Google AI Overviews: How Google Generates AI Answers

Why AI Overviews are rebuilding the search results page

Within two years, Google AI Overviews have turned the search results page from a list of links into an answer engine. The feature is no longer a niche product. According to Sundar Pichai on Alphabet's quarterly earnings call, AI Overviews reach 2 billion monthly users in 200 countries and territories, up from 1.5 billion in May 2025. For B2B companies in the DACH region, this shifts the central question. Beyond ranking in the organic top 10, what counts now is whether Google's AI includes your brand as a source in the generated answer.

The economic impact shows in click behavior. According to a US study by Pew Research Center, users who saw an AI summary clicked a classic search result in only 8 percent of visits, compared with 15 percent without an AI summary. The same users clicked a link inside the summary itself in just 1 percent of visits. The answer is increasingly replacing the click. If you understand how the feature works, you can deliberately work toward becoming part of that answer.

How Google technically generates AI answers

AI Overviews differ fundamentally from the earlier Search Generative Experience (SGE), the beta predecessor. AI Overviews are the rolled-out product above the organic results. Two other products need to be kept apart from them: Google AI Mode, a standalone conversational search experience, and Gemini, Google's chatbot. AI Overviews and AI Mode share the underlying technology but draw their sources differently.

The central mechanism is called query fan-out. Instead of processing a search query as a single string, the language model breaks the question into several thematically related sub-queries and searches for them in parallel. Robby Stein, VP of Product for Search at Google, illustrates this with an example from AI Mode: asked about activities in Nashville for a group, the system thinks through a series of sub-questions such as good restaurants, good bars, and activities with kids, and starts googling in the background. The key point is that the system also researches subtopics the user never explicitly mentioned.

The scope of this fan-out scales with complexity. According to Google, in the Deep Search mode of AI Mode the mechanism can fire dozens or even hundreds of background queries and take several minutes to complete. Then comes synthesis: a Gemini model collects the retrieved passages from the sub-queries, reconciles them, and formulates a coherent answer. For commercial queries, the Shopping Graph, Google's structured product database, feeds in as well.

In practice, this means your content should answer as many of the likely sub-queries as possible, not just the one main question. If you break a topic into clean, standalone answer blocks, you increase the chance that at least one of these blocks matches one of the fanned-out searches.

Which search queries trigger AI Overviews

Not every search produces an AI summary. The trigger rate fluctuates heavily and is the subject of ongoing Google tests. A US analysis by Semrush covering more than 10 million keywords documents the trend for 2025: AI Overviews were triggered for 6.49 percent of queries in January, rose to a peak of 24.61 percent in July, and stood at 15.69 percent in November. This volatility is strategic information in itself: AI Overviews are a moving target, not a stable state.

Search intent is decisive. For October 2025, the same Semrush analysis shows a clear distribution of queries with an AI Overview: informational queries accounted for 57.1 percent, commercial queries for 18.57 percent, and transactional queries for 13.94 percent. AI Overviews appear most often for longer questions that need explanation. Google is much more cautious with clearly transactional or local queries and with sensitive YMYL topics (Your Money or Your Life, meaning health and finance). For B2B, this means that the typical research questions at the start of the buyer journey trigger an AI answer particularly often.

AI Overviews in Germany and Austria

In the DACH region, AI Overviews appear under the German label "Übersicht mit KI". The rollout came later than in the US market, but the feature is now widely available, since according to Google it runs in 200 countries and territories. For German-language queries, the quality of the synthesis has caught up. The US market remains the primary testing ground for new features, though, which is why changes can usually be observed there first.

On the regulatory side, the EU is a special case. The GDPR and the EU AI Act set conditions that Google has to take into account when serving these features. In practice, this means that individual features launch later in the EU or in an adapted form. For a strategic assessment in the B2B market, channel concentration matters most. According to StatCounter, Google held a market share of 80.88 percent in Austria in September 2026, with Bing at 10.93 percent. In Germany, Google stood at 87.84 percent in September 2026, Bing at 6.5 percent. If you want to be visible in German or Austrian search, there is no way around Google's AI answers.

Impact on click-through rate and traffic

The click effects are measurable and getting stronger. An Ahrefs analysis of 300,000 keywords from December 2025 found that the presence of an AI Overview correlates with a 58 percent lower average click-through rate for the top-ranked page, compared with a reduction of 34.5 percent in the April 2025 analysis. A second metric shows the trend over two years: the position-1 click-through rate for AI Overview keywords fell from 0.073 in December 2023 to 0.016 in December 2025.

The effect hits keyword types differently. Click loss is greatest for purely informational queries where the AI answer fully answers the question. For commercial and transactional queries, where users want to compare, configure, or buy, the click survives more often because the summary alone is not enough. This leads to a sober assessment of content: pure definition articles lose the most traffic, while content close to decisions and products holds up better.

How to optimize content for AI Overviews

Optimizing for AI Overviews follows the principles of generative engine optimization. The central question is whether a passage can be extracted as a precise, standalone building block of an answer.

• Front-loading and answer islands: Place the direct answer to a question immediately below the relevant heading. An answer island is a self-contained paragraph that fully answers a question without needing context from other sections. These are exactly the blocks that match the fanned-out sub-queries.
• Statistics and source citations: Back up statements with concrete numbers and named sources. AI systems prefer verifiable, attributed facts over vague claims because they lower the risk of a faulty synthesis.
• E-E-A-T signals: Experience, expertise, authoritativeness, and trustworthiness have to be machine-readable. Clear author attribution, traceable sources, and a recognizable professional background increase the likelihood of a citation.
• Concise structure: Short paragraphs, descriptive subheadings, lists, and tables make extraction easier. Whatever can be cleanly lifted out as a passage is more likely to become part of the answer.
• Question-and-answer format: Phrase subheadings as questions that real users ask, and answer them right away. This covers the most common forms of query fan-out.

Citation logic: why AIO sources rank outside the top 10

The most important strategic shift concerns source selection. The old rule was: whoever ranks at the top gets cited. That logic is breaking down. An Ahrefs analysis of 863,000 keyword SERPs and 4 million AI Overview URLs found that only 38 percent of the pages cited in AI Overviews also rank in the top 10, compared with around 76 percent in July 2025; the remaining roughly 62 percent are split between positions 11 to 100 (31.2 percent) and pages outside the top 100 (31.0 percent).

This does not mean that classic ranking becomes irrelevant. It means that an organic top-10 position is neither a necessary nor a sufficient condition for a citation. Google increasingly selects sources by how well the content fits the respective sub-query, not by ranking alone. For SEO and GEO strategy, this moves the lever. Link building on its own loses its unique standing, while topical depth, brand mentions in relevant contexts, and the extractability of individual passages gain weight. Brand mentions in trustworthy sources can thus become a citation signal, even if your own page is not in first place.

Common mistakes in AI Overview optimization

In practice, a few misjudgments keep recurring, and they cost visibility.

• Measuring success by ranking alone: If you only track organic position, you miss citations. Since only a fraction of the cited pages rank in the top 10, rank monitoring is blind to a large part of AIO visibility.
• Burying answers in running text: If the core statement only comes after several paragraphs, extraction becomes difficult. Query fan-out rewards passages that answer a sub-question immediately and completely.
• Misjudging AI crawler blocks: If you block GPTBot, ClaudeBot, or Google-Extended in robots.txt, you do not lose any visibility in AI Overviews as a result. These rules affect OpenAI, Anthropic, and Gemini apps and Vertex AI. A page drops out of AI Overviews if Googlebot is not allowed to crawl it or if noindex, nosnippet, or max-snippet:0 is set. With data-nosnippet and other max-snippet values, you only limit which text appears. Make both decisions deliberately.
• Confusing a traffic decline with a loss of relevance: Falling click numbers are partly unavoidable for informational keywords. If you cut back good content because of this, you lose exactly the substance that earns citations.
• Relying on a single AI measure: Neither llms.txt nor a single schema snippet replaces substantial, crawlable, and well-structured content. Technical signals only work in combination with real content value.

Structured data and technical requirements

For content to be eligible for AI Overviews at all, it has to be crawlable and machine-interpretable.

• Crawlability for Googlebot: To appear in AI Overviews, a page has to be indexed and eligible to be shown in Google Search with a snippet. So Googlebot needs access. You limit what appears in AI Overviews via nosnippet, data-nosnippet, max-snippet, or noindex. Google-Extended only controls training and grounding for Gemini apps and Vertex AI and, according to Google, has no effect on inclusion in Google Search.
• Schema.org structured data: Markup such as Article, FAQPage, Product, or Organization helps search systems classify content correctly. Structured data does not replace content, but it makes interpretation and extraction easier. Google does not require any special markup for AI Overviews.
• Putting llms.txt in context: The proposed llms.txt file is meant to give AI systems a map of your content. Its actual effect on citations has not been demonstrated so far. It is a small investment, but no substitute for crawlable, well-structured content.
• A clean technical foundation: Fast load times, clear HTML semantics, and a logical heading hierarchy remain the foundation. What a crawler cannot read cleanly, a model cannot reliably cite.

Measurement and tools

AI Overviews escape classic rank tracking because a citation is not a ranking. B2B teams therefore need additional metrics.

• AI referral traffic in GA4: Capture referrals from AI systems by segmenting the relevant source and medium combinations. Part of this traffic shows up as direct traffic because referrer information is missing, which is why clean segments and, where appropriate, UTM parameters matter.
• AIO visibility and share of voice: Specialized GEO tools check which queries show an AI Overview and whether your brand is cited in it. Share of voice in AI answers replaces classic ranking as the measure of visibility.
• Conversion value instead of click volume: Since AI Overviews push click volume down, the focus should be on the quality of the remaining visits. What matters is whether AI-referred users convert, not only how many of them arrive.
• Trigger monitoring: Keep track of which of your core keywords show an AI Overview. Given the documented fluctuation in the trigger rate, this is an ongoing process, not a one-time check.

Outlook: AI Mode, agentic commerce, and next steps

AI Overviews are the first stage. Google AI Mode takes conversational search further: users ask follow-up questions, and query fan-out plays an even larger role. In parallel, agentic commerce is creating a layer in which AI agents research and compare products on behalf of users. Both raise the importance of structured, extractable, and well-sourced content.

For B2B companies in the DACH region, this leads to three concrete steps. First, rebuild your information-rich content into standalone answer blocks that match the likely sub-queries. Second, shift measurement from pure rank tracking to AIO visibility and conversion value. Third, secure Googlebot's access, decide deliberately on blocks for other AI crawlers, and maintain structured data consistently. Given Google's market share of 80.88 percent in Austria (September 2026), optimizing for Google's AI answers is the central lever for search visibility.