GEO optimisation: getting named by AI

GEO optimisation makes AI assistants such as ChatGPT, Perplexity and Google AI name your company in their answers. What matters is less keywords than quotable answers, clean technology and above all evidence from outside, for example in directories and reviews. This guide shows how that works and how you measure it.

More and more people no longer type their question into a search box but put it to an AI assistant. “Which IT provider near me can help?” then no longer appears as a list of results but as a finished answer with one or two names. Whether your company is among them can be checked. We did that on ourselves and show the results openly below, including the uncomfortable ones.

What is GEO optimisation, and how does it differ from SEO and AEO?

GEO stands for generative engine optimisation, meaning optimisation for generative search systems. The aim is that AI assistants name your company in their answers and use it as a source. Online you will meet related terms such as AI SEO, LLM SEO or ChatGPT SEO. At heart the same thing is meant.

The three terms that are most often confused can be told apart like this:

• SEO (search engine optimisation) wants your pages to rank high in a search engine’s results list. Success is a click on your page.
• AEO (answer engine optimisation) wants your answer to be shown directly as an answer box, featured snippet or voice answer. Success is that your text is the answer.
• GEO wants an AI assistant to name, recommend or cite you in an answer it formulates itself. Success is the mention, often without any click.

The three build on one another. A technically sound, well-structured site helps all three. What each needs in addition is described on our pages on SEO, AEO and GEO.

A mention is valuable because an AI answer usually lists only a few providers. Whoever is missing there may not be noticed at all by people who put their question to the assistant first, even if classic search looks good. That is not a forecast, but a reason to check your own visibility soberly once.

How do AI assistants choose their sources?

AI assistants name companies about which they find consistent, verifiable statements in several places. One well-optimised page is often not enough, because the systems also look at what others say about you. The most important factors:

• Outside evidence: mentions on third-party sites often count for more than your own statements. They include industry directories, review portals, trade media and comparison sites.
• Directories and reviews: especially for regional service questions, assistants like to rely on directories and reviews, because there they find providers sorted and tied to a place.
• Quotable answer paragraphs: a paragraph that answers the question in its first sentence is easier to take over than a long running text. This applies to headings as questions, short answers and clear facts such as location, services and contact person.
• Structured data: schema markup (JSON-LD) tells machines unambiguously who you are, what you offer and where you operate.
• llms.txt: a simple text file in the root directory that gives AI systems a compact overview of your content. It is an offer, not a guarantee that it will be used.
• Crawler access: if your robots.txt or server settings lock out the AI providers’ crawlers, your site cannot be read. This can be checked and fixed quickly.

The order matters: technology and answer quality are the entry ticket, outside evidence is often the difference between “present” and “named”.

How do you measure your AI visibility?

You measure it by putting your real customers’ questions to several AI assistants and noting whether you are named and which sources support the answers. A gut feeling does not replace this, because answers differ by assistant and by day. Here is how to proceed:

1. Collect questions: write down ten questions a customer would ask before knowing you, such as “Who offers IT support in my town?”. No brand names, but needs.
2. Ask several assistants: put each question at least to ChatGPT, Perplexity and Gemini. Each assistant draws on different sources.
3. Record mention and source: is your company named, in which position, and which sources does the assistant cite? The sources are the most valuable part of the measurement.
4. Check Google AI Overviews: search the same questions on Google and note whether an AI overview appears and whether you are cited there.
5. Repeat the measurement: answers fluctuate. Repeat the measurement after a few weeks with the same questions to spot changes.

What to record in every measurement

A simple table is enough. For each question and assistant, note:

• the question as asked, word for word, and the assistant,
• whether your company is named and in which position,
• which competitors are named instead,
• which sources are given, with address,
• the date of the measurement.

From the sources named you see where your industry is backed up in AI answers, and therefore where you should appear yourself.

If you would rather not build this yourself, the GEO audit offers a ready-made measurement with a report.

What does our own practical example show?

On 30 September 2026 we measured Laufwerk.cloud ourselves. The result: for ten questions about our services, put to ChatGPT, Perplexity and Gemini, we were named in 3 of 30 answers. All three mentions concerned IT service in Reutlingen and Tübingen, where we ranked first or second each time. In 23 Google AI Overviews that we checked, we were not cited once.

What the assistants cited as sources was revealing. Most often cited were directories and comparison portals such as sortlist.de, agenturtipp.de and wlw.de, not the providers’ own websites. Technology, on the other hand, was not the problem: crawler access works and loading time is fine. What is missing is outside evidence, meaning entries and reviews exactly where the assistants look.

This is a measurement on one day with one set of questions, not a permanent state. But it shows where to start, and it is an example of how even a technically clean presence can be absent from AI answers.

How do we read these numbers? 3 of 30 means we appeared in roughly one answer in ten. The picture is not even, though: for questions with a clear location we were named, for general service questions without a location we were not. Our reading is that the assistants fall back on directories with a location there and that we are too weakly represented in those sources. The 0 of 23 at Google additionally shows that a mention in one system does not automatically apply to the other.

Which steps do we derive from it?

The measurement yields concrete steps that may apply to your company as well. Whether they work we will only know after the next measurement, and we will report on it openly.

1. Identify directories: check which directories and portals the assistants cite for your industry and region, and maintain complete, consistent entries there.
2. Build reviews: ask satisfied customers for a review on the platforms that are consulted anyway.
3. Sharpen answer paragraphs: word service pages so that the core question is answered in the first paragraph, with place, service and contact person.
4. Bundle facts: provide the essential company details in one place, reliably and machine-readable. For us that is the page Facts (in German).
5. Keep the technology up: check crawler access, schema and loading time regularly so that nothing is lost there.
6. Measure again: repeat the same questions after a few weeks and document the change.

The order follows the finding: because the gap lay in outside evidence and not in technology, directories and reviews come first. Anyone who gets a different picture in their own measurement, such as blocked crawlers or missing answer paragraphs, sets the steps differently. The measurement decides where the effort goes first, not a fixed checklist. This matters especially for small teams, because time is limited and every hour should go where the gap actually lies.

We deliberately disclose only the general approach here. How an audit is built in detail is part of how we work and is not in this guide.

What should you realistically watch for with GEO?

Realistically means: GEO is not a switch. No provider can guarantee a mention in ChatGPT, because answers vary by request, assistant and time. Be wary of promises such as “first place in AI”. What is reputable is improving the conditions and measuring the effect.

It is also worth looking at weighting: for local service providers, directories, reviews and a clean link to a place are often more effective than further blog articles. For services that need explaining, on the other hand, well-structured guides are the basis for being cited at all. The two complement each other.

Typical mistakes

• A single sample: one question, one assistant, one day say little. Only several questions and repeats give a picture.
• Brand names in the questions: whoever asks for their own name measures awareness, not findability among people who do not know you yet.
• Texts only for the machine: what does not help people rarely helps assistants. Understandable answers remain the yardstick.
• llms.txt as a miracle cure: the file can help, but it replaces neither outside evidence nor good content.
• Claiming success too early: without repeated measurement, every statement about an effect is only a guess.

Frequently asked questions