Generative Engine Optimization: the GEO checklist

Generative engine optimization (GEO) is to AI assistants what SEO is to search engines: the work of making sure a machine can find, read, and accurately represent your site. The difference is the surface you compete for. A search engine returns ten links and your goal is a high position. An assistant returns one composed answer with a handful of citations, and your goal is to be one of them, described correctly.

The good news: GEO is far more mechanical than SEO. Assistants rely on crawlers and extraction, and crawlers follow rules you can verify yourself, without ever sending data to an AI provider.

Four things decide whether an AI engine can use your site at all:

• Crawler access. GPTBot, ClaudeBot, PerplexityBot, and Google-Extended each obey robots.txt. One site-wide Disallow line and you are invisible to that assistant.
• Text without JavaScript. Crawlers read raw HTML. If your content only appears after scripts run, an assistant sees an empty page. This is the single most common GEO failure on modern sites.
• Extractable structure. A clear title, an honest meta description, an H1 that states the topic, and JSON-LD structured data make it easy for a model to describe and cite you correctly.
• llms.txt. A small Markdown file at the root of your site that lists the pages you want machines to read. Early but cheap: engines that parse it get a curated map instead of guessing.

Once a crawler can read your pages, the question is what it can extract. Models quote what is easy to quote: a title that states the subject, a meta description that summarizes it, headings that form a clean outline, and paragraphs that answer a question directly before elaborating. A useful exercise is to look at the raw HTML of your five most important pages and ask what a machine could cite from it. If the answer is "the title and not much else", your best content is invisible to the machines that recommend you.

You can cover the whole GEO baseline in one sitting:

1. Check robots.txt: allow the AI crawlers you want to be visible to, block the ones you do not.
2. Audit your top pages as raw HTML: is there real text, one H1, a sensible heading outline?
3. Add JSON-LD structured data where it fits (Product, Article, FAQPage, Organization).
4. Publish an llms.txt listing your key pages, and keep it current.
5. Re-check after every major site change, and watch your AI referral traffic to see the effect.

MetriXs automates all five: the visibility audits check crawler access, text, and structure per page, the bot view shows exactly what a crawler could quote from each page, the llms.txt generator drafts the file from your sitemap and top pages, and a weekly re-audit alerts you when something regresses. All deterministic, nothing sent to an AI provider.

GEO work only matters if it sends visitors. AI referral insights show which assistants send traffic, to which landing pages, and on commerce sites how much revenue those visits carry. Pair that with the audit score and you have the full loop: make yourself readable, watch the referrals arrive, and fix what regresses.