Skip to content

GEO & AI Search

GEO (Generative Engine Optimization): How to Get Cited by Google AI Overviews, Gemini & ChatGPT

, 3 minute read

Generative engine optimization is the practice of making a page easy for AI systems to retrieve, understand and cite. It rests on four things: crawler access, server-rendered HTML with clear structure, self-contained answers placed directly under question headings, and consistent facts about the business across its own site and third-party sources.

Start with access, not content

Most advice on AI visibility begins with how to write. Begin lower down. An answer engine cannot cite a page it could not fetch. Think of it as a stack, where a failure at any layer hides everything above it:

  1. Firewall and CDN. Many security settings block unfamiliar bots by default. Check that yours is not turning away AI crawlers you want.
  2. robots.txt. Decide, per crawler, what is allowed.
  3. Raw HTML. The content must be in the server response. Many AI crawlers do not run JavaScript.
  4. Indexing and structure. The page must be indexable, with headings that say what each section answers.
  5. Entity schema. Structured data that confirms who is speaking.
  6. llms.txt. Optional, and last.

Most sites that are invisible to AI systems have a problem in the first three layers, not the last three.

Know which crawler does what

AI companies run different bots for different jobs. Some collect training data. Others fetch a page at the moment a user asks a question, in order to answer and cite it. Blocking the first kind is a policy choice. Blocking the second kind removes you from live answers.

OpenAI, for example, documents GPTBot for training and separate agents for search and for user-initiated browsing. Google uses the Google-Extended token to control use of content for its Gemini models; it does not affect Google Search. Perplexity documents PerplexityBot. Read each company's documentation, then write explicit rules instead of relying on a wildcard.

How a passage gets chosen

Google has been clear that AI Overviews draw on its ordinary index and ranking systems. Its documentation says there is nothing special to add: a page needs to be indexed and eligible to appear with a snippet. In practice, these systems retrieve several relevant passages and assemble an answer from them.

A passage is more likely to be used when it:

  • Answers one specific question
  • Makes sense without the surrounding paragraphs
  • Names its subject instead of saying it or they
  • Contains something checkable: a number, a condition, a named standard
  • Sits directly under a heading that matches the question

This is why we place a 40 to 60 word answer under every primary heading and write it to be quoted alone. The method is on our direct-answer engineering page, and it depends on the page being built from semantic HTML5.

Consistency is the off-page half

When an assistant is asked to recommend a business, it is not ranking pages. It is drawing on everything it has read about the candidates. A business whose name, services, city and phone number agree across its site, its Google Business Profile, review platforms and directories is easy to describe with confidence. One with three addresses and two names is not.

So the unglamorous work matters: one canonical set of facts, published identically everywhere, and expressed on the site as a connected JSON-LD entity graph. Then earn mentions in the places your trade is discussed. We call this entity co-occurrence, and for local businesses it overlaps almost entirely with good profile optimization.

What does not work

Hidden text addressed to AI models. Pages of generated questions nobody asks. Schema describing content that is not on the page. Paying for placement on best-of lists with no editorial standards. These are the same manipulations search engines learned to discount years ago, and systems built on top of search inherit that skepticism.

Measuring it

There is no rank to track. Build a fixed list of twenty or thirty questions your customers actually ask, run them through the main assistants monthly, and record who is cited. Watch referral traffic from AI products in analytics. And track calls, because a recommendation that produces a phone call never shows up as a session.

The uncertain parts of this field, including the status of llms.txt, are covered in our piece on llms.txt and direct-answer blocks.

Questions this article raises

Is GEO replacing SEO?

No. AI Overviews are built on Google's index and ranking systems, and other assistants rely on search results too. GEO is a stricter application of the same fundamentals.

Do I need special schema for AI Overviews?

No. Google states that no special markup is required. Accurate structured data still helps systems confirm who you are and what the page covers.

How long does it take to get cited?

It varies by system and by query. Changes to crawlable content can be reflected within days in systems that fetch live pages, and much more slowly in anything that depends on model training.

Talk to the person who will actually do the work

No account coordinator, no discovery deck. Call and describe what the phone should be ringing for, or send the site and we'll trace where the signal is leaking.

Call 747-249-9821
Call 747-249-9821