GEO & llms.txt:Get Cited by ChatGPT & Perplexity
Generative Engine Optimisation is the new SEO. Here is how to structure your docs so AI assistants quote you, not your competitors.
Search is no longer just ten blue links. A growing share of high-intent questions are answered directly by ChatGPT, Perplexity, Google's AI overviews and Claude — and the brands those assistants cite capture the attention. Generative Engine Optimisation (GEO) is the discipline of structuring your content so AI systems can find it, trust it, and quote it. This guide covers the concrete moves that matter.
DefinitionWhat GEO is, and how it differs from SEO
Classic SEO optimises for a ranked list of links a human clicks. GEO optimises for being the source an AI model synthesises into its answer and cites. The two overlap heavily — clean, crawlable, well-structured content helps both — but GEO adds a specific emphasis: machine-readability, extractable answers, and explicit signals about which content is canonical.
| Classic SEO | GEO | |
|---|---|---|
| Optimises for | Ranked links a human clicks | Being cited inside an AI answer |
| Unit of success | Position on the SERP | A citation / mention in the response |
| Key levers | Keywords, backlinks, page speed | Structure, schema, extractable Q&A, llms.txt |
| Measurement | Clicks, impressions, rank | Citation mentions, branded search lift |
Standardllms.txt: a map for AI crawlers
llms.txt is an emerging convention — a Markdown file at the root of your site that gives AI systems a curated, high-signal map of your most important content, free of the navigation, ads and boilerplate that clutter raw HTML. Think of it as robots.txt's helpful cousin: instead of telling crawlers what to avoid, it tells them what matters.
# liracode.dev
> Private AI infrastructure: every leading model, your data on your
> own disks, one subscription.
## Docs
- [Private AI for Law Firms](/en/resources/private-ai-for-law-firms-keep-privilege-intact): Keep privilege intact.
- [GPU Rental Cost Guide](/en/resources/gpu-rental-cost-guide-own-vs-rent-2026): Own vs. rent in 2026.
- [GDPR & the EU AI Act](/en/resources/gdpr-eu-ai-act-ai-infrastructure-checklist): Infrastructure checklist.
## About
- [Trust Center](/en/security): Security posture and controls.A common pattern is to publish both a concise llms.txt (the map) and an llms-full.txt (the full text of your key docs concatenated), so an assistant can pull complete context in one fetch without crawling your whole site.
SchemaStructured data: speak the machine's language
Schema.org JSON-LD is how you tell a machine, unambiguously, what a page is. For GEO the highest-leverage types are Article / TechArticle (this is a document, here's its author and date) and FAQPage (here are extractable question-answer pairs).
FAQ schema is especially powerful because it hands the assistant exactly what it wants: a clean question paired with a self-contained answer. Structured data is consistently associated with a meaningful uplift in how often content is surfaced and cited — and a well-formed FAQPage tends to be the single highest-return piece of schema you can add. Plan against a roughly 2× uplift from structured data overall, with FAQ markup at the top of the range.
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "What is GEO?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Generative Engine Optimisation structures content so AI assistants can find, trust and cite it."
}
}]
}CraftWrite content an AI can quote
Beyond the technical signals, the prose itself decides whether a model can lift a clean answer. The patterns that get cited share a shape:
- Answer first — lead each section with a direct, self-contained answer, then elaborate. Assistants extract the first clear statement.
- One idea per heading — use descriptive H2/H3s that match how people ask questions.
- Explicit definitions — state plainly what a term means before you use it; models quote definitions.
- Real data and specifics — concrete numbers, named approaches and worked examples get cited over vague claims.
- A genuine FAQ — a short Q&A block, marked up with FAQPage schema, that answers the questions people actually ask.
Every article in this hub follows these rules: descriptive slugs, Article + FAQPage JSON-LD, semantic headings, answer-first sections and an honest FAQ. The hub is its own proof-of-concept for GEO.
MeasurementHow to measure GEO
GEO needs a metric classic SEO doesn't: the citation. Track how often AI assistants mention or cite your content, separately from clicks and impressions, then watch for the downstream effects — direct clicks from AI answers and a lift in branded search. Treat citation mentions as a leading indicator and expect a lag of a few weeks between publishing structured content and seeing the assistants pick it up.
Frequently asked questions
GEO is the practice of structuring content so AI systems like ChatGPT, Perplexity and Google's AI overviews can find it, trust it, and cite it in their answers. Where classic SEO optimises for a ranked list of links a human clicks, GEO optimises for being the source an AI synthesises into its response and credits.
llms.txt is an emerging convention: a Markdown file at the root of a website that gives AI systems a curated, high-signal map of the site's most important content, free of navigation and boilerplate. It complements robots.txt — instead of telling crawlers what to avoid, it tells them what matters. A common pattern is to pair a concise llms.txt with an llms-full.txt containing the full text of key documents.
Yes. FAQPage JSON-LD hands an assistant exactly what it wants — a clean question paired with a self-contained answer — making the content easy to extract and quote. Structured data overall is associated with a meaningful uplift in how often content is surfaced and cited, with well-formed FAQ markup typically at the top of that range; plan against roughly a 2× uplift with FAQ at the upper end.
Track AI-citation mentions — how often assistants reference your content — as a metric distinct from clicks and impressions, then watch the downstream effects: direct clicks from AI answers and a lift in branded search. Expect a lag of a few weeks between publishing well-structured content and assistants beginning to cite it.
Own your baseline, rent the peaks
Track the live GPU market so you never overpay, and own the steady load where the economics flip. We'll help you model the crossover.