/01Technical SEO
robots.txt (20+ AI bots), sitemap validity, redirect chains, security headers, broken links, internal-link graph, server-log analysis, Core Web Vitals.
8 checkers
The SEO audit built for AI search
One Claude Code prompt runs a full SEO + AEO + GEO audit — 29 deterministic checkers, real-browser Core Web Vitals, and live citation testing. Deterministic, confidence-labeled, and open source.
Citation-tested againstChatGPTClaudePerplexityGeminiAI Overviews
What one prompt measurestap a capability → what · why · how
See exactly what the audit checks — and why it matters.
AI-crawler access
- What
- The skill reads your robots.txt and checks, per bot, whether 20+ AI crawlers are allowed in — GPTBot, ClaudeBot, PerplexityBot, Google-Extended and more.
- Why
- If an AI crawler is blocked, the model can never cite you — no matter how good the page is. It's the first failure, and the most invisible.
- How
robots_checker.pyparses every User-agent block, resolves Allow/Disallow per crawler, flags accidental blocks and recommends 301 vs 302 on HTTPS upgrades. Deterministic — a Confirmed finding.
20+
AI crawlers checked per-bot — GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and more.
CrawlerAccess · robots.txt
GPTBotAllow
ClaudeBotAllow
PerplexityBotAllow
Google-ExtendedDisallow
Health Score
- What
- A single weighted 0–100 score across technical, schema, images, links, Core Web Vitals, content and GEO — with a per-category breakdown.
- Why
- You need one figure you can track over time, not forty disconnected warnings that never add up to a decision.
- How
page_score.pyruns every applicable checker in parallel. If a check can't run, its weight is removed from the denominator — never scored as zero. Same input, same verdict.
0–100
weighted & deterministic — the score is the single source of truth, defined in one place in the code.
95 / 100 · example.com
050100
Structured data
- What
- Detects your JSON-LD and checks required and recommended fields, per item, scoring each 0–100 for completeness.
- Why
- A missing required field silently kills rich-result eligibility — and models lean on structured data to understand what your page is about.
- How
schema_recommended_fields.pynames every missing field, not just "schema missing". Knows what Google deprecated (never HowTo; FAQ only gov/health).
names
every missing required + recommended field — so you know exactly what to add, per item.
1
2
3
4
5"@type": "Article", "headline": "…" ok "author": missing — required "datePublished": missing "dateModified": missing — freshness
2
3
4
5"@type": "Article", "headline": "…" ok "author": missing — required "datePublished": missing "dateModified": missing — freshness
Live AI citations
- What
- Five LLM surfaces are asked your target queries; the skill records which ones actually cite your domain in their answer.
- Why
- Ranking #1 in Google no longer means the AI answer quotes you. This is the layer traditional SEO tools never measure.
- How
- A 5-LLM ensemble — Anthropic, OpenAI, Perplexity, xAI, and Gemini-with-Google-Search grounding — the closest public proxy for Google AI Overviews and AI Mode.
25.11%
of Google searches now trigger AI Overviews, up from 13.14% a year earlier — Semrush, 2026.
Engine3 / 5 cite you
ChatGPTcited
Perplexitycited
Claudecited
Gemini · Searchnot cited
Core Web Vitals
- What
- Real-browser LCP, INP and CLS — field data (CrUX, 75th percentile) plus lab Lighthouse.
- Why
- Page Experience feeds ranking, and slow, shifting pages lose visitors before they ever convert.
- How
psi_checker.pypulls PageSpeed Insights field + lab; a Playwright pass catches SPA content only Googlebot's second wave sees. Uses INP — never the retired FID.
INP
measured field + lab at the 75th percentile — the responsiveness metric that replaced FID in 2024.
LCP2.1 s
INP180 ms
CLS0.04
Confidence labels
- What
- Every finding is tagged Confirmed, Likely or Hypothesis — so you always know the evidence behind it.
- Why
- It's the anti-hallucination guard: the skill never presents reasoning as a measured fact, so you act on the right things first.
- How
- Confirmed = a deterministic checker or real-browser measurement · Likely = reasoning plus a signal · Hypothesis = reasoning alone, flagged for validation.
3
labels on every finding, each tied to the exact evidence that produced it.
● ConfirmedLikelyHypothesis
🛑 P0 · hreflangConfirmed
⚠ P1 · securityConfirmed
/03How it works
From one prompt to a ranked action plan.
You type a plain-English request. Behind it, the skill detects your business type, fans out parallel checkers, measures in a real browser, tests live AI citations, and returns a 0–100 score.
/01
Trigger
Type
audit https://example.com — or "сделай SEO аудит example.com". Activates on 25+ keywords, no slash command./02
Detect
Reads the homepage and classifies the site — SaaS, local, e-commerce, publisher, agency or generic — then loads the matching thresholds.
/03
Fan out
Spawns six sub-agents at once — technical, content, schema, sitemap, performance, visual — and de-duplicates contradictions before scoring.
/04
Measure
29 Python checkers produce reproducible verdicts; a real browser measures Core Web Vitals; Playwright diffs server vs JS-rendered HTML.
/05
Test citations
Five LLM surfaces are queried with your target terms; the skill records which cite your domain. Gemini-with-Search proxies AI Overviews.
/06
Score
A 0–100 Health Score with category breakdown, confidence-labeled findings, and a priority plan: Critical → High → Medium → Low.
/04The functionality
Everything it checks, in eight groups.
29 deterministic checkers, real-browser measurement, and a 5-LLM ensemble — every checker outputs structured JSON with a real exit code.
/02Content & E-E-A-T
Flesch reading ease, keyword-stuffing, 134–200-word citable-passage extraction, author + date signals, and an E-E-A-T anti-pattern detector.
content + eeat
/03Schema.org
Detection, required + recommended field coverage per item, per-item completeness 0–100, and JSON-LD generation.
schema
/04Images & media
Alt coverage, format mix vs a ≥70% next-gen target, width/height for CLS, lazy-loading below the fold, and size flags.
images
/05AI search — AEO/GEO
AI Visibility Score (6 components), the 5-LLM citation ensemble, llms.txt checker, JS-render diff, AI-crawler accessibility.
5 checkers
/06Premium diagnostics
Eight v0.8.0 checkers: Google-updates correlation, cannibalization, intent classification, content tiers, backlink toxicity, E-E-A-T sweep.
v0.8.0 · 8
/07Growth
Competitor keyword-gap (Ahrefs MCP), competitor-page generation, programmatic-SEO planning with quality gates, hreflang/i18n, local SEO.
strategy
/08Tooling
Unlimited async crawler, site-wide orchestrator, SQLite history + static dashboard, CMS detection (24+ platforms), self-contained HTML reports.
24+ platforms
/05Commands
26 commands. You never memorize them.
Plain English routes to the right one automatically — but every capability is a named command you can call directly.
Core audits (3)
auditfull-site auditpagesingle-page scoreai_visibilityAI Visibility Score
Technical (9)
technicalschemaimageslinkssecuritysitemapjs_rendercmslogs
Content (2)
contentlocal
AI search (3)
geoaeolive 5-LLM checkserp
Growth (5)
growthplanprogrammaticcompetitor-pageshreflang
History & reports (4)
historydashboardserve_dashboardreport
/06Why this exists
If your site isn't cited by AI, you're invisible to a growing share of search.
Third-party market research, linked below. The skill doesn't ask you to trust it — it measures whether your own pages are reachable, structured, and actually cited.
25.11%
Of Google searches now trigger AI Overviews — up from 13.14% in March 2025.
Search trigger rate
900M
Weekly active ChatGPT users — bigger than most traditional search engines.
ChatGPT WAU
43%
Google searches that now end with zero clicks — the answer never leaves the page.
Zero-click rate
/07Vs standard tools
What a $99–$999/mo SEO tool doesn't tell you.
| Capability | Standard SEO tools | Amazing SEO Skill |
|---|---|---|
| Findings per audit | 5–20 high-level | 50+ specific, actionable |
| Real Core Web Vitals | Often missing | ✓ PSI field + lab |
| Schema recommended fields | "X missing" | ✓ Names every field |
| Live LLM citation testing | ✗ | ✓ 5-engine ensemble |
| Confidence labels | ✗ | ✓ Confirmed / Likely / Hypothesis |
| Runs inside your coding agent | ✗ | ✓ Claude Code skill |
| Cost | $99–$999 / mo | Free + your own API keys |
SEO tools tell you why you don't rank. Amazing SEO Skill also tells you why you don't get cited.
/08Install
One clone. Then just ask.
Lightweight mode works on any Claude Code install — reasoning + WebFetch, no Python. Clone into your skills directory and the trigger keywords do the rest.
$ git clone https://github.com/metawhisp/amazing-seo-skill.git \
~/.claude/skills/amazing-seo-skill
Then, in Claude Code, say: audit https://example.com — the skill auto-activates on keywords like SEO, audit, schema, Core Web Vitals, AEO, GEO. Full mode adds real-browser CWV, the 251-rule deep-audit engine, and live AI-citation testing across 5 LLMs.
/09FAQ
Frequently asked.
/01What is Amazing SEO Skill?+
An open-source skill for Claude Code that runs end-to-end SEO + AEO + GEO analysis from one prompt — Claude reasoning, 29 deterministic Python checkers, real-browser Core Web Vitals, and a 5-LLM ensemble that tests whether AI engines actually cite your site.
/02What's the difference between SEO, AEO and GEO?+
SEO targets traditional search. AEO targets answer surfaces — snippets, voice, AI Overviews. GEO targets generative AI — ChatGPT, Claude, Perplexity, Gemini — where the engine writes an answer and cites sources.
/03How do I install it?+
Clone into your skills directory: git clone https://github.com/metawhisp/amazing-seo-skill.git ~/.claude/skills/amazing-seo-skill. Then ask Claude "audit https://example.com". Lightweight mode needs nothing else.
/04Is it free and open source?+
Yes — Apache-2.0, free to use, fork and ship. The core audit runs with no API key. Optional keys unlock higher limits and live-data layers.
/05How is this different from Ahrefs or Semrush?+
Traditional tools measure what Google uses to rank. This also measures what AI engines use to cite — crawler access, structured data, citable passages, and actual live citations. It runs inside Claude Code and can read your Ahrefs and GSC data via MCP.
Rank in Google. Get cited by the AI that answers for it.
Clone in seconds. Run your first audit in one prompt. Apache-2.0, no signup, no API key to start.