What Is GEO and How to Get Your Company Cited by ChatGPT

GEO (Generative Engine Optimization) is the set of practices that makes a website readable, understandable and citable by AI assistants such as ChatGPT, Google Gemini, Perplexity, Claude and Microsoft Copilot. Where SEO competes for a slot in a list of ten blue links, GEO competes to be the source used to write the answer the user reads first.

The distinction matters because the shape of a search has changed. Someone who used to type "dentist implant price" now asks a full question and gets a paragraph back, with three or four sources cited underneath. If your company is not one of those sources, the visit never happens — and you never see it in your analytics, because there was no click to lose.

What changes between SEO and GEO

Most of the technical foundation is shared: a crawlable, fast site with original content. Three things change in how you execute.

Why AI does not cite your site today

In practice, the reasons repeat. In rough order of how often they turn up in an audit:

• The crawlers are blocked. A generic Disallow inherited from a template, or a robots.txt that never named GPTBot, OAI-SearchBot, ClaudeBot or PerplexityBot. Several of these bots only honour a rule that names them.
• The content only exists after JavaScript runs. Most AI crawlers do not execute scripts. If the text appears after hydration, for them the page is empty.
• The answer is buried. Prices, timelines and definitions sitting three screens down, or hidden behind "request a quote". A model cannot extract what the page refuses to state.
• The entity is inconsistent. A different company name in the footer, in the schema and on the Google profile. Three names read as three companies, none of them authoritative.
• There is nothing verifiable to quote. Pages made of adjectives — "excellence", "innovation", "commitment" — give a model nothing to attribute. A number, a date, a named process does.

What to change, in order

Start with access, because everything else is wasted while the door is shut.

• robots.txt with named rules for the AI crawlers you want reading you. Keep the distinction in mind: -SearchBot and -User agents fetch the page at answer time and cite it with a link; GPTBot, ClaudeBot and CCBot collect for training and grounding.
• Structured data — Organization, LocalBusiness, Service, FAQPage, Article — tied into one graph with a single @id, so every page points at the same entity.
• An llms.txt file: a plain-text index of your canonical pages with one summary line each. It is the map an assistant uses when deciding what to read.
• Answer-shaped content: the definition in the first paragraph, headings that mirror the question as it is actually asked, and a real FAQ with real answers.
• Server-rendered HTML so the content is in the first response, not assembled in the browser.

How to measure something with no ranking

There is no "position 3" in a generated answer, so the metric changes. What is worth tracking: the list of questions that matter in your market and whether you appear in the answer; which competitors are cited in your place; and the traffic arriving from assistant referrers, which shows up separately from organic search in analytics.

Two honest caveats. Answers vary between users and sessions, so a single test is anecdote, not data — you need the same question checked repeatedly over time. And nobody sells placement: any provider guaranteeing a citation is selling something that does not exist.

How long it takes

The technical work ships in days. Citation depends on each assistant's recrawl cycle, so changes typically start showing between 4 and 12 weeks — faster on assistants that fetch pages at answer time, slower where it depends on model training.