What is AI search? How it works, systems and visibility

Basics

What is AI search?

AI search is a search technology that processes the meaning, context and intent of a query and generates an answer with sources from it instead of a mere list of links.

The term covers Google's AI Overviews and AI Mode, ChatGPT Search, Perplexity and Bing Copilot. What they all share is the generated answer directly in search.

How does AI search differ from classic search?

Classic search matches search terms with documents and delivers a results list. AI search additionally recognises synonyms, whole questions and the intent behind them.

AI search therefore also answers long, naturally formulated questions at which pure keyword matching fails. The answer summarises several sources instead of only showing ten blue links.

Is AI search the same as an AI chatbot?

No, AI search is not a mere chatbot. It retrieves current sources from the web and substantiates its answer, while a pure chatbot only answers from its training knowledge.

The difference lies in the binding to sources. An AI search refers to verifiable evidence, without which the risk of invented statements rises.

Technology

How does AI search work?

AI search combines classic web search with large language models. The system retrieves matching sources and summarises them into an answer with references.

The principle behind it is called retrieval. First the system searches for relevant text passages, then a language model formulates the answer from them.

What is retrieval-augmented generation (RAG)?

RAG connects a language model with a current web search. Instead of relying only on old training knowledge, the system fetches fresh content and bases the answer on it.

Systems such as Perplexity and ChatGPT Search work this way. That is why they also provide information about current events that lie after their training cut-off.

What are query fan-out and grounding?

In query fan-out, Google's AI Mode breaks a question down into several sub-questions and searches them in parallel. AI search thus finds more matching content than a single search query would deliver.

Grounding means that the AI binds its answer to the sources found instead of formulating freely. Every statement can thus be traced back to a real search result.

What role do embeddings play?

Embeddings translate words and sentences into numerical vectors that represent meaning. The system thus recognises that two formulations mean the same, even without identical words.

This semantic proximity is the core of AI search. It finds thematically matching content where classic search depends on the exact choice of words.

How does AI search choose its sources?

AI systems do not rank content like Google, but prefer statements that occur frequently and consistently in trustworthy sources. Structure and unambiguity increase the chance of a hit.

Decisive is the citation worthiness of a passage. An AI system is more likely to adopt as a source a section that answers a question completely and with evidence.

Systems

Which AI search systems are there?

The most important AI search systems are Google's AI Overviews and AI Mode, ChatGPT Search, Perplexity and Bing Copilot. They differ in access, freshness and scope of sources.

Google integrates the AI answer into the familiar search, the other systems run as a separate interface. What they all share is the answer with source references.

Google AI Overviews and AI Mode

Google's AI Overviews (formerly Search Generative Experience) show an AI summary above the classic results. AI Mode is the more comprehensive variant with its own answer area.

AI Mode has been available in Germany, Austria and Switzerland since 8 October 2025. Google introduced it at the same time in 36 languages and more than 40 countries.

AI Mode works multimodally and accepts questions via text, voice or camera. Follow-up questions deepen the topic, which sets it apart from the pure summary of AI Overviews.

ChatGPT Search, Perplexity and Bing Copilot

ChatGPT Search extends OpenAI's chatbot with a live web search and is suited to dialogues with follow-up questions. Perplexity is designed for research with many visible sources.

Bing Copilot by Microsoft connects the AI answer with the Bing index and shows results with footnotes. All three name the websites used as sources.

Traffic effect

Who loses traffic through AI search?

Information-heavy guide, how-to and definition pages lose the most. The AI answers the question directly, and the click on the source is dropped.

How much do clicks fall through AI answers?

Clicks fall significantly. According to a SISTRIX analysis of more than 100 million keywords, around 20 percent of searches in Germany show an AI Overview, and the click-through rate on position 1 falls from 27 to 11 percent.

According to the Pew Research Center, only 1 percent of users open the citation link in the AI answer. According to Chartbeat, Google search referrals to publisher sites fell by 34 percent from December 2024 to 2025.

Which pages does the decline hit first?

It hits know pages fastest, pages that answer a question completely without a click being necessary. AI search now delivers exactly this answer itself, so publishers, guide portals and affiliate sites lose visitors.

For publishers, traffic in Google Discover and in search is thus at stake. Of the remaining clicks, Wikipedia, YouTube and Reddit benefit most, which AI systems use particularly often as sources.

Who loses less?

Less affected are brand, navigational and transactional searches, where people head for a specific page or want to buy something. Here the AI shows a complete answer less often.

Strong brands with their own demand also remain visible, because users search for them directly. An unambiguous entity therefore protects against the loss of traffic.

Visibility

How do you optimise for AI search?

Visibility in AI search is managed with Generative Engine Optimization (GEO). Content is built so that AI systems name the brand as a source.

To be distinguished from this is AI SEO. The term means the use of AI as a tool for SEO work itself, from research to drafting.

What is Generative Engine Optimization (GEO)?

Generative Engine Optimization (GEO) describes the optimisation of content for generative AI answers. The term was coined by the study by Aggarwal et al. (Princeton University, Georgia Tech and IIT Delhi) from 2023.

The researchers showed that a substantiated, source-based style with statistics increases visibility in AI answers. The basics are explained in the guide What is GEO.

Which content does AI search name as a source?

AI systems adopt passages that answer a question immediately and with evidence. Answer-first, an unambiguously named entity and a specific piece of evidence such as a figure, date or study increase citation worthiness.

Visible authorship and experience (E-E-A-T), current content and structured data count as well. How this basis relates to classic SEO is shown in the comparison of GEO and SEO.

Ranking and citation increasingly diverge. According to Ahrefs data, Google also adopts sources into AI Overviews that organically rank outside the top 10, so a top position no longer secures a mention.

Being mentioned instead of being cited

The actual goal is the brand mention in the answer, not the click on the source. According to Chartbeat, AI chatbots deliver under 1 percent of publishers' page views anyway, so the mention weighs more than the referral.

How present a brand is across the AI systems is shown by the LLM share of voice, the share of your own mentions compared to the competition. This ongoing build-up of mentions is the responsibility of GEO for publishers.

How publishers build subscription, newsletter and registration pages so that ChatGPT, Perplexity and AI Overviews name them is explained by Moritz Serif in the podcast episode "GEO: how publishers benefit from AI search" (in German).

Limits

What limits and risks does AI search have?

AI search can generate convincing-sounding but wrong answers and prefers large, well-known sources. Both demand a critical check of the results.

How reliable are the answers of AI search?

Reliability depends strongly on the system and the topic. Leading systems have become more accurate through grounding, but the Stanford AI Index 2026 continues to show large differences between the models.

The answers are still not reliable. On niche topics and current events there continue to be hallucinations, meaning invented facts or sources, which is why in health, finance and law a check against the original source remains necessary.

Which sources does AI search prefer?

AI systems cite large, well-known sources disproportionately often, while overlooking smaller and often more relevant pages. This bias disadvantages new providers.

The only remedy against this bias is to build unambiguous entities and substantiated content. Reach arises where the system recognises trust.

Transparency

Sources used

• Own experience from eight years of publisher SEO at publishers and newsrooms.
• Google: "AI Mode launches in Germany, Austria and Switzerland (German version)".
• Google Search Central: "AI features and your website (German version)".
• Pew Research Center: "Do people click on links in Google AI summaries?".
• Ahrefs: "Update: AI Overviews Reduce Clicks by 58%".
• SISTRIX: "AI Overviews in Germany: how much click-through rates really fall (in German)".
• Axios / Chartbeat: "Small publishers hit hardest by search traffic declines".
• Wikipedia (German): "Generative Engine Optimization".
• Stanford HAI: "The 2026 AI Index Report".