GEO: the complete guide to visibility in AI search engines

GEO, short for Generative Engine Optimization, is the set of measures that make sure AI search engines can find your website, understand it and cite it as a source in their answers. Think of ChatGPT, Perplexity, Gemini, Copilot and Google AI Overviews. Where SEO is about a place in a list of ten links, GEO is about a place in a single written answer.

This guide gives you the full overview. Each part has its own in-depth article with examples and steps. If you come across a technical term, you'll find it explained in the GEO glossary.

#What exactly is GEO?

GEO is optimising your website and content so that large language models use your information and refer to you when they answer a question. The term comes from research by Aggarwal and colleagues at Princeton University, Georgia Tech, the Allen Institute for AI and IIT Delhi. In 2023 they described GEO as a new field and studied which changes to a text increase the chance that a generative search engine uses it.

In practice, GEO has three goals:

1. Access. The crawlers of AI services are allowed to read your pages, and are able to.
2. Understanding. The system grasps who you are, what you offer and which facts belong to you.
3. Citation. Your text is so reliable and well structured that an AI system chooses it as a source and links to you.

You'll come across other names too. AEO (Answer Engine Optimization) covers the same idea. Some people talk about LLM optimisation or AI SEO. Flowproof uses GEO because it is the most established term and the research is named after it.

#Why GEO matters now

GEO matters because a growing share of search queries is answered by an AI system that writes its own answer instead of showing a list of links. Google shows AI Overviews and AI Mode above or instead of the classic results. ChatGPT and Perplexity search the web live and give an answer with a few source citations.

For you as a website owner, that changes the rules of the game. In a list of ten results, position six can still bring in visitors. In an AI answer with three sources, you are either in it or invisible. On top of that, some users no longer click through after reading the answer (a zero-click search). If you are mentioned, that counts as brand awareness. If you are not, the answer often mentions your competitor instead.

So GEO doesn't replace your SEO work. It builds on it and shifts the emphasis from rankings to citations.

#GEO and SEO: what is the difference?

The difference between GEO and SEO lies mainly in the goal and the way you measure it: SEO aims for a high position in the search results, GEO aims to be mentioned in a generated answer. The foundation is largely the same. A page that can't be crawled, loads slowly or lacks clear information performs poorly in both. Google itself says you don't need separate optimisation for its AI features on top of good SEO.

Where exactly they overlap and differ, and where to spend your time first, is covered in the article on the difference between GEO and SEO.

#How AI search engines choose sources

AI search engines use two kinds of knowledge: what the language model learned during training, and what it looks up live at the moment of the question. That live lookup is called retrieval. The system sends out one or more searches, fetches pages and then writes an answer based on what it found. Anchoring an answer in sources it has found is called grounding.

For AI Mode, Google describes a technique called query fan-out: one question is split into several sub-questions, each of which is searched separately. A page that clearly answers one sub-question therefore stands a chance, even if it doesn't rank for the main question.

This leads to two practical lessons for GEO:

• For live answers, your current website counts. If your page is crawled today and is clear, it can appear in an answer next week.
• For knowledge inside the model, your reputation and mentions elsewhere count. You build these more slowly, through good content, mentions on other sites and consistent information about your brand.

How this works per platform, which crawler belongs to which service and what makes a passage quotable is explained in the article on how AI search engines choose sources.

#The four steps to an AI answer

A website only ends up in an AI answer if it passes four steps: access, understanding, trust and conversion. If one is missing, the steps after it fall away. The best content won't help if the crawler is blocked.

Flowproof scans exactly these four steps across nine sections. The rest of this guide follows the same order.

#Access: let AI crawlers in

The first requirement for GEO is that the crawlers of AI services are allowed to visit your website. You give that permission in robots.txt. Many sites block AI bots without realising it, for example through a general rule, a plugin or a security setting at their hosting provider.

The most important distinction is between search bots and training bots. OpenAI, for example, has several:

• OAI-SearchBot indexes pages so they can be shown in ChatGPT's search feature.
• ChatGPT-User fetches a page when a user asks for it in a conversation.
• GPTBot collects content that may be used to train models.

If you block GPTBot, ChatGPT can still mention you in search answers, as long as OAI-SearchBot is welcome. If you block OAI-SearchBot, you disappear from those answers. Google works differently. AI Overviews use the regular Googlebot. With the Google-Extended token you only control whether your content may be used for Gemini models.

Watch your firewall too. A CDN or security service can refuse AI bots while robots.txt allows them. You won't see that in robots.txt, only by sending a request with the user agent of such a bot.

A full overview of all AI bots, example files and the firewall test is in the article on AI crawlers in robots.txt.

#llms.txt: a reading guide for language models

llms.txt is a Markdown text file in the root of your website that briefly tells language models what your site is and where the most important information can be found. The idea was proposed in 2024 by Jeremy Howard on llmstxt.org. The file starts with a title, a one-line summary and then sections with links to your key pages.

You need to be honest about its effect: it is a proposal, not an official standard. Google's John Mueller said in 2025 that it plays no role in Google Search. Other companies and documentation platforms do offer it. Because it takes little effort and has no downsides, it is a reasonable addition. It doesn't replace robots.txt, structured data or good content.

How to build one, with a full example and the pitfalls when publishing, is covered in the article on llms.txt.

#Understanding: make facts unambiguous with schema markup

Schema markup is structured data in your page's code that tells machines in a fixed format what something is: an organisation, a product, an article or a frequently asked question. The most widely used format is JSON-LD, a block of code in the head of your page.

For GEO, the principle of entities matters most. An AI system needs to know that "Flowproof" on your site is the same company as the profile on LinkedIn and the entry in the company register. With the Organization type and the sameAs field you make that connection explicit.

The types that make the difference for most business websites:

• Organization or LocalBusiness: name, logo, address, contact details and sameAs.
• WebSite and BreadcrumbList: the structure of your site.
• Article with author and dates: who wrote what, and when.
• Product and Offer: prices and offers, important for shopping features in AI search engines.
• FAQPage: questions and answers that match exactly how people ask.

Which types to use when, with a full JSON-LD example and how to validate it, is in the article on schema markup for AI.

#Trust: write content that AI cites

Content that AI cites answers a question directly, backs up its facts and shows who is behind the information. Google sums up its quality criteria as E-E-A-T: Experience, Expertise, Authoritativeness and Trustworthiness. Google describes these principles in its guidance on helpful, reliable content. They also fit well with what AI systems choose as a source.

What you do in practice:

• Answer first. Start every section with a sentence that fully answers the question, even without the surrounding context. That is the sentence an AI system can lift.
• Headings as questions. Phrase headings the way your customer would ask ChatGPT.
• Facts with sources. Mention specific numbers, dates and names, and link to the original source.
• Structure. Lists, step-by-step plans and tables are easy to process and are often reused.
• Freshness. Show the date the page was last updated and keep it honest.
• Author. Show who writes the content and why that person or organisation knows the subject.

A detailed approach with a before-and-after example and a checklist is in the article on content that AI cites.

#The technical basics that often get forgotten

A common reason for being invisible in AI answers is that a page's text only appears after JavaScript has run in the browser. Many AI crawlers run no JavaScript, or only a limited amount. They then see an empty page. With server-side rendering or pre-rendering, your text is right there in the HTML the server sends.

Also check these points:

• Speed. Crawlers don't wait forever. An HTML page that arrives within about a second is a safe guideline.
• Sitemap. An up-to-date XML sitemap, referenced in robots.txt, helps crawlers find your pages.
• Indexability. A forgotten noindex keeps a page out of search engines, and therefore also out of AI answers that rely on those search engines.
• Language. A lang attribute on the page states which language and market your content is meant for.
• Canonical. For similar pages, a canonical tag indicates which version is the original.

#How do you measure GEO?

You measure GEO with a combination of prompt tests, traffic from AI platforms and your existing SEO data, because there is no generally available "Search Console for ChatGPT" yet. Choose ten to twenty questions your customers ask and test them monthly in ChatGPT, Perplexity, Gemini and Google AI Mode. Note whether your brand is mentioned and whether your site is linked.

In your web analytics you can also see how many visitors arrive via domains such as chatgpt.com or perplexity.ai. Search Console remains important for the basics: a page that isn't indexed won't show up in AI Overviews either.

There are specialised platforms that track mentions in AI answers at scale. Flowproof deliberately focuses on what you can improve on your own website. That's why it makes sense to use technical checks and citation monitoring side by side.

#Common GEO mistakes

The most common GEO mistake is a website shutting out AI crawlers without the owner knowing. That happens through an overly broad rule in robots.txt or through a security setting that is switched on by default. The second is content that is accurate but written in such a way that no single sentence can be quoted on its own. The table lists the mistakes Flowproof sees most often in scans, with the fix.

#What GEO is not

GEO is not a way to trick AI systems. Hidden text for bots, text posing as instructions to a language model, or pages stuffed with repeated keywords work against you. Search engines and AI platforms recognise manipulation, and Google classes techniques like these as spam. GEO is also no substitute for a good product or good service. An AI system that weighs up reviews and mentions elsewhere also sees what customers say about you.

#Step-by-step plan: how to get started with GEO

Start with GEO by sorting out technical access first, then making your key pages unambiguous, and only then working on new content. In this order you get the most out of your time:

1. Scan your website for access, structure and content, so you know where you stand.
2. Check robots.txt and at the very least allow the search bots of OpenAI, Perplexity, Anthropic and Microsoft.
3. Test your firewall with an AI bot's user agent. Set up bot management so that the bots you allow actually get in.
4. Check that your text is in the HTML without JavaScript being needed.
5. Add Organization or LocalBusiness markup, with sameAs pointing to your main profiles.
6. Rewrite the opening sentences of your five most important pages so they answer the main question directly.
7. Make it clear who you are: about us, team, contact details and authors on articles.
8. Publish an llms.txt listing your key pages.
9. Record a baseline with ten prompt tests and your current traffic from AI platforms.
10. Repeat every quarter: scan again, test the prompts and update outdated pages.

#How to check this on your own site

You can check most points in this guide yourself with your browser, your robots.txt and a structured data validator. It's quicker with a scan that checks everything in one go. Flowproof's free scan looks at your site the way an AI crawler does. It checks twelve AI bots in robots.txt, tests whether your firewall lets GPTBot through, reads your llms.txt and schema markup, and gives concrete advice for each finding. Then start with the section that has the lowest score and work through the steps in this guide from top to bottom.