GEO: How to Check in 30 Minutes Whether AI Systems Cite Your Website

What GEO actually is

GEO prepares content so that a language model recognizes it as a solid source, quotes from it, and names the brand while doing so. It is not a second rulebook next to SEO. It is a different target metric.

With SEO, success is a click. Someone searches, sees your result, clicks, lands on your page. With GEO, success is often not a click but a mention. Someone asks ChatGPT for a recommendation, gets three names, and yours is one of them. The contact happens later and directly.

That has an uncomfortable consequence few people say out loud: GEO can be working while your traffic falls. If you only watch sessions, the two look like a contradiction. They are not.

The difference to SEO, concretely

• Target metric. SEO measures position and clicks. GEO measures mentions, citations and whether an AI knows your brand at all.
• Unit of optimization. SEO optimizes a page for a search term. GEO optimizes a passage for a question. A model rarely cites a whole page, it cites three sentences.
• Trust signal. With SEO, links carry a lot of weight. With GEO, what is written about you elsewhere matters as well, because models learn from third-party sources and often pull comparison lists when answering.
• Timing. A ranking moves over weeks. An AI answer can change within days, because it fetches sources live.

So GEO is not a replacement for SEO. The technical foundation is the same: if you are not crawled, not understood, and say nothing verifiable, you do not appear in either world.

Why this matters now

We measured the demand on 28 July 2026 rather than guessing it. The figures below are for the German market, where the term is furthest along, but the pattern is the same in English.

• The German term for GEO is searched around 1,000 times a month, most recently 1,300. That is 171 percent up on the year before.
• In Google's results for that term, the AI overview sits in position 1, above the first organic result.
• That AI overview cited six sources: four agency guides and two YouTube videos. Not one of them was a service page.
• Of the eight organic results on page 1, exactly one is a service page. The rest are five guides, one news article and Wikipedia.

There is a very practical conclusion in that. If you want to be named in this environment, you need a page that answers a question, not a page that advertises an offer. Which is exactly why you are reading a guide and not a sales page.

The seven checks

Work through them in order. The first five need nothing but a browser.

1. Is there a quotable answer at the top of the page?

Why: A model cites passages, not pages. If the first sentences of your page are a claim without substance, there is nothing to quote.

How to check: Read the first three sentences of your most important page. Do they answer the question someone would ask, without prior knowledge and without the rest of the page?

Good looks like: 40 to 60 words, directly after the heading, brand name in the first sentence, one clear statement. Not "We are your reliable partner".

2. Does every page have exactly one main heading?

Why: The heading structure tells the machine what the page is about and what is subordinate. Two main headings on one page cancel that statement out.

How to check: Open the page, right click, view page source, search for `<h1`. There should be exactly one hit.

Good looks like: One H1, H2 for sections below it, H3 for subsections, with no level skipped.

3. Is structured data in place?

Why: Structured data states explicitly who you are, what the page is, and where it sits in the site. It is the one place where you record that unambiguously for machines.

How to check: Search the page source for `application/ld+json`. Check that at least `Organization`, `WebPage` and `BreadcrumbList` appear.

Good looks like: All three on every page, prices as numbers rather than text, every URL carrying its domain exactly once.

4. Is there an llms.txt?

Why: It is a short, machine-readable map of your site. It does not force a citation, but it makes it easier to place you correctly instead of guessing.

How to check: Open `yourdomain.com/llms.txt`. An error means you do not have one.

Good looks like: One sentence explaining what the business does, then the most important pages with a description, grouped by services, industries and company.

5. Does robots.txt let the AI crawlers through?

Why: If you block GPTBot, ClaudeBot, PerplexityBot or Google-Extended, you are absent from those answers. This happens by accident more often than on purpose, usually through an overly broad blocking rule.

How to check: Open `yourdomain.com/robots.txt` and look for those names. Watch in particular for `Disallow: /` underneath any of them.

Good looks like: Either no block at all, or a deliberate decision you can justify. Both are defensible, but it should not be a surprise.

6. Are there verifiable facts in the text?

Why: Models prefer statements that can be supported. A number with a date and a source is quotable. A superlative is not.

How to check: Take your most important page and count. How many concrete numbers, timeframes, methods or names does it contain? And how many adjectives like leading, innovative, holistic?

Good looks like: More numbers than adjectives. Every number with context, meaning where it comes from and when it was measured.

7. Are you mentioned elsewhere?

Why: This is the one point you cannot solve from your own website, and the most effective one. The AI overview in our example cited four third-party guides. Getting into lists like those costs no domain authority.

How to check: Ask ChatGPT, Perplexity and Gemini for the best providers of your service in your region. Note which names come up and which sources are cited. That is your starting position.

Good looks like: Your brand appears. If it does not, that list is a concrete target list of who to approach.

Our own position, in the open

We measured our own website against exactly these points on 28 July 2026. All 84 pages, not a sample.

• Organization structured data: 84 of 84 pages
• Exactly one main heading: 84 of 84 pages
• Language markup for the second language version: 84 of 84 pages
• Breadcrumb markup: 80 of 84 pages
• Question and answer markup: 20 pages
• llms.txt present and reachable

And what is not right yet, because a checklist without open items is not a check:

• The quotable short answer is on 19 of 23 content pages. The four without it are contact, imprint, privacy and the accessibility statement. There is no search question to answer there, so it stays that way.
• Our robots.txt names no AI crawler explicitly. It allows them through the general rule. The four switches for this are set in our CMS but demonstrably do not reach the file that is served. We measured that on 28 July by flipping one switch and fetching the file again: it did not change. That one is on us, and it is on the list.

We are writing this down because an agency that sells GEO and does not measure its own website makes exactly the mistake it flags in a client.

What GEO does not do

• It does not create demand. If nobody asks about your service, a perfect answer will not help.
• It cannot guarantee a mention. No provider can promise a citation, because the answer is generated fresh on every request. Anyone promising that is selling something they cannot deliver.
• It does not necessarily bring more visitors. An AI answer often resolves the question completely. The value then sits in the mention, not the click. If you only measure sessions, you will not see it.
• It is not fast. Mentions on third-party sites build over weeks, not overnight.

Frequently asked questions

What is the difference between SEO and GEO?

SEO aims at a place in the results list and therefore at a click. GEO aims at being named and cited inside the answer itself. The technical foundation is the same, the target metric is different: position and clicks against mentions and citations.

Is GEO more important than SEO?

No, it is additional. Classic search still carries most of the demand. GEO matters where research increasingly runs through AI tools, which is comparison and shortlist questions. And without the SEO fundamentals there is nothing for GEO to build on.

How do you do GEO?

In this order: a quotable short answer per page first, then a clean heading structure and structured data, then llms.txt and crawler access, then verifiable facts instead of adjectives, and lastly mentions outside your own site. The first four are a day of work. The last one takes weeks.

How do I measure whether it works?

Record a starting point before you change anything: which questions you ask, which names ChatGPT, Perplexity and Gemini give, and which sources they cite. Repeat the same questions monthly. Your access log also shows whether GPTBot, ClaudeBot and PerplexityBot fetch your pages at all.

The next step

If you go through the seven points and get stuck on several, that is the normal case, not the exception. What matters is the order: without a clean structure, off-page work achieves nothing, and without mentions, the best structure earns no citation.

How we build this is on our GEO page. If you want to know where your website stands today, the free quick check is the fastest way to find out. And if the technical foundation is still missing, it starts with SEO or web design.