SEO Content Audit

Example output: examples/seo-content-audit-stripe-rate-limiters-20260514/VERDICT.md

Content Quality Audit

Score an existing piece of content against modern E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) and CITE (Clear answer, Include primary stats, Timestamp, Entity authority) rubrics. Surface veto items that block publication regardless of overall score. Produce a clear publish / publish-with-fixes / no-publish verdict with the top 5 fixes.

Prerequisites

• DataForSEO MCP server connected.
• Claude's WebFetch tool available.
• User provides: (a) the URL of an existing piece of content (or pasted content + intended URL), (b) target keyword the content is meant to rank for. Optional: target country (default us).

Process

1. Fetch content WebFetch (always) + mcp__firecrawl-mcp__firecrawl_scrape (when available)

   • Validate target & preflight. See skills/seo-firecrawl/references/preflight.md for the canonical 3-stage preflight (credit balance, Firecrawl availability, Google APIs). Skill-specific notes:
     • DataForSEO bills per call; this skill issues ~10–15 calls typical (AIO context + LLM-mention sampling for the target keyword + audited URL). Use the documented limit/filters params to cap.
     • Firecrawl: optional with WebFetch fallback, 1 Firecrawl credit per URL audited (default cap 50 URLs, hard cap 200). Surface the projected Firecrawl credit count before continuing. Pass --no-firecrawl to force WebFetch-only inspection (lower-confidence veto checks; see step 4 caveat).
     • Google APIs: tier 2 (GA4 available) unlocks step 3b (GA4 organic traffic on the audited URL) after the AIO context step. See skills/seo-google/references/cross-skill-integration.md § "seo-content-audit" for the full recipe.
   • WebFetch first (free, instant): pull the markdown for word count, H-tag hierarchy, source citations (links to authorities, numbered references), images, tables, code blocks, comment thread.
   • Page-type detection. From the URL pattern, H1 phrasing, and JSON-LD @type, classify the page as one of: ultimate guide / pillar, how-to, listicle / best-of, comparison (X vs Y), explainer, review (single product), landing page (commercial). Look up the corresponding word-count floor from references/core-eeat.md → "Word-count floors by page type". Surface the detected type, the applied floor, and the actual word count in evidence/01-content-snapshot.md. If the actual word count is materially below the floor, flag it for the depth E-E-A-T items (auto ✗ unless the auditor justifies the exception).
   • Firecrawl second — recovers what WebFetch's markdown loses:
     • From metadata: canonical URL, robots, lang, og:title.
     • From the returned html: every <script type="application/ld+json"> block. Parse for Article / BlogPosting schema and extract author (name, @type: Person, optional url + sameAs), datePublished / dateModified (ISO 8601), publisher, mainEntityOfPage. Detect Person schema standalone if present.
     • DOM-level byline detection: locate the structural byline (<a rel="author">, <meta name="author">, <span class="byline">, [itemprop="author"]). Distinguish a real byline element from prose mentions ("Written by Jane in collaboration..." in body text is not a byline; <a rel="author">Jane Doe</a> is).
   • If Firecrawl unavailable: WebFetch portion runs unchanged. Mark schema-type detection and structural byline detection as (skipped — Firecrawl required) in evidence/01-content-snapshot.md. Step 4's veto checks #1 and #4 fall back to prose-level inspection (less reliable) — surface that caveat in VERDICT.md.

2. AIO context mcp__dataforseo__serp_organic_live_advanced (AIO block) and mcp__dataforseo__ai_opt_llm_ment_top_domains

   • For the target keyword: is there a Google AI Overview? (read the AIO block from the serp_organic_live_advanced response).
   • Who is cited in the AIO / who gets mentioned across LLMs (ai_opt_llm_ment_top_domains)?
   • Is the candidate URL cited?
   • What patterns characterise the cited sources (publication tier, freshness, structure)?

3. LLM mention sampling mcp__dataforseo__ai_opt_llm_ment_search

   • Sample LLM mentions where the target URL's domain appears as a source. For an actual live answer, use mcp__dataforseo__ai_optimization_chat_gpt_scraper / mcp__dataforseo__ai_optimization_llm_response.
   • Cross-reference with the candidate URL — does it show up in any sampled mentions?

3b. GA4 organic traffic on the audited URL (only if google-api.json is present, tier ≥ 2)

• Replaces the implicit traffic estimation with actual measured organic sessions for the audited URL.
• Pull the top organic landing pages (last 28 days): python3 scripts/ga4_report.py --report top-pages --days 28 --json
• Filter the result client-side for the audited URL's path. Surface in VERDICT.md "## Snapshot" alongside the existing AIO citation cross-check:
  • GA4 organic last 28d: {sessions} sessions / {users} users / avg engagement {n}s
  • If the URL doesn't appear in the top-100 organic landing pages: "GA4: not in top-100 organic landing pages last 28d — low or zero traffic."
• This is a signal, not a veto. Low GA4 traffic on a YMYL page with high E-E-A-T is informative ("we're not earning the visibility our content quality should support") but doesn't change the publish decision.
• See skills/seo-google/references/cross-skill-integration.md § "seo-content-audit" for the full recipe.

4. Score E-E-A-T using references/core-eeat.md

   • 60-item rubric across 4 dimensions (15 items each).
   • Per-item: yes/no/partial. Compute dimension scores (0–100% each).
   • Score the 8-item AI-content markers subsection (see references/core-eeat.md → "AI-content markers"). Mark each fired/not-fired.
   • Apply 4 veto checks. Any veto = no-publish.
     1. Anonymous author on YMYL topic. (High-confidence with Firecrawl-recovered author schema + DOM byline; medium-confidence with prose-only inspection.)
     2. Factual claims with no sources cited.
     3. Undisclosed affiliate / sponsored relationships.
     4. AI-generated YMYL content with no human-review markers (≥4 AI-content markers fired AND YMYL topic AND no editor byline / "reviewed by" credit / "last reviewed" or "fact-checked on" date). The "editor byline / reviewed by" check uses Firecrawl-recovered DOM byline + Article-schema author when available; falls back to prose-level pattern match if Firecrawl is absent (lower confidence — note in VERDICT.md).

5. Score CITE using references/cite.md

   • 30-item CITE rubric (Clear answer in 1st 200 words, Include primary stats, Timestamp freshness, Entity authority).
   • Per-item: yes/no/partial. Compute dimension scores.
   • Apply 3 veto checks (no answer in first 300 words / no datestamp on time-sensitive content / no entity disambiguation for proper-noun queries).

6. Cross-check against AIO winners

   • For the patterns characteristic of cited sources (from step 2), evaluate the candidate against each: does it have what the cited sources have?
   • Surface specific gaps.

7. Synthesise verdict using templates/verdict.md

   • Publish: E-E-A-T ≥ 75%, CITE ≥ 70%, no vetoes triggered.
   • Publish with fixes: E-E-A-T 60–74% OR CITE 55–69%, no vetoes. Top 5 fixes specified.
   • No publish: any veto triggered, OR E-E-A-T < 60%, OR CITE < 55%. Substantial rewrite needed.

Output format

Create a folder seo-content-audit-{target-slug}-{YYYYMMDD}/ with:

seo-content-audit-{target-slug}-{YYYYMMDD}/
├── VERDICT.md                       (publish / publish-with-fixes / no-publish — primary deliverable; inlines content snapshot + AIO context)
├── 03-eeat-scoring.md               (60-item rubric scored — load-bearing reference an editor consults item-by-item)
├── 04-cite-scoring.md               (30-item rubric scored — load-bearing reference)
├── 05-aio-winner-comparison.md      (gap vs cited sources — must remain top-level; live AIO competitive evidence per EVAL_RESULT_v2.md §9)
└── evidence/
    ├── 01-content-snapshot.md       (HTML extracts + page metadata — raw step output)
    └── 02-aio-context.md            (AIO presence, citations, patterns — raw step output)

Step files 01 + 02 are inlined as a "Snapshot" / "AIO context" section in VERDICT.md; the copies in evidence/ preserve raw step output. 03-eeat-scoring.md, 04-cite-scoring.md, and 05-aio-winner-comparison.md stay at top level — editors consult the rubric scoring detail directly, and the AIO winner comparison is the live competitive evidence the rubric verdict rests on.

VERDICT.md follows this shape (also see templates/verdict.md):

# Content Audit: {URL or title}

> Audited {YYYY-MM-DD} · Target keyword: "{keyword}" · Country: {country}

## Verdict: {PUBLISH | PUBLISH WITH FIXES | NO PUBLISH}

{One sentence summary of why}

## Scores

| Dimension | Score | Threshold | Status |
|---|---|---|---|
| Experience | {n}% | 75% | {✓/✗} |
| Expertise | {n}% | 75% | {✓/✗} |
| Authoritativeness | {n}% | 75% | {✓/✗} |
| Trustworthiness | {n}% | 75% | {✓/✗} |
| **E-E-A-T composite** | {n}% | 75% | {✓/✗} |
| Clear answer | {n}% | 70% | {✓/✗} |
| Include stats | {n}% | 70% | {✓/✗} |
| Timestamp | {n}% | 70% | {✓/✗} |
| Entity authority | {n}% | 70% | {✓/✗} |
| **CITE composite** | {n}% | 70% | {✓/✗} |

## Veto checks

- Anonymous author on YMYL: {triggered / not triggered}
- Unsourced factual claims: {triggered / not triggered}
- Undisclosed affiliate / sponsored: {triggered / not triggered}
- AI-generated YMYL with no human review: {triggered / not triggered} ({n}/8 AI-content markers fired)
- ...

## AI Search readiness
- AIO present for "{keyword}": {yes/no}
- Top citation patterns: {list}
- Candidate URL cited in any sampled AIO: {yes/no}
- Gap vs cited sources: {bulleted gaps}

## Snapshot (measured)
- GA4 organic last 28d: {sessions} sessions / {users} users / avg engagement {n}s  *(or `not in top-100` / `not configured (Tier 2 required)`)*

## Top 5 fixes

1. {Specific fix linked to a low-scored item or veto}
2. ...
5. ...

## Detailed scoring

See:
- 03-eeat-scoring.md (item-by-item E-E-A-T)
- 04-cite-scoring.md (item-by-item CITE)
- 05-aio-winner-comparison.md (gap analysis)

Tips

• DataForSEO allows up to 2,000 calls/min, 30 concurrent. AIO + LLM-mention queries are ~5–10 calls; plenty of headroom.
• DataForSEO bills per call; no pre-run credit gate. ~10–15 DataForSEO calls typical, plus 1 Firecrawl credit per URL audited when Firecrawl is installed (default cap 50 URLs). Use the documented limit/filters params to cap.
• The thresholds (75% E-E-A-T, 70% CITE) are starting points. Tune per domain — a YMYL site (medical, financial) should require higher (85%/80%); a general-interest blog can run lower (65%/60%).
• The veto checks are not negotiable. A piece with anonymous authorship on a YMYL topic doesn't pass regardless of score.
• For pieces that score "publish with fixes," the top-5 list is the deliverable. Hand it to the writer; re-audit after fixes.
• Pair with seo-content-brief for the new-article counterpart: this skill audits existing content, content-brief produces new content.
• Pair with seo-sxo if the page has technical/page-type issues — that's a different diagnosis.
• The 60-item E-E-A-T rubric and 30-item CITE rubric are in references/. They are opinionated — adjust for your domain's editorial standards.