What is GEO and how does your company become visible in AI answers?

What is Generative Engine Optimization?

Generative Engine Optimization, GEO for short, means positioning content and brand so that AI systems such as ChatGPT, Perplexity, Microsoft Copilot and Google’s AI answers find them, understand them and name them as a source. A place in a list of links is no longer enough, your company should appear in the answer itself.

The term comes from a study by researchers at Princeton University and IIT Delhi, published in November 2023 and presented at the KDD conference in 2024. Using 10,000 search queries, it tested which changes to a text increase visibility in AI answers. Quotations, statistics and source citations worked best, with up to 40% more visibility. Keyword stuffing, on the other hand, achieved little.

In everyday use, an AI answer differs greatly from a classic results page:

• The answer usually names only a few sources instead of ten blue links
• The user reads the summary and clicks through to a website less often
• The AI combines several sources and decides by itself whom it names
• Contradictory or vague information drops out, specific and evidenced information stays in

Why is GEO becoming important for companies right now?

Because searching in Germany has changed within two years. According to Bitkom, half of internet users at least occasionally use AI chats instead of a search engine, among 16 to 29 year olds it is two thirds. At the same time, Google itself shows AI answers above the search results.

For websites this means fewer clicks from search, and the decision is made more often within the answer. In 2025 the Pew Research Center analysed around 69,000 Google searches in the US. With an AI summary, users clicked a search result in 8% of cases, without a summary in 15%. Only 1% clicked the links inside the summary itself.

Anyone named in the answer still lands on the list of providers someone looks at more closely. Anyone missing is not even asked. That is exactly why GEO pays off even when clicks decline.

Does GEO replace classic search engine optimisation?

No. GEO builds on SEO. In its guide from May 2026, Google explicitly writes that optimising for AI features in search is still SEO. There is neither special markup nor special treatment for AI. What is new is what you pay attention to and where you measure.

How little a ranking alone guarantees is shown by an Ahrefs analysis from March 2026: only 37.9% of the URLs Google cites in AI Overviews rank in the top 10 for the same query. In the background, Google searches for related questions and also takes sources from there. A page that answers many sub-questions of a topic well therefore has better chances than one that serves only a single keyword.

How do ChatGPT, Perplexity and Google choose their sources?

Most AI answers are created in two steps. First the system searches an index for suitable pages. Then the language model reads the results and writes an answer in which it names some sources. Anyone not found in the first step cannot be cited in the second.

1. 1

   ChatGPT

   According to OpenAI, ChatGPT’s web search uses third-party search engines, for business accounts OpenAI names Bing. Crawling is done by the OAI-SearchBot. According to OpenAI, anyone blocking it in robots.txt does not appear in ChatGPT’s search answers.

2. 2

   Microsoft Copilot

   Copilot gets information from the web through Bing’s search service. What Bing does not know, Copilot cannot cite.

3. 3

   Google AI Overviews and AI Mode

   The basis is the normal Google index, crawled by Googlebot. A page must be indexed and allowed to show a snippet to be considered as a source.

4. 4

   Perplexity

   Perplexity searches the web with its own PerplexityBot and shows its sources very visibly below every answer.

For your website this means it must be fully indexed in Google and in Bing, and the AI providers’ search crawlers must be allowed to read it. How to set up Bing in a few minutes is explained in the AI search section of our guide to the cost of a website.

Which measures get your website into AI answers?

These seven points have the greatest influence. The first three are technical and done in a few days, the others need some time and work for a long time.

1. 1

   Allow AI search crawlers

   Check robots.txt and your host’s firewall. OAI-SearchBot, PerplexityBot and Claude-SearchBot should be allowed to read your site. Training crawlers such as GPTBot can be blocked independently.

2. 2

   Be fully indexed in Bing and Google

   Set up Search Console and Bing Webmaster Tools, submit a sitemap, report new pages to Bing via IndexNow. Bing explicitly recommends IndexNow so that changes reach AI answers faster.

3. 3

   Deliver content as readable HTML

   Some AI crawlers see text that is only loaded via JavaScript to a limited extent. Important content belongs directly in the source code.

4. 4

   The answer in the first sentence

   Every section starts with a direct answer to its heading, followed by details. These are exactly the sentences AI systems take over into their answers.

5. 5

   Figures with source and date

   The 2023 GEO study shows the greatest effect for statistics, quotations and source citations. According to the ARD/ZDF media study 2026, 81% of people also consider source citations important in AI answers.

6. 6

   Consistent facts about your company

   Year of founding, locations, services and contacts the same everywhere: on the website, in the Google Business Profile, in directories and in structured data. Contradictions make a source uncertain for the AI.

7. 7

   Mentions on other websites

   AI systems rely on what several sources confirm. Industry directories, review portals, trade articles and client websites that name you count for more than any text on your own site.

What do llms.txt and Markdown versions really achieve?

Less than is often claimed. llms.txt is a text file in the root directory meant to guide language models to a website’s most important content. Jeremy Howard of Answer.AI proposed it in September 2024. To this day it is not an official standard.

In its guide from May 2026, Google explicitly says that you need no separate AI text files, no special markup and no Markdown versions for the AI features of Google Search. They neither help nor hurt there. There is also no confirmed use by OpenAI or Microsoft. Creating an llms.txt is an hour invested in an option. The seven measures above matter more.

From our practice

How we put GEO into practice on pageworkers.com

Our own website is our test bed. Every page is also available as Markdown, plus an llms.txt listing all pages in German and English. We report every new page to Bing via IndexNow at launch, and the structured data comes from the same data as the visible text, so nothing drifts apart.

Since November a tracker measures every week: it asks ChatGPT, Perplexity, Gemini and Google’s AI Overviews the same questions and shows whether and how we are named. For searches for agencies in Aachen, according to Sistrix, Google’s AI Overviews now name Pageworkers first. Nationwide there is still a long way to go, and that is exactly what we work on with content like this.

37questions we test in AI systems every week

4AI systems in the tracker

No. 1in Google AI Overviews for Aachen searches

Malek ArabManaging Director, Pageworkers

How can you find out whether AI systems know your company?

With a simple test you can do yourself in ten minutes. Ask the following five questions in ChatGPT, Perplexity, Copilot and Google Search, each with your company name or your industry and region.

• What does Sample Ltd from your city do?
• Which providers of your service are there in your region?
• Roughly what does your service cost?
• Which provider of your service is recommended, and why?
• What experiences have customers had with Sample Ltd?

Pay attention to three things: are you named, are the facts about you correct, and which websites does the answer link as sources. The sources show you where you need mentions. Since February 2026, Bing Webmaster Tools also show how often your pages are cited in Copilot.

Which mistakes cost visibility in AI answers?

The most common mistakes are technical and quickly fixed once you know them.

Where is the best place to start with GEO?

With the self-test and the technology. Check whether AI crawlers and Bing can read your website, then build out the pages that answer customer questions directly. Mentions on other websites are the slowest but strongest lever.

GEO is not a one-off project. AI systems change quickly, and those who measure regularly see early what works. If you are looking for support, talk to our GEO agency.