Skip to content
All articles
AI SEO10 min readUpdated

AI SEO: how ChatGPT, Perplexity and Google AI Overviews choose their sources

AI SEO and GEO: how ChatGPT, Perplexity, Gemini and Google AI Overviews choose sources, which content gets cited and how to measure it.

By Editorial team

AI assistants answer questions for which people used to scan ten blue links. For businesses, this changes the question visibility revolves around: not just “Which position do we rank in?”, but also “Are we mentioned in the answer, and with which content?”.

Optimizing for these systems is called AI SEO, also known as GEO (generative engine optimization). This article explains how ChatGPT, Perplexity, Gemini and Google AI Overviews choose their sources and what follows for content, technology and measurement. Because the systems change quickly, we cite a dated source for every statement about products and features. This article reflects the state of October 2026.

In Germany, Google has shown AI Overviews in search results since the end of March 2025, an AI-generated summary above the organic results (t3n, 26 March 2025). Since October 2025, AI Mode has also been available, a conversational search with follow-up questions (onlinemarketing.de, 8 October 2025). At the same time, more and more people research directly in ChatGPT, Perplexity or Gemini.

This has three consequences for visibility:

  • Fewer clicks per search: Many questions are answered in the response itself. Those who are cited get a share of the clicks. Those who are not cited are invisible for that research, even with a good organic position.
  • Longer, more specific questions: Users phrase full sentences with context, such as “Which agency in Cologne handles SEO and Google Ads for online stores?”. Content has to answer such questions directly.
  • Recommendation instead of list: AI answers often name only a few providers. Whether a brand is mentioned becomes more important than which position it holds.

How do AI assistants compile answers?

An AI answer draws on two fundamentally different sources of knowledge.

Training knowledge: During training, a language model processed large amounts of public text. What was written about a brand frequently and consistently, the model “knows”. This knowledge has a cut-off date and only changes with the next model. It can only be influenced indirectly and over the long term, through a strong presence across the web.

Live web search: For many questions, the assistant also searches the web in real time. It formulates its own queries, retrieves matching pages and selects passages to include in the answer. Google describes this approach for AI Mode as “query fan-out”: a question is broken down into several sub-questions, each of which is searched separately (Google Search Central, “AI features and your website”).

For businesses, web search is the more important lever. It determines which pages are cited and linked, and it responds to changes within weeks.

Source selection follows a recognizable pattern. First, a page must be in the system’s search index. Then it must be relevant to one of the sub-questions. Finally, it must contain a passage that can be taken into an answer directly. If a page fails at any of these three stages, it is not cited, however good it may be overall.

Differences between ChatGPT, Perplexity, Gemini and Google AI Overviews

The systems work similarly but draw on different indexes. This has direct consequences for optimization.

System Where the sources come from What counts for you
ChatGPT Own crawler OAI-SearchBot plus external search providers Access for OAI-SearchBot, good discoverability in Bing as well
Perplexity Own index via PerplexityBot, plus fetching individual pages on user request Access for PerplexityBot, current and well-sourced content
Gemini Based on Google Search Indexing and rankings in Google
Google AI Overviews and AI Mode Google index, pages must be indexed and eligible for snippets Traditional SEO, no special markup required

Some details are particularly relevant in practice:

  • ChatGPT: OpenAI distinguishes three crawlers. GPTBot collects training data, OAI-SearchBot builds the index for ChatGPT search, ChatGPT-User retrieves individual pages when a user asks for them. Sites that block OAI-SearchBot will not appear in ChatGPT search answers (OpenAI, “Overview of OpenAI Crawlers”).
  • Perplexity: Perplexity explicitly recommends allowing PerplexityBot in robots.txt so that a website can appear in its results (Perplexity, “Perplexity Crawlers”).
  • Google: According to Google, AI Overviews and AI Mode have the same requirements as regular search. A page must be indexed and eligible to be shown as a snippet. Additional files or special markup are not required (Google Search Central, “AI features and your website”).

It follows that ranking well in Google gives you a good starting point for AI Overviews and Gemini. For ChatGPT and Perplexity, crawler access and visibility in other indexes also matter.

Which content gets cited?

AI systems do not take over entire pages, but individual passages. Content that gets cited therefore shares three characteristics.

Clear statements: A citable passage answers a question completely, in two to four sentences, without the reader needing the rest of the page. Definitions, concrete recommendations and figures with a source are taken over more often than general introductions. Start sections with the answer and give the reasoning afterwards.

Structure: Headings phrased as a question or a precise topic help the system find the right passage. Lists, tables and step-by-step instructions are easier to take into answers than continuous text. It is important that this content is in the HTML. Many AI crawlers do not execute JavaScript and may not see content that is loaded later.

Freshness: For questions about prices, products, rules or developments, the systems prefer current sources. A visible date, regularly revised content and specific years signal that a page reflects the current state.

An example: asked “How long does SEO take to work?”, an assistant is more likely to cite the sentence “Usually three to six months, longer in competitive markets” than a paragraph that begins with the history of search engine optimization.

The role of mentions on trusted sites

Your own website is only part of the picture. When an assistant makes a recommendation, for example for a service provider or a product, it often relies on third-party sources: trade articles, comparison portals, industry directories, review platforms, forums and media coverage.

A brand that only talks about itself on its own website appears less reliable to these systems than a brand that is described consistently by independent sources. Three types of mentions have proven particularly effective:

  • Expert mentions: Guest articles, interviews and quotes in trade media of your own industry.
  • Structured profiles: Entries in relevant directories, on review platforms and in partner lists with a consistent name, range of services and location.
  • Comparisons and lists: Articles like “The best providers for …” are frequently used by AI systems when users ask for recommendations.

Consistency is decisive. If the website, profiles and third-party sources describe a brand with the same key statements, the system can identify it unambiguously and recommend it with more confidence. Building targeted mentions is therefore part of link building and digital PR, even if not every mention carries a link.

Structured data and technical foundations

Structured data based on Schema.org is not a switch for AI visibility. Google stresses that no special markup is needed for AI Overviews. It still helps because it makes information unambiguous: who the company is, where it is based, which services it offers, who wrote an article and when it was published.

The technical foundations at a glance:

  • Crawler access: Configure robots.txt so that OAI-SearchBot, PerplexityBot and Googlebot may fetch the relevant pages. Whether you allow GPTBot for training is a separate decision.
  • Indexing: Important pages must be indexed in Google and must not carry nosnippet, otherwise they are not available for AI Overviews.
  • Server-rendered content: Text, prices and facts directly in the HTML, not loaded later via JavaScript.
  • Structured data: Organization, LocalBusiness, Product, Article, FAQPage and BreadcrumbList where they fit the content.
  • Clear freshness signals: Publication and modification dates in the content and in the structured data.

In addition, some websites publish an llms.txt file with an overview of their most important content. It is not an official standard and has not been confirmed as a ranking factor by the major providers. It does no harm, but it does not replace any of the foundations above.

How can visibility in AI answers be measured?

Measurement is the least mature part of AI SEO. There is no complete data source, but several partial views that together give a usable picture.

Google Search Console: Impressions and clicks from AI Overviews and AI Mode are included in the regular performance report under “Web”. Since June 2026, Google has also been introducing dedicated reports for generative AI features that show how often URLs appear in AI Overviews and AI Mode. They are being rolled out gradually and currently contain impressions only (Google Search Central Blog, 3 June 2026).

Web analytics: Visits from ChatGPT, Perplexity, Gemini and Copilot appear as referrals from their domains. ChatGPT often adds the parameter utm_source=chatgpt.com to linked URLs. Set up a dedicated “AI assistants” channel in your analytics tool so that these visits do not get lost in general referral traffic.

Sampling and prompt monitoring: Define 20 to 50 questions your target group actually asks and check regularly whether and how your brand appears in the answers. Specialized tools automate this. It is important to read the results as a trend. Individual answers fluctuate because the systems do not choose the same sources for every request.

Asking new leads: Add the option “AI assistant” to the question “How did you hear about us?” in your contact form. Many recommendations from AI answers do not lead to a measurable click, but to a later direct inquiry.

What sets AI SEO apart from traditional SEO, and what stays the same?

Our assessment from practice: AI SEO is not a new discipline, but an extension of SEO with a different emphasis. The foundations stay the same, the yardstick for success changes.

What stays the same:

  • Without technical foundations, indexing and relevance there is no visibility, neither in traditional results nor in AI answers.
  • Good content answers the target group’s questions better than the competition.
  • Trust comes from mentions and links from credible sources.

What shifts:

  • From keywords to questions: Content is written for complete questions and their context, not for individual search terms.
  • From pages to passages: Every section must be citable on its own, not just the page as a whole.
  • From links to mentions: Mentions without a link also shape how AI systems classify a brand.
  • From positions to presence: The most important metric is not the rank, but whether and how a brand appears in answers.

Investing in the foundations today improves both: rankings in traditional search and the chance of being mentioned in AI answers. Anyone who sees AI SEO as a shortcut around SEO will be disappointed in both areas. Our approach is described under AI SEO.

Sources

All information on products and features reflects the state of October 2026.

More articles

All articles