Most GEO advice starts with content. That is backwards. If the infrastructure is wrong, the best content on earth is invisible to the engines it was written for. Run these checks in order.
1. The Curl Test — Is Your Content In The Raw HTML?
Open a terminal and run curl -s yourdomain.com. Read the output. If your headline, service list, and prices are not there — if all you see is meta tags and an empty root div — then every AI crawler on earth sees the same nothing. JavaScript-only rendering is the single most common and most fatal GEO failure.
The fix is server-side rendering or build-time prerendering. Every public route should serve complete HTML before a single script runs.
2. robots.txt — Are AI Crawlers Actually Allowed In?
Many sites block AI crawlers without knowing it, either through default security settings or a blanket disallow. The crawlers that matter: GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Google-Extended (Gemini and AI Overviews), and CCBot (Common Crawl, which feeds many training sets). Each should be explicitly allowed.
3. llms.txt — The File Written For The Machine
llms.txt sits at the site root and speaks directly to language models in plain language: what the business is, exactly what it sells, real prices, credentials, differentiators, and the specific questions it wants to be the answer to. It removes ambiguity the model would otherwise have to resolve by inference — and models reward sources that do not make them guess.
4. Structured Data — Machine-Readable Facts
JSON-LD schema is how a page tells engines what it means, not just what it says. The core set for a service business: Organization and ProfessionalService with a real OfferCatalog including prices, FAQPage for extractable question-answer pairs, and speakable markup pointing at the page's key summary elements.
One rule outranks all the others: the schema must agree with the visible page, and both must agree with llms.txt. Engines cross-check. A contradiction anywhere lowers confidence everywhere.
5. Fact Consistency Across Every Surface
Same business name, same prices, same service descriptions on the site, the schema, Google Business Profile, and every directory. This is unglamorous data hygiene, and it is weighted heavily — an entity that agrees with itself across surfaces is safe for an engine to recommend.
6. Quick-Answer Content Structure
Every important page should open with a direct, liftable answer to the question the page exists for — one or two sentences a model can quote without surgery. Detail and story follow after. If a human skimming only the first paragraph gets the answer, so does the model.
The Order Matters
Checks one and two decide whether engines can see you. Three and four decide whether they understand you. Five decides whether they trust you. Six decides whether there is anything worth quoting. Teams that jump straight to content are building on a layer the engines cannot reach.
UM Media runs this entire checklist as part of every web build — it is infrastructure, not an upsell. See how sites are built citable from day one, or read how engines decide who to recommend.