GEO explained: How your company becomes visible in AI answers

More and more people no longer type search terms but ask ChatGPT, Perplexity or Google's AI Mode a question. Instead of a list of ten links, they get a complete answer, often with two or three recommendations and their sources.

That changes the goal. Ranking first in search results is still valuable, but what really matters is whether your company is named in the answer. That's exactly what Generative Engine Optimization (GEO) is about. In this article, we show how these systems choose their sources, what current research shows and which levers we use to increase our clients' visibility in AI answers.

AI search has arrived in German-speaking markets

AI answers are no longer an experiment. Google shows AI Overviews (in German, “Übersicht mit KI”) above search results and offers AI Mode in German as well. ChatGPT, Perplexity and Gemini answer purchase-related queries with specific recommendations.

For companies, this means more and more purchase decisions start with a question to an AI. If you're missing from those answers, you're invisible to some of your future customers.

Fewer clicks, but visitors with clearer intent

Search clicks drop as soon as an AI summary appears. The Pew Research Center analyzed browsing data from 900 US adults in March 2025: with an AI summary, users clicked a result on 8% of searches; without one, on 15%. They clicked a source within the summary itself on just 1% of searches.

But in practice, we also see the other side: people who still click after reading an AI answer usually want to act, not just learn. Fewer clicks don't necessarily mean fewer inquiries.

GEO is SEO with a new goal

In its July 2026 guide to AI search, Google explicitly states that optimizing for generative AI search is still SEO. The usual prerequisites apply to AI Overviews and AI Mode: the page must be indexed and eligible to show with a snippet. According to Google, you don't need additional files, special structured data or to chop your content into small chunks.

What changes is the goal, and with it the priorities:

How language models choose their sources

Language models work differently from traditional search, which ranks pages by how well they match a query. Three mechanisms matter for businesses.

Entities instead of search terms

Language models think in entities: clearly identifiable things like a brand, a person, a product or a technical term. If your company is described inconsistently across the web, a model can't confidently link it to a topic and will name someone else instead.

Meaning over exact wording

Text is converted into numerical vectors (embeddings). Related topics sit close together. If someone asks about CRM software for a trades business, the model looks for content covering invoicing, mobile use and ease of use, even if the phrase “CRM for trades” never appears.

Retrieval before the answer

Most AI search tools use retrieval-augmented generation (RAG): the system first retrieves current sources, then writes its answer based on them. To be cited, you have to be found as a credible source in that first step.

That's why cited sources only partly overlap with traditional rankings. Beyond websites, these systems often draw on forums, specialist portals, video platforms and networks like LinkedIn. A strong ranking helps, but it doesn't guarantee a mention.

What research reveals about visibility in AI answers

The study that coined the term GEO (Aggarwal et al., KDD 2024) tested across 10,000 search queries which changes to a text increase its visibility in generated answers. The result: up to 40% more visibility. Adding quotes from relevant sources, statistics and citations had the strongest effect. Traditional keyword stuffing, by contrast, reduced visibility.

The study confirms what we see in practice: language models favor content that adds something verifiable. A text that says the same thing as a hundred others gives a model no reason to cite it in particular.

Seven levers we use to increase your visibility in AI answers

1. Clear positioning as the foundation

What does your company stand for, and which questions should it be the obvious answer to? Describe your services, target groups and strengths consistently in your own profiles. Incorrect or contradictory information on important external sites should be corrected. We spell brand and product names identically everywhere.

2. Content that answers a question directly

Every section opens with the answer, followed by reasoning and evidence. Comparisons go in tables, enumerations in lists, and headings state the question the section answers.

3. Original data and experience

We back up every claim in the same sentence with a figure or a source. The most valuable figures are the ones only you have: your own analyses, market analyses, project results. When you supply others with original data, you become a source yourself.

4. Mentions on third-party sites

Language models recommend what others recommend. That's why we deliberately build mentions in trade media, interviews, podcasts, comparison lists and industry directories, along with a visible presence on platforms like LinkedIn, YouTube and Reddit.

5. Balanced answers to real scenarios

People ask AI more specific questions than they type into a search engine: not “buy running shoes,” but “Which running shoes work for marathon beginners with wide feet under €150?” Content that addresses scenarios like this, with pros and cons, is more useful to a model than pure promotional copy.

6. Current, clearly dated content

The date in the article, the sitemap and the structured data all match. When we update an article, we change the content, not just the date.

7. Clean technical foundations

AI can only cite what it can read. Pages need to be indexable, load fast and deliver their content without JavaScript, because many AI crawlers don't execute it. Learn more in our article on AI crawlers and the JavaScript problem.

Structured data helps search, but it's not a lever for AI answers

For a long time, structured markup was considered the “language of AI.” Google has since clarified that AI Overviews and AI Mode don't require special structured data. It's still useful for rich results and must match the visible content.

Google Search doesn't use llms.txt files either, according to its own guide. So we don't treat them as a lever.

A clean basic setup still makes sense: Organization with name, logo and profiles, Article with author and date, and BreadcrumbList. A simplified example:

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "GEO: So kommt Ihr Unternehmen in die KI-Antworten",
  "datePublished": "2026-01-21",
  "dateModified": "2026-10-09",
  "author": { "@type": "Person", "name": "Alex Sperber" },
  "publisher": { "@type": "Organization", "name": "UnitedAds", "url": "https://unitedads.de" }
}

How to measure your visibility in AI answers

For a long time, visibility in AI search was hard to measure. That's changing:

• Google Search Console: A dedicated report now shows impressions in AI Overviews and AI Mode. It doesn't yet show clicks or search queries.
• Visitors from AI sources: In Google Analytics, you can analyze visits from chatgpt.com, perplexity.ai, gemini.google.com or copilot.microsoft.com as a separate group. Clicks from AI Overviews and AI Mode still count as organic search there.
• Mentions in answers: Specialized tools measure how often, in what context and with what attributes a company is mentioned by asking a set of questions on a regular basis. We combine them with our own spot checks.

A mention in an AI answer works like a recommendation. That's why we don't judge it solely by the click that follows.

Conclusion: In AI search, recommendations decide

GEO is the logical next step for SEO. The fundamentals stay the same: technically clean, indexable pages with content that answers a question well. What's new is that a model decides whom to recommend, and it weighs credibility across many sources.

If you sharpen your positioning now, publish original data and make sure others write about you, you'll be named in tomorrow's answers. No plugin or handful of AI-generated articles will get you there. It's work that pays off, precisely because your competitors often underestimate it.