Nearly 30% of Google AI Overview Sources Sit Outside Page One: New Research Changes the SEO Playbook

The headline finding was that nearly 30% of the sources cited by Google AI Overviews did not appear among the conventional first-page results shown for the same search. This means a website could be absent from the visible top results and still become one of the sources used to construct the answer appearing above them. The researchers concluded that AI Overview source selection appears to be meaningfully different from Google’s conventional ranking system.

This does not mean that organic rankings no longer matter. Google still needs to discover, crawl, interpret and evaluate a page before it can use that page confidently. However, it does mean that page-one ranking position is no longer a complete measure of search visibility. A business might rank below established competitors in conventional results while still providing the particular fact, explanation or supporting evidence Google uses within an AI-generated response.

Traditional SEO reporting has often concentrated on whether a page ranks in positions one to ten. That remains commercially important because prominent organic positions can generate visits, leads and sales. The new research suggests, however, that AI Overviews can create another form of exposure outside that familiar threshold. A page may influence the answer a user sees even when it is not one of the first conventional results presented.

This changes how businesses should interpret an apparent ranking failure. A page ranking outside the top ten may still have value if it is being cited, summarised or used to support an AI-generated answer. Conversely, a page holding a strong traditional position may receive less attention when Google places a complete response above it. Businesses therefore need to measure rankings, citations and AI appearances separately rather than assuming one metric represents the others.

The study found that AI Overviews appeared for 13.7% of all queries in the dataset, but the activation rate increased dramatically when the search was written as a question. Question-form queries triggered an AI Overview in 64.7% of cases, compared with 9.5% for searches that were not phrased as questions. Searches beginning with “how” activated AI Overviews in 84.3% of cases, while those beginning with “why” did so in 73.4% of cases.

Longer searches were also more likely to generate an AI response. Among non-question queries, the activation rate increased from 9.9% for single-word searches to 38.7% for searches containing six or more words. These are often the detailed, explanatory and comparison-based searches people make when they are researching a problem rather than looking for a specific website. For businesses, these questions can reveal where AI visibility is most likely to influence discovery.

The researchers found that the sources selected for AI Overviews were, on average, more credible than the conventional first-page sources displayed alongside them. This is encouraging because it suggests that Google’s AI layer is not automatically selecting lower-quality pages. The AI citation pool was also less concentrated around a small group of dominant websites, potentially leaving more room for specialist and industry-specific sources to appear.

However, source credibility and answer accuracy were not the same thing. After breaking the AI responses into more than 98,000 individual claims, the researchers found that approximately 11% were not supported by the cited pages. In many cases, the issue was omission rather than direct contradiction, meaning the generated answer contained information that could not be found in the references attached to it. The study cautions that some of this may result from technical limitations in analysing social or video sources, but the wider finding remains important.

Businesses should not interpret the findings as permission to abandon conventional SEO. Crawlability, useful pages, clear site structure and strong organic performance still help search systems find and understand information. The opportunity is to look beyond broad ranking targets and create content that provides distinctive evidence, direct explanations and specific answers that an AI system can use when constructing a response.

Generic articles that repeat information already available across hundreds of websites are unlikely to create much additional value. Businesses have a stronger opportunity when they publish original research, first-hand experience, specialist commentary, clear definitions, practical comparisons and well-supported answers to specific customer questions. Google’s own recent guidance has similarly emphasised unique, non-commodity content created for readers rather than material produced merely to target search queries.

An AI Overview does not always need an entire article. It may use one statistic, one explanation, one comparison or one clearly expressed piece of evidence. This makes information design increasingly important. A business may have valuable knowledge on its website, but that knowledge is less useful when it is hidden inside vague marketing language, unexplained claims or long passages that never answer the central question directly.

Strong pages should make important information explicit. That can include named services, locations, prices where appropriate, eligibility criteria, product differences, original data, expert quotations and answers to common customer concerns. The aim is not to write mechanically for an algorithm. It is to make the page so clear and well-supported that both a person and an AI system can understand what it contributes.

A traditional rank tracker cannot show whether a page has been used inside an AI Overview when the page does not appear in the top ten. It also cannot explain whether the business itself was named, whether a competitor was recommended or whether a third-party source shaped the final answer. Ranking data therefore remains useful, but it should sit alongside direct testing of the searches and questions customers are likely to use.

Businesses can begin by creating a set of priority questions covering early research, comparisons, recommendations and purchasing decisions. Those searches should then be tested repeatedly because AI answers and citations can change. The results can reveal where the company appears, which competitors are selected, which sources influence the response and whether existing content is being used without the brand receiving meaningful visibility.

The research offers a potentially positive message for smaller organisations competing against more established domains. Conventional search results can be difficult to break into when large publishers, directories and long-standing competitors dominate the first page. If Google’s AI layer is selecting a broader set of sources, a specialist business may still earn visibility by supplying particularly relevant or authoritative information.

That opportunity is not guaranteed, and appearing as a citation is not the same as being recommended as a business. Nevertheless, the findings show why organisations should not assess their entire search presence through ranking position alone. A page can contribute to an AI answer from beyond page one, while a highly ranked page can remain absent from the generated response.

Google Search now contains at least two related selection systems. One orders conventional results, while another identifies the sources and claims used to generate an AI response. The overlap between them is significant, but the new research shows that it is far from complete. Businesses that treat AI Overviews as nothing more than shortened search results may therefore miss how visibility is actually being distributed.

The practical response is not to replace SEO with a new set of tricks. It is to retain strong SEO foundations while separately testing how the business is represented within generated answers. As AI Overviews become more common for detailed customer questions, the organisations that understand both layers will have a clearer picture of how people discover, evaluate and choose them.