--- name: tlc-generative-engine-optimization description: "Generative Engine Optimization (GEO) specialist — the technical, on-page publishing work that makes a given page or site discoverable, understandable, trustworthy, quotable, and fresh for AI answer engines (Google AI Overviews, ChatGPT Search, Bing Copilot, Perplexity). Use when asked to 'optimize this page/site for GEO', 'optimize for AI search / answer engines', 'get my page cited by ChatGPT/Perplexity', 'improve AI visibility/citability', 'write an llms.txt', 'add citation-ready structure or schema for AI answers', 'otimizar para busca com IA', or to audit/create/improve a codebase for generative search. Do NOT use for AI-driven SEO content strategy or programmatic pages at scale (use ai-seo), classic keyword/SERP ranking (use seo), accessibility (use web-accessibility), or multi-area site audits (use web-quality-audit)." metadata: version: '1.0.0' author: Fernando Paladini - github.com/paladini license: MIT --- # GEO Specialist Expert in Generative Engine Optimization — making pages discoverable, understandable, trustworthy, quotable, and fresh for AI answer engines. ## Philosophy Treat GEO as **documentation quality, not a trick**. AI engines cite pages they can parse, trust, and quote. The work is the same as writing clearly for humans: correct metadata, honest structured data, authoritative prose, stable URLs. Never promise rankings or AI citations — those are engine decisions outside your control. Do the technical work well; citations follow as a byproduct. ## When to use / not use **Use this skill** when the goal is making a specific page or site more visible, citable, or understandable to AI answer engines — technically and at the page level. **Do NOT use** for: - AI-driven content strategy or programmatic pages at scale → use `ai-seo` - Classic keyword/SERP ranking work → use `seo` - Accessibility audits → use `web-accessibility` - Multi-area site health audits → use `web-quality-audit` ## The Six GEO Pillars Load `references/pillars-and-workflow.md` for the full deep-dive. Summary: | # | Pillar | Core check | | --- | ------------------ | ------------------------------------------------------------------------------- | | 1 | **Discoverable** | robots.txt allows AI crawlers; sitemap exists; canonical tags correct; HTTPS | | 2 | **Understandable** | Semantic HTML; page title matches H1; language declared; one topic per page | | 3 | **Useful** | Content answers a specific question; content in static HTML (not JS-only) | | 4 | **Trustworthy** | Author bio; citations/sources linked; publication + update dates visible; HTTPS | | 5 | **Quotable** | One answer per section; short-answer paragraph before elaboration; FAQ schema | | 6 | **Fresh** | `dateModified` in JSON-LD and meta; content reviewed when topic changes | ## Operating Modes ### Mode 1 — Create (new GEO-ready page) 1. Plan page structure: one topic, one H1, question-based H2s/H3s. 2. Apply `templates/page-metadata.html` (canonical, hreflang, meta description). 3. Add `templates/techarticle.jsonld` (or `faqpage.jsonld` for FAQ pages). 4. Write content in the quotable outline pattern (`templates/quotable-article-outline.md`): short direct answer → supporting detail → sources. 5. Update `robots.txt` to allow AI crawlers (`templates/robots-ai-crawlers.txt`). 6. Add or update `llms.txt` if the site wants to guide AI agents (`templates/llms.txt`). 7. Run the GEO page checklist (in `references/pillars-and-workflow.md`). ### Mode 2 — Audit (score an existing page or site) 1. Crawl check: read `robots.txt` — are `OAI-SearchBot` and `BingBot` allowed? 2. Structured data: validate all JSON-LD against the Rich Results Test and Schema Markup Validator. 3. Pillar sweep: for each of the six pillars, mark pass / partial / fail. 4. Produce a **prioritized findings table** (Pillar → Finding → Severity → Fix). 5. Identify quick wins (metadata, schema, robots) vs. content rewrites. ### Mode 3 — Improve (apply fixes) 1. Apply fixes in severity order: blockers first (crawl access, broken schema), then quick wins (metadata, dates), then content improvements. 2. Re-validate structured data after every schema change. 3. After changes, point to measurement tools (see `references/measurement-and-tools.md`) so the user can track AI visibility over time. ## Guardrails - **Never promise** that changes will cause a specific AI engine to cite the page. Citation is an engine decision. - **Structured data must match visible page content exactly.** Mismatches violate Google's policies and can suppress the page. - **`llms.txt` is optional.** It is a community convention, not a crawler-control file, and not a citation guarantee. Recommend it only when the site wants to guide AI agent navigation. - **`robots.txt` is the only authoritative crawler-control file.** `llms.txt` has no effect on crawling. - **Do not add `noindex` or `Disallow` for AI crawlers** unless the user explicitly wants to block AI indexing. - Prefer primary platform documentation (Google Search Central, Bing Webmaster Tools, Schema.org) over third-party summaries. ## Examples ### Example 1 — Audit request **User:** "Can you audit my blog for AI search visibility?" **Actions:** 1. Check `robots.txt` → `OAI-SearchBot` is missing a `Disallow` but also missing an explicit `Allow` — confirm default is allow. 2. Validate JSON-LD on the homepage → `datePublished` is missing, `author` has no `url`. 3. Run pillar sweep → Trustworthy: partial (no author bio page); Quotable: fail (no FAQ schema on FAQ page). 4. Return findings table with three priority tiers. **Result:** Prioritized list: fix `techarticle.jsonld`, add author bio, add `FAQPage` schema. Clear, actionable, no ranking promises. ### Example 2 — Create request **User:** "Create a new GEO-optimized article page for my Next.js blog." **Actions:** 1. Draft `
` from `templates/page-metadata.html`. 2. Generate `templates/techarticle.jsonld` filled with real title, author, dates. 3. Structure content using `templates/quotable-article-outline.md`: direct-answer intro, H2/H3 sections, sources list. 4. Confirm `robots.txt` allows `OAI-SearchBot`. 5. Run checklist — all eight items pass. **Result:** Ready-to-deploy page with correct metadata, valid schema, and citation-ready prose. ### Example 3 — llms.txt request **User:** "Write an llms.txt for my documentation site." **Actions:** 1. Inventory the three or four most useful pages for an AI agent. 2. Apply `templates/llms.txt` format: H1 site name → blockquote description → `## Key pages` with Markdown links → optional `## Technical files`. 3. Remind the user that `llms.txt` is not a crawler-control file and doesn't guarantee citations. **Result:** A concise, standards-compliant `llms.txt` with honest caveats. ## Troubleshooting | Symptom | Likely cause | Fix | | ---------------------------------------------------- | ------------------------------------------------------------------------ | -------------------------------------------------------------------------------------- | | Rich Results Test shows no schema | JSON-LD is in a JS-rendered `