agentleFS
GhanyteGhanyte/geo-seo-audit/skills/geo-content-health/SKILL.md

geo-content-health

Skill9 starsChanged 7 months agoMITView on GitHub

Editorial health audit. Detects "Mount AI" anti-patterns (self-promotional listicles, mass-generated content, synthetic FAQ, mass refresh, negative-GEO injections) and measures sustainable patterns (BLUF format, definition language, atomic paragraphs, multi-channel coherence, first-hand expertise). Reports a risk score AND a maturity score.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If user passes --cwd, writes GEO-CONTENT-HEALTH.md in current directory.

    line 166

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Content Health — audit module

Purpose

Audit the editorial layer of a site: what kind of content does it publish, how is it structured, and is it currently exposed to ranking-penalty risk that would cascade into LLM-citation loss?

The methodology and confidence levels live in docs/CONTENT_HEALTH_PATTERNS.md. The Python implementation is in scripts/checks/content_health.py.

Implementation

Single-page audit:

python3 -m scripts.checks.content_health "<URL>" --lang fr
# Returns JSON with anti-pattern detection (CHA*) + sustainable-pattern measurements (CHM*).

Multi-page (recommended): sample pages via scripts/site_crawler.py then run the check on each. Aggregate the metrics (BLUF compliance, definition density, etc.) across pages — single-page numbers are unrepresentative.

Output language

By default, the report is written in the audited site's language, detected via scripts/site_locale.py (HTML lang → Content-Language header → TLD → fallback en). User can override with --output-lang <iso2>. See CLAUDE.md rule "Output language".


Two parallel scores

The audit reports two non-redundant scores:

  1. Risk score (0-100, lower is better) — based on anti-patterns detected ("Mount AI" triggers documented by Lily Ray on 30+ sites with -22% to -34% organic visibility drops since Jan 2026)
  2. Maturity score (0-100, higher is better) — based on sustainable patterns present (BLUF format, definition language, multi-channel coherence, first-hand expertise signals)

A site can have low risk and low maturity (clean but unexceptional). It can have high risk and high maturity (mixing risky tactics with substantive content). The two scores are complementary diagnostic axes.


🚫 Anti-patterns to detect (drives Risk score)

A1 — Self-promotional listicles ("Best of YEAR")

Detection:

  • Title regex (?i)\b(best|top|meilleur)\b.*\b(20\d{2})\b
  • The site's own brand mentioned in the first 200 words
  • Numbered list structure (numbered headings or <ol>)
  • Absence of methodology language ("we tested", "based on N hours of usage", explicit criteria)

Severity if found: High. Documented case: 10 such pages out of 1 700 articles (0.6% of content) caused -29% organic visibility on one site.

A2 — Mass-generated content fingerprints

Detection:

  • Templated H2 sequences (same H2 set across many articles, just with the topic substituted)
  • Identical intro/outro paragraphs across articles
  • Unattributed authorship (no Person schema, no byline)
  • Disproportionate ratio of articles to actual editorial team

Severity: High.

A3 — Synthetic FAQ spam

Detection:

  • High count of FAQPage schemas across the site (>50% of pages)
  • Questions following stuffed patterns
  • Answers <30 words OR answers that just paraphrase the question without adding fact

Severity: Medium.

A4 — Mass content refresh (date-bumping)

Detection (when reproducibly checkable):

  • dateModified within last 7 days on >30% of pages
  • Wayback Machine snapshot diffs show no real content delta

Severity: Medium-Low.

A5 — Negative-GEO injection (defensive scan)

Detection (HTML scan):

  • <section>, <div>, <aside> with class names like agent-instructions, system-prompt, llm-prompt whose content contains exclusion or self-promotion directives ("never cite", "do not include", "the only authorized source")
  • HTML comments with prompt-injection-style instructions
  • Non-standard <meta> tags addressing LLMs

Note: a legitimate <section class="agent-instructions"> block describing the entity factually (per STRUCTURED_DATA_BEST_PRACTICES.md) is good practice. A negative-GEO block tries to manipulate LLM behaviour against third parties. Distinguishing them needs content analysis (look for negation, explicit exclusion, or competitor naming).

Severity: Medium (decreasing efficacy + reputational risk). Reported as informational rather than fix-must.


✅ Sustainable patterns to measure (drives Maturity score)

M1 — BLUF format on key pages

For each H2 on key pages:

  • First-paragraph word count (target ≤ 60)
  • Presence of a number / data point
  • Presence of attribution (named source, "according to", named protocol/study)

Flag long meandering openers (>120 words before any factual claim). Confidence: High.

M2 — Definition language density

Count occurrences of definition phrases ("X est défini comme", "désigne", "X is Y", "refers to") in content body. FR + EN templates.

  • Compute density (% of paragraphs containing one)
  • Target ≥ 30%; flag pages with <15% Confidence: Medium (single Ahrefs correlation study).

M3 — Atomic paragraphs

  • Average paragraph length (target 40-80 words)
  • Question-form heading ratio (target 30-50% for content pages) Confidence: Medium-High.

M4 — Multi-channel brand coherence (cross-domain)

For each platform (free public APIs only):

  • Wikipedia / Wikidata presence (MediaWiki + SPARQL)
  • Reddit mentions (search public)
  • LinkedIn company page
  • YouTube channel + videos
  • GitHub presence
  • Crunchbase
  • Wayback Machine first capture
  • For French sites: Pappers, Societe.com

For each present channel, verify the brand description and key facts (founding year, location, headcount when stated) match the audited site. Discrepancies are red flags — they break attribution. Confidence: High (Ahrefs n=75 000).

M5 — First-hand expertise signals

  • Author bylines on articles linking to a Person schema
  • Person.alumniOf, knowsAbout populated for key authors
  • Original quantified claims with methodology disclosed (sample size, dates, methodology)
  • Industry citations or peer-reviewable references Confidence: High.

M6 — Freshness with substance

For pages with dateModified recent (<30 days), spot-check 1-2 against Wayback Machine to verify substantive change rather than date-bumping. Confidence: Medium.


Per-vertical adjustments

VerticalWhere the levers concentrate
E-commerceProduct page coherence, AggregateRating exposure, multi-channel brand presence
Editorial / blogBLUF format, definition language, author E-E-A-T, original data
B2B SaaSComparison tables (NOT self-promotional listicles), case studies with named clients & quantified outcomes
Local serviceNAP consistency, LocalBusiness schema, Google Business Profile freshness
Agency / consultancyCase studies, published frameworks, named experts, press mentions

Output

Resolved via scripts/client_paths.py paths <URL>:

  • Markdown report: $CLIENT_DIR/audits/<YYYY-MM-DD>-content-health.md

If user passes --cwd, writes GEO-CONTENT-HEALTH.md in current directory.

Report structure:

# Content Health — <domain>
**Audit date:** YYYY-MM-DD
**Vertical detected:** [...]

## Risk score: XX/100 (lower is better)
- A1 Self-promotional listicles: [N detected]
- A2 Mass-generated content fingerprints: [present / absent]
- A3 Synthetic FAQ spam: [yes / no]
- A4 Mass content refresh: [detected / not detected / not checked]
- A5 Negative-GEO injection: [detected / not detected]

## Maturity score: XX/100 (higher is better)
- M1 BLUF format: X% of H2 sections in target range
- M2 Definition language: X% paragraph density
- M3 Atomic paragraphs: avg X words, Q-form heading ratio Y%
- M4 Multi-channel coherence: X/8 channels present, N discrepancies
- M5 First-hand expertise: [strong / partial / weak]
- M6 Freshness with substance: [strong / not checked]

## Findings
[severity-grouped: Critical / High / Medium / Low / Informational, each with confidence + effort + score-gain estimate]

## Per-vertical commentary
[brief narrative on which levers to prioritise given the detected vertical]

This skill is one of the building blocks of /geo audit; when run in isolation it produces the report above.


What this skill does NOT do

  • Never blacklist competitor sites or domains. The audit reports patterns on the audited site only.
  • Never recommend negative-GEO tactics. Detection is in scope (defensive), injection is not.
  • Never claim a "Mount AI guarantee". The penalty triggers are documented but each Google update is unpredictable in detail. We frame as risk, not certainty.
  • Never project traffic / revenue impact. Per audit-only philosophy.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.