In July 2025, 76% of pages cited in Google AI Overviews also ranked in the top 10 for the same query. By February 2026, that number had dropped to 38%. An Ahrefs study examining 863,000 keywords and 4 million AI Overview URLs documented the full trajectory — and the implication for teams that have been optimizing for rankings as a proxy for AI visibility is significant: the proxy broke.
This post unpacks what drove the citation-ranking decoupling, where AI Overviews are now pulling sources from instead, and what a correct GEO strategy looks like given the current data.
The 76% to 38% Drop: What Changed
The Ahrefs study tracked top-10 overlap with AI Overview citations from late 2024 through February 2026. The trajectory:
- Late 2024: ~75% of AI Overview citations from top-10 pages
- July 2025: 76% top-10 overlap
- October 2025: 54% overlap (BrightEdge)
- February 2026: 38% overlap (Ahrefs, 863K keywords) / 17% (BrightEdge, different methodology)
The divergence between methodologies — 38% vs. 17% — reflects differences in query sample and how partial overlaps are counted. The directional finding is consistent across both: citations from top-10 pages have declined sharply and substantially. For context, a BrightEdge analysis published February 12, 2026 puts the overlap even lower than Ahrefs.
Two factors drove the shift. First, the Gemini 3 upgrade on January 27, 2026 replaced approximately 42% of previously cited domains and increased the number of source URLs per response by 32% compared to its predecessor. A model upgrade that simultaneously expands source diversity and replaces nearly half of existing citations is effectively a large-scale citation reset — regardless of whether the underlying pages changed. Second, AI Overview’s retrieval system has been progressively weighted toward E-E-A-T signals, entity authority, and content freshness rather than organic ranking position — a reweighting that has been documented since mid-2025 and accelerated after the December 2025 Core Update extended E-E-A-T requirements beyond YMYL topics to all content categories.
Where Citations Are Actually Coming From Now
The Ahrefs distribution data from February 2026 shows how citations are spread across the organic ranking spectrum:
- Rank 1–10: 37.1% of cited URLs
- Rank 11–100: 26.2% of cited URLs
- Outside top 100: 36.7% of cited URLs
The outside-top-100 category is now roughly equal in citation volume to the top-10 category. This is structurally different from traditional SEO, where traffic drops off sharply below position 10. In AI Overview citation mechanics, pages that rank nowhere near the top 10 are competing in the same citation pool as pages that dominate organic SERPs — if they meet the retrieval criteria AI systems apply.
What drives citation eligibility for pages outside the top 10? The research points consistently to three factors: strong E-E-A-T signals (96% of AI Overview citations come from sources with measurable E-E-A-T), high entity density (pages with 15 or more recognized entities show 4.8x higher citation selection probability), and content freshness (85% of AI Overview citations came from content published within the last two years, with 44% from 2025 alone). A page that ranks #47 but has a named expert author, cites primary sources with data, was updated in 2025, and contains 15+ named entities outperforms many top-10 pages for AI Overview citation.
YouTube’s Citation Share in AI Overviews
The most unexpected finding in the Ahrefs citation distribution data is YouTube’s position: it is now the most-cited domain in AI Overviews overall, with 34% growth in citation share over six months and 5.6% of all citations coming from YouTube. YouTube accounts for 18.2% of all citations from outside the traditional top 100 — meaning YouTube is a major source of the “non-ranking” citation pool that now accounts for over a third of all AI Overview citations.
For brands that have no video presence, this is a concrete citation gap. A brand with comprehensive written content but no YouTube videos is structurally absent from a category that contributes 5.6% of all AI Overview citations. The practical investment — one substantive video per week on topics where citation is the goal, with transcripts published to the brand’s site — addresses both the YouTube signal and the content freshness signal simultaneously, since recent video publications are counted as fresh content in the 2025-published citation pool.
The Gemini 3 Reset and Citation Volatility
The January 2026 Gemini 3 upgrade replaced 42% of previously cited domains in a single model update. This isn’t an edge case — it’s a documented characteristic of how AI citation pools work when the underlying model changes. And model updates are not rare or predictable events that teams can anticipate and prepare for.
Separate research on AI Overview citation stability found that AI Overviews change their content roughly 70% of the time when the same query is run on different days, and when the answer updates, approximately half of the citations are replaced with new sources. For a team tracking citation presence on a weekly basis, this means a significant fraction of the citation set they see on Monday is a different set from what their competitors see on Tuesday.
The implication is not that AI visibility tracking is futile — it’s that single-point measurements are unreliable. Citation presence tracking requires rolling windows (4 weeks minimum for general queries, longer for low-volume queries) to distinguish genuine citation presence from variance. The Gemini 3 event also illustrates why brand presence across multiple citation sources — not just organic ranking — provides resilience that ranking-dependent strategies don’t.
Brand Search Volume as a Predictor
One of the more surprising findings from 2026 AI visibility research is the role of brand search volume in AI citation likelihood. Brand search volume shows a 0.334 correlation with AI citation frequency — materially stronger than total backlink count (0.218) and in a similar range to branded web mentions (0.664). The interpretation is that when users search for a brand by name on Google at high rates, it establishes that brand’s relevance and authority signal in Google’s entity graph in a way that flows through to AI Overview retrieval.
This creates a feedback loop: brands that already have strong organic brand search performance have a citation advantage in AI Overviews that purely technical GEO investments don’t fully bridge. For brands with weak brand search volume, the investment in building genuine brand recognition — PR, community presence, word-of-mouth channels — becomes part of the AI citation strategy, not separate from it. A brand with high domain authority but low brand search volume may rank well in organic results and still be underrepresented in AI Overviews relative to lower-DA brands with stronger brand recognition.
Platform Variance: Why AI Overview Results Don’t Represent Your Full AI Visibility
A 900-prompt study published by Zen Media on September 25, 2026 tracked a brand’s AI visibility score across ChatGPT, Gemini, Claude, Perplexity, and Grok simultaneously. The range: 15.5% visibility in the worst-performing platform to 59.5% in the best-performing. The same brand, the same time period, the same category — 4x variance across platforms.
This matters for teams that test AI visibility on one platform and generalize the result. Google AI Overviews, ChatGPT, Perplexity, Gemini, and Grok use different retrieval architectures, different source weighting systems, and different citation selection logic. A brand that tests well on Perplexity (which weights Reddit, community sources, and direct retrieval heavily) and poorly on Gemini (which has a narrow citation window and weights established editorial sources and brand authority) needs different interventions for each platform. Tracking citation presence on a single platform produces a platform-specific answer, not a category-wide visibility picture.
What to Prioritize Given This Data
The Ahrefs citation distribution data and the platform variance findings converge on a specific set of investments for teams that want to build AI visibility that doesn’t evaporate with the next model update:
- Content freshness audit (Week 1): 85% of AI Overview citations come from content published in the last two years, with 44% from 2025 alone. Pages last updated in 2023 or earlier are structurally disadvantaged regardless of their ranking or E-E-A-T signals. Identify your 10-15 highest-value pages and schedule substantive content updates — not minor edits, but additions of current data, new examples, or fresh statistics. Updated publication dates matter, but they need to be accompanied by genuine content additions to register as meaningfully fresh.
- Entity density review (Week 1-2): Run your top pages through a named entity recognition tool and count recognized entities per 500 words. Pages averaging fewer than 8-10 entities per 500 words are below the threshold where the 4.8x citation selection multiplier activates. Add named studies with publication dates, specific organization references, data points with sources, and defined terminology. This is a structural property of the content, not a quality judgment — a page can be well-written and have low entity density if it discusses topics in general terms rather than named, resolvable references.
- YouTube publishing (Week 2-4): YouTube’s 5.6% share of total AI Overview citations and 18.2% share of outside-top-100 citations makes video presence non-optional for brands serious about AI visibility. One substantive video per week on core category topics, with full transcripts published on the brand’s site, covers both the YouTube citation signal and the content freshness signal. The transcript doubles as fresh text content on the brand’s site for the AI Overview retrieval system.
- Cross-platform citation testing (Week 2): Run your core category queries across Google AI Mode, ChatGPT, Perplexity, Gemini, and Grok. Map where you’re visible and where you’re absent. Given the 4x variance documented in the Zen Media study, assume the result on one platform is not representative of others. Platform gaps that are narrow (15% vs 25% citation rate) may be variance; gaps that are wide (15% vs 59%) require platform-specific investigation.
- Brand search volume building: The 0.334 correlation between brand search volume and AI citation frequency means that building genuine brand recognition — the kind that leads people to search for you by name — is an AI citation strategy. This doesn’t reduce to a specific tactic, but the investment in PR, community building, podcast appearances, and earned media that drives brand name awareness has a measurable AI visibility return alongside its other benefits.
The 76% to 38% drop in AI Overview citations from top-10 pages is the clearest evidence yet that organic ranking and AI citation are operating on partially independent selection logic. Teams that have been treating ranking as a sufficient proxy for AI visibility have been operating on an assumption that expired sometime in late 2025. The citation data from February 2026 makes the new reality specific and measurable: more than 60% of AI Overview citations now come from pages that don’t rank in the top 10, and a correct AI visibility strategy needs to address that source pool directly.
To see exactly where your domain stands across Google AI Overviews, ChatGPT, Perplexity, and Claude — and which content and authority gaps are creating the biggest citation losses — run a free audit at ai-visibility.llmagnet.com.