AI Summary
Google AI Overviews appear for 9.46% of all keywords on desktop and 16% of all US keywords, according to Ahrefs’ analysis of 55.8 million AIOs across 590 million searches. They reduce position-1 click-through rates by 34.5% for those keywords. By search volume, AI Overviews cover at least 12.8% of all Google searches. The playbook for earning citations inside them is well-defined, and it starts with the same foundation as good SEO.
When AI Overviews Trigger and When They Do Not
Ahrefs’ 55.8 million AIO dataset shows that 97.70% of keywords triggering AI Overviews are informational in intent. Branded queries, local queries, and transactional queries are significantly underrepresented. AIOs are more likely on longer search queries and higher search volume keywords than on short-tail or low-volume terms. This means the bulk of AIO exposure comes from mid-to-long-tail informational research queries, the same queries where strong pillar content already wins organic rankings.
For site owners: the question “does an AIO appear for my target keyword” is measurable. Third-party tools track AIO presence by keyword. Google Search Console as of June 2026 has begun including generative AI performance data in the Performance report, though separating AIO-specific clicks from organic clicks remains limited.
How AI Overviews Work: The Retrieval Mechanism
AI Overviews use retrieval-augmented generation (RAG). Gemini decomposes the user’s query into sub-queries, retrieves relevant passages from Google’s index, and synthesizes a single answer with citations. The citations are grounded in specific passages, not just pages. This architecture has direct implications for how you structure content.
Citation slots favor passages that directly answer a sub-query. A page that broadly covers a topic but does not contain a passage that crisply answers one specific sub-question is a poor candidate for citation. This is why pillar and cluster architecture matters: each cluster page should own one narrow intent, and each H2 within it should answer a distinct sub-question.
Our May 2026 study of 153,425 citations found that 74.9% of cited sentences appear in the first half of the source document. Mean cited position sits at 37% through the document. Front-loading factual claims is the structural pattern that matches how the RAG system retrieves passages.
What Google’s Official Guidance Actually Says
Google’s official position, stated consistently across Search Central documentation and the May 2025 “succeeding in AI search” post, is that no separate AI optimization is needed beyond solid SEO. The pre-verified pool of guidance confirms: Google explicitly states that optimizing for AI features does not require a separate strategy from traditional SEO best practices. Google also ignores llms.txt files for AI Overview citation purposes.
What that means practically: your crawlability, indexability, Core Web Vitals, and overall content quality are the same inputs for AIO citations as for organic rankings. There is no secret AIO-only signal. The separation is in emphasis: the RAG system rewards passage-level precision, which standard SEO underweights compared to page-level signals.
AI Overviews vs AI Mode: Different Animals
AI Overviews are the inline summary boxes that appear above standard SERP results. AI Mode is Google’s full conversational search interface, launched in the US in May 2025. They share underlying technology but serve different use cases and have different citation behavior. The AI Mode optimization playbook covers the conversational side. This post focuses on AI Overviews specifically: the inline citations in standard search results.
The CTR Reality: Why Citation Presence Matters More Than Rank
Ahrefs analyzed 300,000 keywords and found that AI Overview presence correlates with a 34.5% lower average CTR for the top-ranking page versus similar informational keywords without an AIO. This matches the broader zero-click trend: Bain’s December 2024 consumer survey (n=1,117) found 80% of consumers rely on zero-click results in at least 40% of their searches, with organic traffic down an estimated 15-25%.
The implication is structural. Being ranked number one for a keyword with an AIO is worth less than it was two years ago. Being cited inside the AIO partially compensates, because citation links within AIOs do generate clicks. The citation slot becomes more valuable than the rank position below it. We analyze the full traffic picture in our AI Overviews CTR impact analysis.
The Optimization Playbook
| Tactic | What It Targets | Priority |
|---|---|---|
| Match each H2 to a specific sub-query Gemini would decompose from your target keyword | RAG sub-query resolution | High |
| Lead every section with a 1-2 sentence direct answer before expanding | Passage extraction for citations | High |
| Place key factual claims in the first 35% of the document | Citation position bias (mean 37%) | High |
| Add Article schema with datePublished, dateModified, and Author with sameAs links | E-E-A-T signals + freshness | High |
| Write factual claims in 6-15 word declarative sentences | Passage-level citation extraction | High |
| Build deep single-topic pages (3,000+ words) over many thin pages | Topical depth signal | Medium |
| Use structured lists and comparison tables for data-dense sections | Structured data extraction | Medium |
| Include named numerical sources in first 200 words | Verifiability signal | Medium |
| Ensure crawlability: no robots.txt blocks, no noindex on target pages | Index access | Required |
Passage-Level Optimization in Practice
The core shift from page-level to passage-level optimization is mechanical. Every H2 section should be able to stand alone as a complete, citable answer. If you remove every other section from your article, each remaining H2 block should still make sense and provide useful information independently.
This is the discipline behind the BLUF writing format for AI content: Bottom Line Up Front, where the most important claim appears in the first sentence of every section. RAG systems extract the most relevant passage per sub-query. If your most relevant sentence is the fourth sentence in a section preceded by context-setting prose, it is less likely to be extracted cleanly than if it is the first sentence.
Our May 2026 study found cited sentences average 9.27 words, median 10, none exceeding 18 words. Long subordinate clauses and hedging language reduce citation rate. Short, declarative facts increase it. Writing atomic sentences for AI citation is the same skill applied to AIO optimization.
Schema Markup and E-E-A-T for AIO Citation
Article schema with Author markup and sameAs links to professional profiles is a verified E-E-A-T signal. FAQPage schema for Q&A content and HowTo schema for step-by-step guides align content structure to query types that frequently trigger AIOs. Schema markup for AI search covers implementation specifics. The short version: structured data does not directly inject content into AIOs, but it helps Google’s systems classify and surface passages with higher confidence.
E-E-A-T signals matter because AIOs are more likely to cite sources Google has assessed as authoritative on the topic. The top 50 domains hold 28.90% of all AIO mentions across Ahrefs’ 55.8M AIO dataset. Wikipedia, Reddit, and Mayo Clinic dominate not because they optimize for AIOs, but because they have deep entity authority and consistent factual accuracy. For non-institutional brands, the path to AIO citation share runs through topical authority built over dozens of closely related pages on a narrow subject.
Comparing AI Overviews to Featured Snippets
Featured snippets and AI Overviews both extract passages from indexed pages, but they behave differently. Featured snippets are a single source. AI Overviews synthesize from three or more. Featured snippets require position one or near-top ranking in most cases. AI Overviews pull from a wider pool of sources. The overlap in optimization tactics is high: both reward passage clarity, direct answers, and structured content. The key difference is that AIOs are more likely to cite pages that are not rank-one, creating a citation opportunity for position 5-15 pages with superior passage quality.
We compare the two features in depth at Featured Snippets vs AI Overviews. The takeaway: if you optimized for featured snippets five years ago, you already understand 70% of the AIO optimization framework. The remaining 30% is passage density, sentence length, and multi-source synthesis signals.
Content Freshness and AIO Citation
AI Overviews for time-sensitive queries strongly prefer recently updated content. Surface datePublished and dateModified in both visible HTML and Article schema. Perplexity shows the most acute freshness bias among platforms, but Google’s AIO system also weights recency for queries with temporal intent. Content freshness and AI citation recency bias is a documented pattern: pages that have not been substantively updated in 12 or more months underperform in AIO citations for competitive informational queries, even when organic rank is maintained.
The GEO Research Foundation
The arXiv GEO paper (KDD 2024, arXiv 2311.09735) showed up to 40% visibility gains from combining cite-sources, quotation, and statistics methods. The leveling effect is sharpest at the extremes: in the paper’s Table 2, the Cite Sources method moved a rank-5 source’s visibility up 115.1% while reducing an already rank-1 source’s visibility by 30.3%. These methods map directly to AIO optimization: including verifiable statistics inline, citing named sources, and writing factual claims as independently verifiable statements.
In our client work, we track AIO citation share using our open-source GEO/AEO Tracker. We run a fixed query set weekly, extract cited URLs, and measure share over a 60-day window. Typical time to measurable change after passage-level restructuring is 4-6 weeks. Schema updates and freshness signals register faster, usually within 10-14 days.
What Does Not Work for AI Overviews
- llms.txt files. Google has explicitly stated it ignores llms.txt for AI Overview citation purposes.
- Shallow keyword-matched pages. AIOs favor topical depth. Thin pages that match a keyword but contain minimal unique information are not extracted.
- Generic AI-written filler. The RAG system extracts the most information-dense passage per sub-query. Low-density prose loses to dense factual passages regardless of keyword frequency.
- Hiding content behind login walls. Pages that Googlebot cannot access cannot be cited. Standard indexability requirements apply fully.
- Expecting rank-to-citation automatic transfer. Rank 1 does not guarantee AIO citation. Passage quality, topical authority, and E-E-A-T are independent signals.
Connecting to Your Zero-Click Strategy
AIO optimization exists within a broader zero-click strategy. Bain’s research (n=1,117, December 2024) found 60% of all searches now end without a click to any website. That means for the majority of informational queries, the only way to reach the user is through the AI-synthesized answer itself: via citation. A brand that earns consistent AIO citation share for its category’s core queries is visible at the moment of decision even when clicks never happen.
We cover the full framework for zero-click AI search strategy separately. AIO optimization is one component: it maximizes citation presence on the platform that still drives the most total query volume globally. Combined with structured data implementation and topical cluster depth, it creates a compounding visibility moat that pure organic rank no longer provides on its own.