatim quotes from your article (exact match)?
- Which competitors are mentioned?
- Tone of mention (positive / neutral / context-only)
Step 4: Analyze results
Group findings by category. Compute:
Citation rate (per category) = N queries with brand mention / total queries
Direct URL discovery rate = N queries where Article URL appears / total
Verbatim citation rate = N queries with direct quote / total
Competitor share-of-voice = competitor mentions / total mentions
Apply decision rules from LLM-Visibility-Audit-Methodology.md:
- Categories A+B+C work, D+E don't → normal for LinkedIn-only. Continue.
- No category works after 4 weeks → problem in Article structure. Recommend revision.
- Competitors dominate D+E → confirms you need your own blog with SEO for cold acquisition.
Step 5: Save audit report
Create file Articles/_Audits/YYYY-MM-DD-[article-slug]-audit.md using the template from LLM-Visibility-Audit-Methodology.md ("Audit report template" section):
# LLM Visibility Audit — [Date] — [Article slug]
> **Article:** [title]
> **Age:** [days since publish]
> **Method:** [manual / WebSearch + WebFetch]
## TL;DR — 3 key findings
1. ...
2. ...
3. ...
## Per-category results
[5 categories with metrics + examples]
## Verbatim citations found
[list of direct quotes from Google AI Overview / other engines]
## Verdict: [X/10]
[1-line summary]
## Recommendations
[3–5 action items with owner + deadline]
Output
- Report file in
Articles/_Audits/
- Inline chat summary — TL;DR + verdict
- Suggested next actions — what to improve in article structure, which queries to add in the next monitoring round
What this skill does NOT do
- Does NOT predict future visibility (only current state)
- Does NOT auto-optimize the article (only shows where it fails)
- Does NOT track competitor mentions in historical view (need
--full mode for baseline)
- Does NOT replace professional AEO-tracking tools at scale (Profound, Athena AI, Otterly.AI) — this is manual baseline + early warning
Anti-patterns
- ❌ Running 1 week after publication (too early)
- ❌ Only Category A (creates illusion of visibility)
- ❌ Only one LLM engine (need at least 2 for reliability)
- ❌ Audit without action items (becomes a shelf document)
- ❌ Running more often than every 2 weeks (changes don't surface that fast)
Relation to the system
| Document | How it relates |
|---|
LinkedIn-Articles-GEO-Framework.md | What an article needs to have to pass this audit |
Articles-Plan-Template.md | Where to write geo_target_queries in advance |
Growth-Loop-Architecture.md | LLM citation rate can be added as a metric in the Growth Loop |
.claude/skills/create-article/SKILL.md | Creates the article with the frontmatter this skill reads |
