GEO – generative engine optimisation

Generative engine optimisation: how your business appears in AI answers

More and more people ask ChatGPT, Gemini, Perplexity or Google’s AI answer instead of going through a list of results. Generative engine optimisation is about this: if you are named there, you win – if not, you don’t show up at all. I’ll tell you beforehand whether it is even worthwhile for your business.

Since 2004 on the webNo minimum term
Oliver HansteinOliver HansteinWeb designer & SEO
In short: GEO is not some new kind of magic, but good structure, clean technology and mentions outside your own website – judged by what language models can read from them.

What GEO is – and what it isn’t

Generative Engine Optimization (GEO) means making sure your business appears in the answers of language models. In other words, where someone asks “Who builds websites for trade businesses near me?” and gets a ready-made answer with two or three names.

It is often confused with something else: writing content with AI. That is a question of tools and has nothing to do with visibility. You can even see the confusion in search queries – under “ChatGPT SEO” you find almost nothing but tool comparisons. The optimisation discussed here starts with your content, not with the question of what that content was written with. This page is about the other thing: being named.

Along the way you will come across other abbreviations. LLMO (large language model optimisation) and AEO (answer engine optimisation) essentially mean the same as GEO. No single term has caught on yet.

How AI systems generate answers

To understand where GEO comes in, it helps to look under the bonnet. Generative AI systems work in two ways, often combined:

  • From training: the language model reproduces what it learned from millions of texts during training. What was often written about a brand appears in the answer. Recent changes are missing.
  • With live search: the system first sends the question to a search engine, reads the pages found and builds the answer from them – with sources. This is how AI Overviews and AI Mode at Google, Perplexity and ChatGPT with web search work. Specialists call this retrieval augmented generation.

With live search, the system often breaks a question down into several search queries and pulls the results together. Those who rank well in the classic sense for these queries and offer sections that can be quoted are named as a source. That is why good search engine optimisation is the foundation of every GEO strategy.

GEO and classic SEO: what stays the same, what changes

Stays the same

  • Clean technology, fast pages, accessible content
  • Content that fully answers a question
  • Trust through mentions and links

Changes

  • A section is quoted, not the whole page
  • Whole questions instead of short keywords
  • The brand is named instead of a click to the website – visible, but harder to measure

Who it’s worthwhile for – and who it isn’t

Here is the uncomfortable truth missing from most providers’ pages: language models favour large, frequently mentioned brands. They repeat what has often been written about someone – and that is rarely small businesses. Their sources are publicly accessible content: websites, directories, forums, specialist articles.

  • Usually worthwhile: businesses with a name of their own in the market, manufacturers, specialist providers, service providers with customers beyond the region, industries with long decision-making processes.
  • Rarely worthwhile: the trade business that gets its customers from within a twenty-kilometre radius. For local searches, Google still delivers the better results – and that is where people search.
  • Almost never: anyone who has only a few pages and doesn’t appear anywhere outside their own website. There simply isn’t the data an answer could be built from.

If you belong to the second or third group, I’ll tell you in the first conversation. In that case, your budget is far better spent on local visibility .

What I actually do

None of this is magic, and much of it also benefits traditional search engine optimisation :

  • Sections rather than pages. Language models rarely cite a whole page, but rather a paragraph. Each section therefore has to make sense on its own – with a clear subheading and without references to something two screens further up.
  • Mentions outside your website. Industry directories, business profile, reviews, specialist articles, local press. Models learn from what others write about you, not just from how you present yourself.
  • Closing question gaps. People ask an AI full questions, not keywords. I collect the questions being asked in your market and check which ones your website doesn’t answer.
  • Technical accessibility. AI systems send their own crawlers. Can they reach your content, or does the configuration block them? Structured data also helps them assign your offer and area of expertise correctly. This can be checked in ten minutes and is surprisingly often the cause.
  • Measuring. Mentions in answers, search volume for your company name, enquiries with the sentence “ChatGPT recommended you”.

Why sections matter more than pages

AI systems break your content down into blocks of text and look for the passage that fits the question. That block is then cited as the source, not the page. What wins is the paragraph that answers the question completely – not the page with the most words.

A page as one block compared with a page in sectionsOn the left, a page made of one continuous block of text: the language model has to search for the right passage in it. On the right, the same page divided into four sections with their own headings; the matching section is cited as a whole. Continuous text ? The model first has to dig out the right passage — and if in doubt, cites nothing at all. Divided into sections gets cited Each section answers one question on its own — the matching one is taken over as a whole.

In practice, that means: a subheading that names the question. A paragraph below it that answers it without the reader having to have read anything before. Lists instead of convoluted sentences. Exactly the kind of work readers appreciate too – which is why optimisation for AI systems is not an extra project alongside the copywriting, but a yardstick for it.

Where my knowledge comes from

GEO is two years old, so a lot of claims are made about it. I learned the fundamentals at SISTRIX – in the seminar series “SEO und AI – Was ändert sich und was bleibt gleich” (SEO and AI – what changes and what stays the same), with 14 online sessions and a final session in Bonn in June 2026. Alongside the GEO fundamentals, the syllabus covered keyword research in the AI era, visibility in AI systems and the tools for it.

A certificate is no substitute for results – but it tells you that I haven’t pieced the topic together from blog posts.

SISTRIX certificate for Oliver Hanstein for the seminar series "SEO und AI" (SEO and AI) covering GEO fundamentals, completed in June 2026 in Bonn
Seminar series at SISTRIX, completed on 3 June 2026 in Bonn.

Authority: why your brand counts outside your website

Language models prefer to name providers that others write about too. Mentions in specialist articles, business directories, the local press or forums work like recommendations – even without a link. For a GEO strategy, this means: besides your own website, what counts is how often and in what context your name comes up elsewhere. This brand work takes time, but pays off for search, AI and trust at the same time.

llms.txt, robots.txt and the AI crawlers

The AI providers send their own crawlers: GPTBot and OAI-SearchBot from OpenAI, ClaudeBot from Anthropic, PerplexityBot, and at Google the Google-Extended token controls whether content may be used for Gemini. What these crawlers may read is set in your website’s robots.txt. Often a security setting or a plugin locks them out without anyone knowing.

llms.txt is a proposal for a file that gives language models an overview of the most important content. So far, no major provider has confirmed that it evaluates it. I create it when it takes little effort, but don’t expect anything from it – accessible pages and structured data matter more.

Measuring AI visibility

There are now dedicated tools for visibility in AI answers. I work with the AI section of SISTRIX: it shows in which answers from ChatGPT, Gemini, Perplexity and Google’s AI Overviews a domain appears, and for which questions. An example from my own business: hansteinmedia.com currently appears there in around 90 recorded answers, almost all in Google’s AI Overviews – without my ever having optimised for it specifically. That shows how closely GEO and solid search engine optimisation are connected.

Entities: how AI systems recognise your business

Language models don’t think in keywords but in entities – clearly identifiable things such as a company, a person, a place or a product, and the relationships between them. If the information about your business is consistent everywhere – name, address, services, founder, location – AI systems can recognise it as one entity and associate it with the right topics. Contradictory details across your website, directories and profiles do the opposite.

Structured data and schema markup

Schema markup is code in the HTML of a page that describes its content in a machine-readable way: this is an organisation, this is its address, these are its services, this page answers these questions. Search engines and AI systems use it to interpret pages correctly. For most businesses, a handful of types is enough: Organization or LocalBusiness, Service, Person for the people behind the business, FAQPage where it fits, and sameAs links to official profiles.

Content that gets cited

AI-generated answers prefer content that can be quoted without context. In practice, that means:

  • Clear definitions: a short sentence that explains a term, right under the heading that names it.
  • Comparisons: X vs. Y, with the differences in a list or table.
  • Facts with sources: numbers, dates and prices that are specific and up to date.
  • Answers to real questions: the questions people ask in conversational search are longer and more specific than classic keywords – and often show the intent behind them.

Measuring citations and clicks

Being named in an AI answer is visible, but harder to measure than a ranking. Useful signals are:

  • citations of your website in the answers of ChatGPT, Gemini, Perplexity, Claude and Google’s AI Overviews for a fixed set of questions
  • visits from AI assistants in your analytics – referrals from chatgpt.com, perplexity.ai and similar sources
  • searches for your brand name, which often follow a recommendation
  • enquiries in which customers mention an AI assistant

How you can tell whether it’s working

The honest answer first: there is no rankings table. Answers come out slightly differently for every query; there is no position 3. What can be observed:

  • whether your name appears in answers to typical questions in your market – checked regularly, with the same questions, across several providers (ChatGPT, Gemini, Perplexity)
  • whether search volume for your company name increases – anyone who gets a recommendation searches for it
  • whether enquiries come in that name an AI as their source

How it works

  1. Initial audit. I ask language models the questions your customers would ask and see who gets named – you, your competitors or nobody.
  2. Assessment. You get a clear statement on whether the effort is worthwhile for a business of your size and in your market. Even if the answer is no.
  3. Implementation. Structure, technology, content, mentions – in the order in which the optimisation will achieve the most for you.
  4. Re-measuring. The same questions, the same approach, again after a few months.

Frequently asked questions about generative engine optimisation

Do I need an agency for this?

No. The work consists of text structure, technology and mentions – that doesn’t take a team, but someone who understands the website. With me, you talk to the person who also does the work.

Does GEO replace search engine optimisation?

No. By far the largest share of searches still goes through search engines, especially local ones. GEO comes on top; it doesn’t take its place. If you are still invisible on Google, that’s where to start.

Does it work for a local business?

Usually only to a limited extent. Language models favour frequently mentioned brands, and a business with customers from the surrounding area hardly features in that volume. For local questions, Google delivers the better answers – that is where the effort pays off more.

What does it cost?

The initial audit has a fixed price; after that you decide whether to continue. If more work is needed, you are given a total beforehand – no open-ended hourly budget and no minimum term.

Can you guarantee that ChatGPT will name me?

No, and nobody can do that seriously. Which sources a model draws on is decided by the provider, and the answers change constantly. What I can promise is work you can follow and an honest assessment up front.

What is the difference between GEO, LLMO and AEO?

Essentially none. GEO stands for generative engine optimisation, LLMO for large language model optimisation, AEO for answer engine optimisation. Three terms for the same thing – none of them has caught on yet.

What is llms.txt – and do I need it?

A text file in the website’s root directory that is meant to give language models an overview of the most important content. It is a proposal, not a standard, and so far no major provider demonstrably evaluates it. It does no harm; it isn’t a priority.

What are AI Overviews?

The AI summaries that Google shows above the search results. They answer a question directly and link to the sources the answer comes from. Being shown as a source there brings visibility – even if fewer people click than on a classic result.

What is schema markup – and does it help with AI search?

Code that describes the content of a page in a machine-readable way, for example your organisation, address and services. It helps search engines and AI systems interpret your pages correctly. It doesn’t guarantee citations, but it removes ambiguity.

How do I know whether AI assistants send me visitors?

In your analytics, look for referrals from chatgpt.com, perplexity.ai, gemini.google.com and similar sources. Visits from Google’s AI Overviews are counted as normal Google traffic and can’t be separated.

I’ll check whether you appear in AI answers – and tell you honestly whether doing more is worthwhile.

Get your visibility checked →