What Is GEO? Getting AI Engines to Cite You

1. What GEO is, and how it relates to SEO

GEO stands for generative engine optimization. The goal is that ChatGPT, Claude, Google's AI overviews and similar systems can reach your content when answering a question, and are willing to credit you as the source. SEO wants a higher position and a click. GEO wants your sentence lifted into the answer, with a link left behind.

The foundations overlap heavily. The site has to be crawlable, load quickly, have a clear information structure and say something of substance. SEO and GEO both need all of that. The difference is the last mile: SEO optimises one result in a list, GEO optimises a passage that can be copied into an answer.

So the first move is not to add something new. It is to confirm the SEO basics hold: pages are not blocked from being crawled, every page has a correct title and summary, headings follow a logical order, and the site reads properly on a phone. Those are the foundation for GEO as well. If the basics are broken, nothing you build on top connects.

After that, track "position" and "cited" as two separate measures. They do not move together, and reporting them in one table leaves both unreadable. A useful way to start: write down the questions you want customers to ask, phrased the way they would say them out loud, for example "how much does a WordPress site cost for a small business". Those sentences become the subheadings in your content later on.

2. Why it matters now: the answer arrives and the click does not

The way people ask has changed. It used to be keywords into a search box and three tabs open for comparison. Now plenty of people ask an AI to compare A and B directly, or read Google's AI summary and close the page. The answer finished on the results page, and your traffic report never saw the impression.

Two consequences for a small business. First, your position in a list no longer describes your visibility. Second, if the engine quotes a competitor's explanation instead, you are not even in the comparison. The early sign shows up in Search Console: impressions flat while clicks slide, which usually means the question is being settled before anyone reaches you.

What you can do is concrete. Pricing, specifications, how the service runs, anything customers ask about: write it on the site as plain text. Content that only exists inside a PDF or an image may as well not be written, because a machine cannot read it and so will never use it to answer somebody's question.

Separately, take your brand name, your service names and your common questions to a few AI engines on a regular basis, and record who they cite and which page they cite. That record is closer to reality than any score a tool sells you, and it costs nothing.

3. What an AI engine reads is not what you see

What a machine collects is the document the server sends first, not the screen you get in a browser. On some sites the content only appears once the browser has run the page code, so the machine receives an empty shell. Google is able to wait, but it defers the work. Other AI engines vary, and some do not wait at all.

If the source is empty, the fix is to have the site generate the text of every page before it goes live, so a machine gets complete content the moment it asks. Systems that build the page on the server, WordPress among them, never have this problem in the first place. Sites built with Next.js, Nuxt or Astro are worth checking, because each page needs to be server-rendered or statically generated rather than assembled by the browser.

The same logic applies to anything that only expands on click. Prices, opening hours and contact details that load only after a user taps something are usually out of reach for a machine. Those happen to be the most asked-about details on any site, which makes them a strange place to hide things.

4. Structured data: introducing yourself in a form machines read

Structured data is a short self-introduction written inside the page for machines only, invisible to visitors. It saves the engine from guessing: this string of digits is a phone number, this name is the author, this block is a question and answer. For GEO the benefit is fewer misreadings, so the engine attaches the right facts to the right business.

Four types are enough for a small business. Put Organization on the home page with the name, URL, logo and contact details. If you have a physical location or serve a defined area, use LocalBusiness instead and add the address and the areas you cover.

The other three each have their place. FAQPage belongs on pages that genuinely have a question and answer section, and its shape is the closest thing on your site to the shape of an AI answer, which makes it the easiest block to lift whole. Write the questions the way a person would say them. BlogPosting belongs on article pages and states the title, author, publication date and last updated date, which is what an engine uses to judge credibility and freshness. BreadcrumbList says where the page sits in the site, which helps an engine understand how your content is organised.

5. The shape content needs to be in to get quoted

A generative engine usually lifts a paragraph or a section. If your paragraph only makes sense after the reader has gone back two paragraphs, it means nothing once it has been pulled out. Write each one as an answer that can stand alone: state the conclusion in the first sentence or two, then add the conditions and the exceptions.

That runs against how most people write. The habit is to set the scene and then arrive at the point, but for a machine lifting an extract, those first three sentences are noise and the engine may stop before it reaches the substance. The fix is small: move the concluding sentence to the front of the paragraph and leave the rest in order.

6. The common misreading: not blocked does not mean indexed

This is the part that gets oversold. llms.txt is a proposal still under development. robots.txt is a long-established way of declaring crawler permissions. Both files deal with the same single question: are you allowed to fetch this.

So robots.txt has exactly one job: do not block by accident. Check that you are not disallowing GPTBot, ClaudeBot, Google-Extended and the rest, unless keeping your content out of AI systems is a deliberate choice. llms.txt costs almost nothing, so treating it as a small bonus is fine. Treating it as a guarantee of being indexed is not.

One expectation to set early. There is currently no official way to look up whether a given AI system has indexed your site. If a service guarantees you a listing or guarantees citations, ask them how they intend to verify that. A new domain also needs time to build weight, and once the technical side is maxed out, mentions and links from elsewhere are still the material an engine uses to decide whether you are credible.

7. How to check it yourself: three actions, half an hour

You do not need to buy a tool first. These three actions show you what your site looks like to an AI, and take about half an hour.

Run all three again after a redesign, a framework change or a hosting move. When the environment changes, what the crawler sees changes with it, and that kind of change almost never shows on screen. Only the page source and the AI's own answers report it honestly.