How Google AI Overviews Selects Sources to Cite

Google AI Overviews now appear for roughly 30% of U.S. desktop searches, and their frequency on mobile is growing 475% year-over-year. These AI-generated summaries sit above the traditional search results and cite specific websites, so they change what an organic ranking is worth.

Google's official documentation says "there are no additional requirements" to appear in AI Overviews. Third-party data adds detail. I compared the official sources with large-scale studies covering 500+ million keywords and 16 months of longitudinal data to see what correlates with being cited. This is part of our series on how AI platforms select sources, alongside our guides on how ChatGPT selects sources and how Claude selects sources .

The Query Fan-Out Technique

The most distinctive thing about Google AI Overviews is query fan-out. Instead of running a single search, the Gemini-powered system breaks your question into subtopics and runs multiple searches simultaneously.

From the official Google blog :

With the Gemini 3 upgrade , Google says its latest model "more intelligently understands user intent" and "can find new content that it may have previously missed."

A study analyzing 173,902 URLs and 33,000 fan-out queries quantified the impact:

• Pages ranking for both the main query and at least one fan-out query accounted for 51% of AI Overview citations
• Ranking for fan-out queries is 49% more likely to earn a citation than ranking for the head term alone
• Spearman correlation of 0.77 between fan-out breadth and citation likelihood

In practice, pages that cover related subtopics are cited more often. A page that answers the main question and covers related angles is more likely to be cited than a page focused on a single keyword. That is the reverse of the single-keyword content that traditional SEO favored.

The Organic Ranking Connection

Several large-scale studies find a clear relationship between traditional organic rankings and AI Overview citations. The strength of that relationship varies by study and by industry.

What the Studies Found

seoClarity analyzed 500+ million keywords and found that 97% of AI Overviews cite at least one source from the top 20 organic results. The #1 position appears more than half the time.

Originality.ai found that 52% of AI Overview citations come from top-10 Google results. The top-ranked document alone has a 58% chance of being cited. By the top 30, nearly 90% of all citations are covered.

BrightEdge's 16-month longitudinal study across 9 industries found that AI Overview citations from organically-ranking pages grew from 32.3% to 54.5% (a 69% relative increase). Only 16.7% of citations came from top-10 results, meaning positions 21-100 drive most of the overlap growth.

68% of cited pages didn't rank in the top 10 for either the main query or any fan-out query. A page outside the top 10 can therefore still earn a citation.

Industry Variation in Overlap

The BrightEdge study found large differences by industry:

• Healthcare, Education, B2B Tech, Insurance: 68-75% overlap between AI Overview citations and organic rankings (trust-sensitive YMYL content)
• E-commerce: Only 22.9% overlap with virtually no change over 16 months
• Restaurants and Travel: Under 24% overlap

What Google Says vs. What the Data Shows

Google's official documentation says the following about what it takes to appear in AI Overviews:

• "There are no additional requirements to appear in AI Overviews or AI Mode."
• "You don't need to create new machine readable files, AI text files, or markup."
• "There's also no special schema.org structured data that you need to add."
• To be eligible, a page must be indexed and eligible to be shown in Google Search with a snippet.

Third-party research finds that certain factors correlate with more citations. Google requires nothing special, yet the data points to quite a lot of factors that go with being cited:

Structured Data: +73% Selection Rate

Google says no special schema is needed. Yet Wellows' analysis found that schema markup correlates with a +73% selection rate for AI Overview citations. A Search Engine Land analysis found that "only the page with well-implemented schema appeared in an AI Overview and achieved the best organic ranking, suggesting that schema quality, not just its presence, may play a role."

Google says no special structured data is required. Even so, structured data is associated with stronger organic rankings (the main route to AI Overview citations), helps Google understand entities, and makes content easier for machines to parse. Those factors tend to accompany a higher chance of citation. A similar correlation shows up on another platform: ChatGPT research shows sites with FAQPage schema are cited more often (a ChatGPT-visibility correlation of roughly 6.2% vs 0.8%, about 8× more likely to be cited), though this is a correlation in ChatGPT visibility, not a Google causal result. Our guide to AI agent readiness covers why structured data matters for all AI platforms.

E-E-A-T: 96% of Citations Come from Strong Signals

Wellows' research found that 96% of AI Overview citations correlate with sources that show strong E-E-A-T signals (Experience, Expertise, Authoritativeness, Trustworthiness). E-E-A-T is a Google quality concept, so this is a correlation in the cited set, not proof that schema or any single signal causes citations.

• Pages with expert authorship correlate with about 3.2x more citations in Google AI Overviews than generic staff-written content ( Relixir , correlational, Google AI Overviews only)
• Author schema that connects content to real experts in the Knowledge Graph may strengthen these signals
• For YMYL topics (healthcare, finance), E-E-A-T verification matters more, and these industries show the highest overlap with organic rankings

Entity Richness: 4.8x Higher Selection

One analysis suggests pages with 15+ recognized entities correlate with substantially higher AI Overview selection (reported as roughly 4.8x, correlational and specific to Google AI Overviews, not a causal result). Google's Knowledge Graph stores entities and their relationships: people, products, organizations, concepts. Content with many well-connected entities is easier for Google's AI to verify, which may help explain the correlation.

Google's Livegraph system assigns confidence weights to every identified triple (subject-predicate-object). Pages without strong entity signals may be filtered out early, even when they are well written, because the system needs to connect content to verified entities in the Knowledge Graph.

Content Format: Answer Units of 134-167 Words

These format patterns match how AI Overviews are structured:

• Self-contained answer units of 134-167 words perform best
• Pages using lists, tables, or FAQs align with how AI summaries are structured
• 44.2% of citations correlate with the first 30% of text ( Kevin Indig / Growth Memo ). Front-load key information
• Multi-modal content (text + images + video) shows +156% selection rate
• Content scoring 8.5/10+ on semantic completeness is 4.2x more likely to be cited

The Google-Extended Confusion

Blocking Google-Extended in robots.txt does not affect AI Overviews.

• Google-Extended is a robots.txt user-agent token specifically for AI model training data collection
• AI Overviews use standard Googlebot for crawling, not Google-Extended
• Blocking Google-Extended has no impact on search rankings, indexation, or AI Overview visibility
• Only blocking Googlebot itself would remove you from search entirely

Playwire's analysis confirms that you can block AI training crawlers without hurting your AI search visibility. The opt-out rules may still change. As of early 2026, Google is exploring ways to let sites opt out of AI Overviews specifically, separate from traditional search.

Most Cited Domains in AI Overviews

The SurferSEO AI Citation Report (36 million AI Overviews, 46 million citations) shows that video content leads:

1. YouTube (~23.3%), the most cited domain across every vertical
2. Wikipedia (~18.4%)
3. Google.com (~16.4%)
4. Reddit, LinkedIn, Facebook round out the top tier

Domain-specific sites such as NIH, Shopify, and ScienceDirect appear often within their niches, and AI Overviews distribute citations more evenly among niche sites than ChatGPT does.

The Semrush study (150,000+ citations) found that the top 20 domains account for 66% of all citations. That is still concentrated, though it leaves room for specialized content.

The Semrush AI Overviews study (10+ million keywords tracked from January-November 2025) found that 84% of AI Overviews appear for informational queries, 12.5% for transactional keywords (rising trend), and just 0.01% for local keywords.

When Do AI Overviews Appear?

AI Overviews don't appear for every search. The trigger patterns show where optimization is most likely to matter:

• seoClarity : 30% of U.S. desktop keywords trigger AI Overviews (September 2025)
• Mobile AI Overview frequency grew 475% year-over-year
• AI Overviews peaked at ~25% in July 2025, then declined to 15.69% by November, which suggests Google is showing them more selectively
• Average AI Overview text length dropped 70% (from ~5,300 to ~1,600 characters)

Google also reports positive engagement. According to the official Google blog , "when people click from search results pages with AI Overviews, these clicks are higher quality (meaning, users are more likely to spend more time on the site)."

What You Can Do Today

• Front-load key information . 44% of citations come from the first third of content
• Use specific entities . Pages with 15+ recognized entities show 4.8x higher selection
• Add images and video. Multi-modal content shows +156% selection rate
• Don't block Googlebot. Google-Extended is separate and safe to block
• Server-side render your content . Put important information in the HTML source. See our article on how AI agents see your website

Wrapping Up

Google officially says "just do good SEO." The data suggests that is necessary but not sufficient. Structured data, E-E-A-T signals, entity richness, comprehensive content, and clear formatting all correlate with higher citation rates. This matters most outside YMYL topics, where organic ranking overlap is lowest.

The most useful finding is that 68% of cited pages don't rank in the top 10. A page stuck on page two of Google can still be cited in an AI Overview. For more on the shift from traditional SEO to AI optimization, read our SEO vs AEO comparison .

Sources

• Google: AI Features and Your Website (Official documentation on AI Overviews requirements)
• Google Blog: Expanding AI Overviews and AI Mode (Official description of the fan-out technique)
• Google Blog: Gemini 3 in Search (Latest model upgrades for AI Overviews)
• Google Blog: AI in Search Driving Higher Quality Clicks (Click quality from AI Overviews)
• Search Engine Land: Fan-Out Rankings Study (173,902 URLs analysis)
• seoClarity: AI Overviews Impact Study (500+ million keywords analysis)
• Originality.ai: Google Ranking and AI Citations (Organic ranking correlation)
• BrightEdge: 16 Months of AI Overviews (Longitudinal overlap study by industry)
• Wellows: Google AI Overviews Ranking Factors (Schema, E-E-A-T, and entity correlation)
• Relixir: How to Earn Citations in AI Overviews (Expert authorship impact)
• SurferSEO: AI Citation Report (36 million AI Overviews analyzed)
• Semrush: Most Cited Domains by AI (150,000+ citations analysis)
• Semrush: AI Overviews Study (10+ million keywords tracking)
• Search Engine Land: Schema and AI Overviews Visibility (Schema quality impact)
• DataDome: Google-Extended Explained (Training vs. search crawler distinction)