site-geo-optimization

name

site-geo-optimization

description

Make a website discoverable and citable by AI answer engines (ChatGPT, Perplexity, Claude, Google AI overviews) — i.e. GEO / generative engine optimization, the AI-era cousin of SEO. Use this whenever someone wants their site, blog, or docs to show up in AI answers, asks about llms.txt, "getting cited by ChatGPT", AI SEO, structured data for LLMs, or wants to audit/score how AI-ready their site is. Trigger it even if they just say "I want AI to find my site" or "why doesn't ChatGPT know about my blog". Covers the concrete signals (llms.txt, AI-crawler robots, author identity, schema, FAQ markup), how to score a site, the correctness rules that avoid penalties, and how to prompt an AI to do the work.

Site GEO optimization

What this is and why it matters

People increasingly ask an AI instead of opening Google. The AI answers, often with a link back to a source it trusts. That link is the new front door. GEO is the work of making sure your site can be found, read, trusted, and cited by those answer engines.

The good news: the basics are concrete, mostly a handful of plain files plus some structured data. Much of it can be generated by an AI — the human's job is the judgment calls (which profiles are real, where FAQ markup is honest, keep it simple). Explain it to the user in those terms; most people doing this are not deep technical SEO folks.

The signals (the checklist)

Work these in order; each is independently valuable.

1. llms.txt (+ llms-full.txt) — the llmstxt.org standard. A clean, plain-text "menu" of the site: an H1, a one-line summary, then link lists of pages with short descriptions. llms-full.txt is the same idea but contains the full text of every page in one file. Generate both from the site's content at build time so they stay current automatically. This is the single highest-signal "I'm AI-ready" file.

2. AI-crawler-friendly robots.txt — explicitly Allow: / the AI bots (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, anthropic-ai, PerplexityBot, Google-Extended, Applebot-Extended, CCBot, Bingbot) and reference the sitemap. Many sites accidentally block these. You can't be cited if you can't be read.

3. Author identity (entity + sameAs) — the biggest authority lever. Emit a Person in your structured data with url and a sameAs array linking the author's real profiles (LinkedIn, GitHub, X, YouTube, newsletter). This is how engines recognize the author as a known entity and attribute/cite work by name. The links must be real — a 404 in sameAs hurts more than helps.

4. Core structured data (JSON-LD) — WebSite site-wide; Article on posts with author (the Person above), publisher (Organization), datePublished/ dateModified, image, mainEntityOfPage. Plus the per-page basics: unique <title>, meta description, canonical URL, <meta name="author">, Open Graph

   • Twitter cards.

5. FAQ markup on "how do I…" content — FAQPage JSON-LD is what engines pull into question-style answers. Make it opt-in and visible: render a real FAQ section on the page AND matching FAQPage schema from the same source. Never ship schema that doesn't match visible content — mismatched/invisible FAQ markup gets ignored or flagged. A central /faq hub page aggregating them is a strong bonus asset.

6. Fundamentals that still matter — semantic HTML (one <h1>, real heading hierarchy, <article>, <time>), a sitemap, an RSS feed, fast load (static output helps), and clean readable URLs. AI parsers reward the same clarity humans do.

Scoring a site (audit rubric)

When asked to "score" or audit, fetch the live pages and check real signals — do not score from assumptions. Weighted dimensions that work well:

Dimension	Weight	Checks
LLM-readable files	20%	llms.txt, llms-full.txt, AI-friendly robots, sitemap, RSS present
Crawlability & access	15%	AI bots allowed, HTTPS, clean URLs, canonical, static/fast
Structured data	20%	Article + Person + Organization + WebSite + FAQPage
Content semantics & meta	15%	one h1, headings, meta description, author, dates, OG/Twitter
Authority & citability	15%	author entity with real sameAs, bylines, freshness, internal links
Distribution & feeds	15%	RSS, OG/Twitter, full-text endpoint

Score each 0–100, weight, and report the total with the top 3 highest-leverage fixes. Verify each claim against the live HTML (curl + grep) so the score is defensible. See references/implementation.md for an audit command and the generator code for each file.

Correctness rules (avoid penalties)

• sameAs links must resolve — find the user's real profile URLs (search their site/content) rather than guessing; a wrong link is worse than none.
• FAQ schema must mirror visible on-page Q&A. Don't mark up content that isn't there, and be cautious duplicating the same FAQ across many URLs.
• Generate llms.txt and feeds from the content source so they never drift out of sync with the site.

How to prompt an AI to do this

This is very delegable; the human supplies judgment, the AI does the mechanics. Useful prompt shapes:

Audit first:

"Score my site [domain] for GEO — check llms.txt, robots/AI-crawler access, structured data, author identity, and FAQ markup against the live pages, and tell me the top fixes."

Then implement the highest-leverage items:

"Add an llms.txt and llms-full.txt generated from my posts, an AI-friendly robots.txt, and an author Person entity with sameAs links to my real profiles: [LinkedIn], [YouTube], [GitHub]. Keep them self-updating on each deploy."

Add FAQ correctly:

"On my tutorial posts, add an opt-in FAQ that renders a visible section AND matching FAQPage schema from the same source — only where the Q&A is genuinely on the page."

Keep yourself in the loop:

"Tell me which links you used for sameAs so I can confirm they're mine, and verify each change against the live site after deploying."

Reference implementation

A complete, live build of this (Astro static site): generator endpoints for llms.txt / llms-full.txt, the author Person + Article JSON-LD, an opt-in FAQ component, and a /faq hub — https://github.com/alnutile/astro-and-zapier-chat — with condensed, adaptable code in references/implementation.md.