| name | ai-seo |
| description | This skill MUST be invoked when the user says "ai seo", "seo analiz", "GEO audit", "site optimize", "AI arama optimizasyonu", "citability", "ai citation", "crawlers check", "llms.txt", "brand mentions", "structured data audit", "schema audit", "technical seo", "E-E-A-T", "ai visibility", "generative engine optimization", or any variation requesting website optimization for AI-powered search engines. Analyzes websites for AI search readiness across citability, crawler access, brand authority, structured data, technical foundations, and content quality. Accepts a live URL or a local project path (scan source files before deploy).
|
| argument-hint | [audit | citability | crawlers | llmstxt | brands | schema | technical | content] <url | path> |
AI SEO — Generative Engine Optimization
Optimize websites for AI-powered search engines (ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews) while maintaining traditional SEO foundations.
AI search is eating traditional search. This skill optimizes for where traffic is going, not where it was.
Usage
/ai-seo audit <url|path>
/ai-seo citability <url|path>
/ai-seo crawlers <url|path>
/ai-seo llmstxt <url|path>
/ai-seo brands <url|path>
/ai-seo schema <url|path>
/ai-seo technical <url|path>
/ai-seo content <url|path>
The argument is either a live URL (https://example.com) or a local project path (., ./src, /path/to/project). See Input Modes: Web vs Local below for how each subcommand reads from disk instead of fetching.
Routing
The first token after /ai-seo selects the mode:
- If it matches a known subcommand keyword —
audit, citability, crawlers, llmstxt, brands, schema, technical, content — run that subcommand.
- Otherwise treat the whole argument as the target (URL or path) and run the full audit, which launches all 7 analyses in parallel.
Keyword match always wins: /ai-seo content runs the content subcommand, not an audit of a directory named content. To audit a path that collides with a keyword, name it explicitly: /ai-seo audit ./content.
Why GEO Matters
Figures below are directional and dated — verify current numbers before quoting them to a client.
| Signal | Trend (named source) |
|---|
| AI-referred traffic | Rising fast — one 2025 report tracked a 527% jump over five months (Previsible), though AI is still a small share of total traffic |
| AI traffic intent vs organic | Reported as higher-converting (Semrush 2025 modeled ~4.4x; other studies report very different multipliers) |
| Brand mentions vs backlinks for AI | Mentions correlate more strongly than backlinks (Ahrefs 2025-26) |
| Gartner: search engine volume | -25% by 2026 |
| llms.txt adoption | Still low among top sites — early-mover advantage |
Subcommands
| Subcommand | Command | Description |
|---|
audit | /ai-seo audit <url|path> | Full GEO audit with composite scoring |
citability | /ai-seo citability <url|path> | Passage-level AI citation readiness |
crawlers | /ai-seo crawlers <url|path> | AI crawler access analysis (robots.txt) |
llmstxt | /ai-seo llmstxt <url|path> | llms.txt standard analysis or generation |
brands | /ai-seo brands <url|path> | Brand authority on AI training platforms |
schema | /ai-seo schema <url|path> | Structured data + entity recognition |
technical | /ai-seo technical <url|path> | Technical SEO for AI crawlers |
content | /ai-seo content <url|path> | Content quality + E-E-A-T scoring |
GEO Composite Score (Full Audit)
The full audit produces a weighted composite score (0-100):
| Category | Weight | Subcommand |
|---|
| AI Citability & Visibility | 25% | citability |
| Brand Authority Signals | 20% | brands |
| Content Quality & E-E-A-T | 20% | content |
| Technical Foundations | 15% | technical |
| Schema & Structured Data | 10% | schema |
| Platform Optimization | 10% | crawlers + llmstxt |
In Local mode, any category whose source is not in the repo (e.g. CMS-driven content, web-only brand reputation) is excluded from the composite and the remaining weights are renormalized to total 100 — never scored 0. See Input Modes: Web vs Local below.
Score interpretation:
| Range | Grade | Meaning |
|---|
| 80-100 | A | AI-optimized — strong visibility in AI search |
| 60-79 | B | Good foundation — targeted improvements needed |
| 40-59 | C | Significant gaps — AI crawlers may struggle |
| 20-39 | D | Major issues — minimal AI search presence |
| 0-19 | F | Not optimized — invisible to AI search |
Business Type Detection
When running a full audit, detect the site type to adjust analysis focus:
| Type | Signals | Adjusted Focus |
|---|
| SaaS | /pricing, /docs, /api, app subdomain | Documentation citability, API schema |
| Local | Address, map embed, NAP data, reviews | LocalBusiness schema, Google Business |
| E-commerce | /products, cart, price elements | Product schema, review markup |
| Publisher | /blog, /articles, bylines, dates | Article schema, E-E-A-T, citability |
| Agency | /services, /portfolio, /case-studies | Organization schema, brand authority |
Input Modes: Web vs Local
Every subcommand accepts either a live URL or a local project path. Detect the mode from the argument before doing anything else:
| Argument looks like | Mode | How the site is read |
|---|
http://… or https://… | Web | Fetch the live site |
A bare domain (example.com) with no matching local path | Web | Treat as https://… and fetch |
., ./src, ../site, /abs/path, ~/proj, or any existing file/dir | Local | Read project files from disk |
If the argument is ambiguous, check whether it exists as a local path (Glob/Read). If it does, use Local mode; otherwise treat it as a URL.
Tool mapping
| Need | Web mode | Local mode |
|---|
| Page / HTML content | WebFetch | Read + Glob + Grep on source files |
| Raw files (robots.txt, llms.txt, sitemap.xml) | Bash(curl -s <domain>/<file>) | Read from project root, then public/, static/, dist/, build/ |
| Brand mention discovery | WebSearch | WebSearch — Local mode changes WHERE the site is read, not WHETHER the web is allowed |
Do NOT use Python scripts or external dependencies. The LLM's context-aware analysis replaces regex-based scoring.
Local project layout
Where to look in Local mode (framework-dependent — Glob to find the real location):
- Control files:
robots.txt, llms.txt, llms-full.txt, sitemap.xml → repo root or public/, static/, dist/, build/
- Content & templates:
*.html, *.md, *.mdx, *.vue, *.svelte, *.astro, *.jsx, *.tsx → content/, src/, pages/, app/, templates/
- Structured data:
Grep "application/ld+json" across source files
- Framework signals:
package.json dependencies, next.config.*, nuxt.config.*, astro.config.*, gatsby-config.*, svelte.config.*, vite.config.*
Local mode honesty rules
Local source is what you wrote, not what an AI crawler renders. Two rules keep local scores honest:
- Not in the repo → exclude, never zero. If a category's source is not present locally (content is CMS- or API-driven, only empty templates exist, brand reputation needs the web), mark it "Not assessable locally — run against the live URL" and DROP it from the composite, then renormalize the remaining weights to total 100. Never score it 0 — that fabricates a low grade.
- Static analysis is a lower bound. Runtime-injected markup (Next.js
<Head> JSON-LD, CMS-injected schema, client-rendered content) is invisible to a static file scan. Local Schema / Citability / Content / SSR scores are a FLOOR, not authoritative. Say so in the report and recommend verifying against the live URL.
Output Format
All subcommands return analysis in the response as structured markdown. No files are created unless explicitly requested. Each analysis includes:
- Score (0-100) with grade (A-F)
- Key findings with specific file/line/URL references
- Prioritized recommendations (Critical → High → Medium → Low)
- Quick wins highlighted separately
