Written by: Mariana Fonseca, Editorial Team, AI Growth Agent
Key Takeaways
- Passage extractability, entity authority, and original data determine whether a page earns AI Overview citation.
- Only 37.9% of AI Overview citations come from the top 10, and 31% come from pages ranked beyond the top 100.
- AI systems read only a median of 377 words per page, so the first one to three sentences of every section must deliver the direct answer.
- Schema markup supports comprehension, and types like Organization, Article, and FAQPage provide the most value.
- AI Growth Agent maps your full universe of fan-out queries and builds content that earns selection. Book a demo to get started.
Candidate Pool Versus Selection In AI Overviews
Ranking earns a page a place in the candidate pool. Passage quality and entity authority decide which pages get selected into the final answer. These are two distinct stages. Treating them as the same stage creates most of the confusion in current AI Overview advice.
Ahrefs’ March 2026 study of 863,000 SERPs and 4 million AI Overview URLs found that only 37.9% of cited pages ranked in the top 10. Another 31.2% ranked at positions 11 to 100, and 31.0% ranked beyond the top 100 entirely. Ahrefs’ July 2025 predecessor study found 76.1% top-10 overlap, but that work sampled only the three most visible citations per overview. Ahrefs notes that citation parsing improved between the two studies, so the drop signals a directional shift rather than a single confirmed algorithm change.
BrightEdge’s February 2026 analysis found about 17% overlap between top-10 organic results and AI Overview citations. A separate BrightEdge 16-month analysis across nine industries from May 2024 to September 2025 found that only 16.7% of AI Overview citations came from the top 10. The SIGIR 2026 peer-reviewed study ran 11,500 queries and found an average source Jaccard similarity of 0.18 between AI Overviews and the SERP. That converts to roughly 30% overlap and places the peer-reviewed measurement within eight points of Ahrefs’ vendor estimate.
The mechanism that explains this pattern is Google’s query fan-out system, confirmed by Google Search Central. It splits a query into related sub-queries and cites pages appearing most often across those sub-query SERPs, instead of relying only on the original search result page. The three major studies converge on the same overlap range, as the table below shows.
| Study | Sample | Top-10 Citation Share | Date |
|---|---|---|---|
| Ahrefs | 863,000 SERPs, 4 million AI Overview URLs | 37.9% | March 2026 |
| BrightEdge | 9 industries, May 2024 to September 2025 | 16.7% | February 2026 |
| SIGIR 2026 (NJIT, NTU Singapore, Indiana University) | 11,500 queries across SERP, AI Overviews, and Gemini | ~30% (Jaccard 0.18) | December 2025 |
AI Growth Agent maps your full universe of fan-out queries and builds content that earns selection, not just consideration. Schedule a consultation to see how.
If ranking only earns consideration, selection depends on what AI systems actually read on each page.
How Much Of A Page AI Systems Actually Read
AI systems read pages in short passages instead of full documents. They split pages into chunks, retrieve a small set into a fixed context window, and focus most attention near the beginning. The practical unit of reading is the surviving passage.
Dan Petrovic’s Dejan.ai analysis of 7,060 queries and 2,275 tokenized pages found Google operates on a roughly 2,000-word grounding budget per query, selecting a median of about 377 words from any individual page. Grounding coverage drops as pages grow longer: pages under 1,000 words receive roughly 61% coverage, pages of 1,000 to 2,000 words receive roughly 35%, pages of 2,000 to 3,000 words receive roughly 22%, and pages over 3,000 words receive roughly 13%.
RESONEO’s July 2026 study of 1,249 ChatGPT responses, published in Search Engine Land on August 17, 2026, found ChatGPT opened and read only 759 pages, roughly one out of every 80 pages it retrieved. Pages it opened were cited 74% of the time versus 7% for pages merely fetched. Most sites are evaluated based only on their title, URL, and an excerpt of about 200 characters.
Andre Alpar’s late-July 2026 test of 14 AI assistants on a 705,216-word page confirmed that reading depth follows an absolute word count. Claude stopped at word 16,000 on both a 40,140-word page and a 705,216-word page. Gemini stopped at word 2,000. Alpar describes these numbers as directional rather than lab grade.
The practical implication is a ski-ramp structure. Place the direct answer in the first one to three sentences of every section, then elaborate. Surfer SEO’s late-2025 study of 57,253 URLs across 1,591 keywords found AI-cited pages averaged 31% Fact Coverage versus 24% for non-cited pages.
Ranking Requirements For AI Overview Citations
Ranking in the top 10 is not mandatory for citation, yet rank still influences the odds of selection. Digital Applied’s Q2 2026 study found that ranking in the top 5 for a query makes a URL likely to be cited in AI Overviews for related prompts. Pages outside the top 100 still earn citations when their passages answer fan-out sub-queries directly.
Surfer SEO’s November 30, 2025 study of 10,000 keywords found 67.82% of all citations were not in the top 10 for the query or any fan-out query, and 45.86% of top-3 visible citations were not in the top 10. Pages ranking for fan-out queries are 161% more likely to be cited than pages ranking only for the main query, according to a study reported by Search Engine Land on December 18, 2025.
Cyrus Shepard’s May 2026 meta-analysis of 54 studies scored URL accessibility at 9.5, search rank at 9.4, and fan-out rank at 9.3 as the top factors for AI citation. seoClarity’s October 12, 2025 study of 362,000 US desktop keywords and 5.1 million citations found 90% of AI Overviews cite at least one top-10 page, yet only 56% of citations come from the top 20. Brands therefore cannot rely on rank reports as a proxy for AI visibility.
Schema Markup Types That Matter For AI Overviews
Schema markup helps AI systems understand content structure, but it does not guarantee citation. Google Search Central states that structured data is not required for a page to appear in AI Overviews or AI Mode.
Ahrefs’ schema study tracked 1,885 pages that added JSON-LD schema against 4,000 control pages and found Google AI Overviews citations fell 4.6% relative to matched controls. Google AI Mode rose 2.4%, and ChatGPT rose 2.2%, with the latter two changes statistically indistinguishable from zero. A searchVIU experiment cited in Ahrefs’ report found none of five major AI systems used schema markup during real-time page retrieval and instead extracted only visible HTML content.
The schema types that matter for AI Overview citation factors are specific. Menra’s July 2026 guide ranks them as follows:
- Organization with sameAs for entity disambiguation
- Article with author as Person and dateModified for freshness and E-E-A-T tiebreaks
- Product with typed offers for shopping-flavored answers
- FAQPage for parse-proof Q&A pairs matching fan-out sub-queries
Schema cannot rescue a page that fails to answer the query, and it cannot guarantee citation on pages already inside the AI consideration set. As Menra’s July 2026 guide frames it, schema functions as “anti-hallucination infrastructure more than a ranking lever” because a typed offers.price value is harder to garble than a number buried in styled markup. CiteFlow’s 2026 guide adds that the freshness layer in both Google’s AI Overviews and Perplexity reads the dateModified field literally, which makes that field critical for AI citation.
Proving E-E-A-T For AI Citation
Entity authority travels across domains and surfaces. Off-domain mentions on YouTube, Reddit, and industry publications feed AI Overview citation directly.
Ahrefs’ study of 75,000 brands across ChatGPT, AI Mode, and AI Overviews found YouTube mentions show the strongest correlation with AI visibility at approximately 0.737. Branded web mentions followed at 0.66 to 0.71, branded anchors at 0.51 to 0.63, and link volume at near zero. Ahrefs’ March 2026 data shows YouTube accounts for 18.2% of non-ranking AI Overview citations and 5.6% of all AI Overview citations, and grew 34% in six months.
An AirOps analysis of 21,311 AI brand mentions across GPT-5, Claude Sonnet 4.5, and Perplexity found 85% of brand mentions came from external domains, and brands were 6.5 times more likely to be surfaced through third parties than through their own site. DeepSmith’s analysis found roughly 75 to 90 percent of citation slots in a Google AI Overview go to sources other than the brand being discussed, and earned media accounts for 82 to 95 percent of AI citations while paid or advertorial content accounts for about 0.3 percent.
AI answer systems cross-check four families of off-site sources: community (Reddit, niche forums, LinkedIn threads), reference (Wikipedia, Wikidata, structured knowledge bases), peer-review (G2, Capterra, Trustpilot), and editorial (best-of listicles, comparison articles, analyst roundups). A brand present across all four reads as consensus. A brand present in only one looks like an outlier.
Refresh Cadence For AI Overview Content
Content should refresh when signals indicate decay, not on a rigid calendar. Trigger a refresh when Search Console impressions decline, when dateModified contradicts visible dates, or when a claim has gone stale.
Ahrefs found that AI platforms cite content that is 25.7% fresher than the content ranking organically, which shows that recency carries more weight in retrieval than in ranking. SweetReed’s June 2026 analysis found 83% of AI citations for commercial and evaluation-stage queries come from pages updated within the past 12 months, and pages not refreshed quarterly are three times more likely to lose existing AI citations.
Freshness acts as a signal rather than a fixed schedule. Those same signals of declining impressions, changed competitor claims, pricing or feature updates, and a stale dateModified field should drive your refresh plan. CiteFlow’s 2026 guide notes that the freshness layer in both Google’s AI Overviews and Perplexity reads the dateModified field literally, which makes accurate timestamps essential.
Industry-Specific AI Citation Patterns
Citation winners vary sharply by industry and query type. BrightEdge’s vertical analysis measured a 52-point spread: 75.3% top-10 overlap in healthcare versus 22.9% in e-commerce, with education at 72.6% and insurance at 68.6%. A single citation strategy therefore cannot cover every vertical.
Digital Applied’s Q2 2026 study of 5,000+ intent-weighted queries across five AI surfaces found citation winners by vertical:
- SaaS: G2, Reddit, vendor docs, and Stack Overflow
- Health: Mayo Clinic, NIH/PubMed, and CDC.gov, the most concentrated vertical with little room for brand content
- Finance: Bloomberg, Reuters, SEC.gov, and Investopedia
- B2B services: HBR, McKinsey, Gartner, and firm-published original research
DeepSmith’s analysis found that on YMYL queries where a real authority has jurisdiction, .gov and .edu sources can take 60 to 90 percent of AI Overview citation slots. For commercial queries, review and comparison content dominates. Winning listicles share clear ranking criteria, a comparison table, an explicit methodology, and a takeaway near the top.
The practical implication by segment is straightforward. SaaS brands should prioritize original research and G2 presence. E-commerce brands should prioritize comparison tables and Product schema. Local brands should prioritize Google Business Profile signals. Regulated finance and health brands should prioritize primary source data and conservative language.
AI Overview Fix List: Week One, Month One, Quarter One
Teams should fix passage extractability first, entity authority second, and original data third. Other tactics support these three pillars.
Week One: Rewrite the first one to three sentences of every priority section as a self-contained answer block. Add Article schema with author as Person and dateModified. Confirm crawl access and indexability. These moves deliver fast gains because they unlock citation selection without requiring new content.
Month One: Build entity authority off-domain. Earn placement in independent third-party roundups, grow verified G2 and Capterra reviews, and land comparison-article mentions. AirOps found that brands present across all four off-site source families read as consensus to AI systems.
Quarter One: Publish original research with a clearly stated methodology and sample size. Digital Applied’s Q2 2026 finding is that original research with a clearly stated methodology and sample size gets cited across multiple AI surfaces for months after publication and compounds faster than summary content.
Schema cannot rescue a page with no extractable passages. Backlinks move little without entity authority behind them. Content volume only compounds when it carries original data. These dynamics explain why many popular recommendations sit behind the three core priorities.
Schedule a demo to see if you are a good fit and to get a week-one action plan tailored to your industry and citation gaps.
How AI Growth Agent Delivers AI Overview Growth
AI Growth Agent executes passage extractability, entity authority, structured data, and living content on a site the brand owns. It also provides incremental visibility reporting that proves what the engine actually generated.
The engine maps a brand’s full universe of seed terms and long-tail queries from real-time Google and ChatGPT data. It produces authoritative content that validates every claim and source, and it stands up a fully optimized site the client owns within the first week. It provisions the full technical and agentic SEO stack automatically: schema, Blog MCP, agent discovery via /.well-known/, llms.txt and llms-full.txt, sitemaps, and instant indexing. The client needs no technical skill.
Across the first twelve weeks, clients average more than 12,000 additional AI citations and mentions, over 100,000 additional bot visits, and a 20%+ lift in impressions. Content indexes in as little as ten days, and the first article goes live within a week. The content behaves as a living asset that updates and self-heals over time instead of going stale, so the brand’s presence keeps pace with a changing market.
AI search monitoring tools follow a different architecture. Those tools focus on monitoring first, meter prompts, and rely on action layers that still hand the work back to a human. They rarely close the loop of mapping, publishing, and self-healing on a site the client controls. AI Growth Agent starts from the opposite direction. Content creation sits at the core, and the engine maps, writes, publishes, and self-heals on a site the client owns.
Frequently Asked Questions
Do You Have To Rank In The Top 10 To Be Cited In AI Overviews?
Rank alone does not determine AI Overview visibility. As the Ahrefs and BrightEdge data above show, most citations come from outside the top 10, and fan-out rankings influence selection more than head-term rankings. Use rank as one signal among many, not as a stand-in for AI citation share.
How Much Of A Page Does AI Actually Read?
AI systems read only a small portion of each page. The Dejan.ai and RESONEO studies above show that models work with a limited grounding budget, sample a median of a few hundred words per page, and often rely on a short excerpt near the top. Treat the first one to three sentences of every section as the primary input to AI answer systems.
What Schema Markup Actually Influences AI Overview Citations?
Schema markup helps AI systems parse entities, dates, and offers, yet it does not act as a direct citation switch. The most useful types are Organization with sameAs, Article with author as Person and the dateModified field CiteFlow flagged as read literally, Product with typed offers, and FAQPage for clean Q&A pairs. Focus on these types after you have strong passages in place.
How Often Should You Refresh Content For AI Overviews?
Refresh content when performance and accuracy signals change. The Ahrefs and SweetReed findings above show that AI systems favor fresher pages and that stale content loses citations faster. Watch for declining impressions, updated competitor claims, product changes, and mismatched dates, then update the page and its dateModified field together.
Conclusion
Most AI Overview advice recycles traditional SEO guidance and centers on top-10 rankings, even though data from Ahrefs, BrightEdge, and SIGIR 2026 show that most citations come from outside that band. The contradiction disappears once you treat retrieval and selection as separate stages. Ranking earns consideration. Passage extractability, entity authority, and original data earn selection.
The brands cited in AI search this year are training the next generation of models with their own story. The brands that wait allow the next generation to learn from whatever happens to sit on the open web.
Traditional search tools show you where your brand stands. AI Growth Agent turns your brand into the answer. Book a kickoff and see your first article live within a week.
