SEO Geo

When to Use

Building or auditing any public page/landing/site that must be FOUND — in Google AND in AI answer engines (ChatGPT, Perplexity, Google AI Overviews). Keywords/intent come from marketing-strategy (the ICP); words from brand-voice; OG/social tags from web-assets. Run the checklist before shipping any page.

THE HONESTY RULE (read this first — it governs all of GEO)

Review / AggregateRating, testimonials, and entity facts (founder, ratings, sameAs, awards, foundingDate) are emitted ONLY from REAL data the research/brand provided. NEVER fabricate them — fake review/rating schema earns a Google manual penalty, is dishonest, and teaches AI engines to distrust the brand. If a fact isn't real, OMIT that schema gracefully. A thinner-but- true entity graph beats a rich lie. When in doubt, leave it out.

The 3 layers (do in order)

1. Technical — crawlable, fast, structured. Foundation; nothing ranks without it.
2. On-page — title/meta/headings/content match the searcher's intent.
3. GEO/AEO — get CITED by AI answer engines (ChatGPT, Perplexity, Google AI Overviews, Claude, Copilot) — the 2026 edge. Five moves: let AI crawlers in (a robots.ts that deliberately welcomes GPTBot/ClaudeBot/PerplexityBot/Google-Extended/CCBot/… with a documented opt-out), feed them clean text (llms.txt + llms-full.txt per llmstxt.org), describe your entities (richer + entity schema: BreadcrumbList, SoftwareApplication, WebSite+SearchAction, Article with dates, entity-deep Organization with real sameAs/founder), write extractable answers (self-contained answer chunks, comparison tables, FAQ, definitions, sourced stats), and measure it (AI-referral analytics by referrer hostname → a GA4 event).

Full checklists, the robots.ts/llms.txt patterns + JSON-LD templates: references/checklist.md.

Hard Rules

• Validate every JSON-LD block (valid JSON, real schema.org @type, https://schema.org context) BEFORE shipping.
• Emit schema only for what the page ACTUALLY is, only from REAL data. No Review/AggregateRating unless real; no SearchAction without a search route. Omit gracefully when a fact isn't there.
• One page = one primary intent / keyword cluster. Answer the query in the first 100 words, in the searcher's words.
• Core Web Vitals targets: LCP <2.5s, INP <200ms, CLS <0.1 — measure with Lighthouse/PSI, never guess.
• NEVER keyword-stuff or fabricate. Human first, machine second. <title>/<h1> must match the real content (no bait).
• GEO: write self-contained answer chunks (a clear claim + its supporting fact together) with real numbers/dates — that's what AI engines quote.
• Welcome AI crawlers for a marketing site (so it can be cited) but keep /api/ and sensitive paths blocked for every bot; ship the opt-out toggle so a privacy-sensitive client can decline.
• Keep llms.txt + llms-full.txt IN SYNC with the real pages — a fact in them that contradicts the page is debt.
• Put real datePublished + dateModified on pages/articles and <lastmod> in the sitemap — freshness strengthens a citation.

Output Contract

Return: the meta/title/canonical/OG block, the JSON-LD schema, the heading outline mapped to intent, the Core Web Vitals check, and any gap (missing sitemap, slow LCP, thin content).

Escalate to the heavy pack when needed

For a deep competitive audit (0–100 score, backlink/keyword data, multi-site rank tracking, PDF/Excel reports) the claude-seo pack (by AgriciDaniel) is registered and one command away: claude plugin install claude-seo@agricidaniel-claude-seo It adds a GLOBAL Edit|Write schema-validation hook (safe, local-only) — enable it for a focused SEO push, then claude plugin disable it. For build-time on-page / technical / GEO work, THIS skill is enough.