Skill: GEO (Generative Engine Optimization)
Purpose
Classic SEO optimizes for ranking in result lists; GEO optimizes for being retrieved, quoted and cited by generative engines that compose answers. This skill audits a site and applies the fixes. It complements a classic SEO pass and never replaces it (meta, sitemap and Core Web Vitals stay SEO territory; if a dedicated seo skill is available, use it for those).
Audit
- Renderability. Fetch key public pages with
curl(no JS). What does a crawler actually see? SPA/Inertia pages render an empty shell, and generative crawlers mostly do not execute JS. If the content that should be cited is invisible without JS, that is finding #1 and nothing else matters until it is fixed (SSR, prerendering, or dedicated server-rendered pages for public content). - Machine surface. Check for
/llms.txt(curated map of the site's content for LLM crawlers),robots.txtrules for AI crawlers (GPTBot, ClaudeBot, PerplexityBot…: blocked intentionally or by accident?), and schema.org structured data (JSON-LD: Organization, Product, FAQPage, Article, BreadcrumbList as relevant). - Citability of the content itself. Generative engines quote passages, not pages. Look for: self-contained paragraphs that answer one question (quotable without surrounding context); headings phrased as the questions users actually ask; concrete facts with units, dates and sources rather than marketing prose; visible freshness (dated content); consistent naming of the product/entity so the model can attribute facts to it.
Apply
- Fix in the order above: renderability → machine surface → citability.
llms.txt: a short markdown index (what the site/product is, links to the pages worth reading, one line of context each).- JSON-LD blocks server-side (in the HTML head, not injected by JS).
- Rewrite key sections so each answers one question in its first sentence; keep the H2/H3 as the question.
- Never degrade the human page to please crawlers: GEO changes must be invisible or beneficial to human readers.
Report
List findings ordered by impact, each with: what a generative engine currently sees, what it should see, and the concrete fix applied or proposed.