Jonathon's AI Wiki

Ahrefs 38% AI Overview Citations from Top-10 Update (2026-03)

9 min read

Source: ahrefs-38pct-aio-top10-update-2026-05-19.md — Louise Linehan & Xibeijia Guan, reviewed by Ryan Law. Published 2026-03-02 on the Ahrefs Blog.

Ahrefs ran a longitudinal update to their July 2025 study tracking AI Overview citation overlap with organic top-10 rankings. The headline finding: top-10 overlap fell from 76% to 38% in seven months, with the drop attributed to Google’s Gemini 3 rollout on January 27, 2026 and a corresponding expansion of query fan-out behavior. The refreshed analysis covers 863K keyword SERPs and 4M AI Overview URLs pulled from Ahrefs Brand Radar — more than double the original July 2025 sample. The engineering implication is concrete: ranking for the user’s exact query is no longer the primary path into an AI Overview citation, because the model is increasingly pulling sources from the SERPs of fan-out sub-queries rather than the original query SERP.

Key Takeaways

  • The number that moved. 37.9% of URLs cited in AI Overviews also appear in the first 10 blocks of the same SERP (including ads, featured snippets, PAA, video packs, organic). Restricted to blue-link organic only, it’s 37.10%. Down from ~76% in July 2025. Seven months, half the overlap.
  • Methodology — full-block view + organic-only view. Ahrefs ran both a “all SERP blocks including features” pass and a “blue links only” pass. Numbers nearly identical (37.9% vs 37.10%), confirming the drop isn’t an artifact of feature-counting.
  • Where the citations now come from. 31.2% rank in positions 11-100. 31.0% don’t rank in the top 100 at all. The remaining 36.7% (organic-only view) sit outside the top 100 entirely.
  • Fan-out is the proposed mechanism. Google has confirmed that AI Overviews perform a “query fan-out” — splitting the user’s initial query into multiple sub-queries. Pages cited tend to be those ranking well across those sub-query SERPs, not necessarily the original query SERP. Gemini 3 appears to fan out more aggressively or more broadly than prior Gemini variants. ^[inferred]
  • YouTube dominates non-ranking citations. Among AI-Overview-cited pages that don’t rank in Google’s top 100 for the query, 18.2% are YouTube URLs. YouTube accounts for 5.6% of all AIO citations across the dataset and is the most-cited domain overall in Brand Radar — grew 34% over six months. Reinforces the Similarweb most-cited-domains data showing YouTube as a top citation surface for AI search broadly.
  • Gemini 3 launched January 27, 2026 — five weeks before this study. Ahrefs published March 2, 2026. The 38% number is post-Gemini-3; the 76% baseline was pre-Gemini-3 (July 2025). The model swap is the proposed cause.
  • Engineering implication. “Win the top 10 first, AEO second” is now insufficient on its own. The fan-out surface has grown to roughly two-thirds of citation sourcing. Practitioners need parent-topic / cluster coverage that wins fan-out sub-query SERPs, not just the head query.
  • Cluster-level reconciliation. This update sits cleanly inside the broader Zyppy meta-analysis ranking, which puts Search Rank as the #1 factor (8.5/10) but already weights cluster coverage and topic depth heavily. The 38% number is the empirical recalibration of how heavily.
  • Practitioner tools called out. Qforia (Mike King, iPullRank), Gemini API + Screaming Frog (Dan Hinkley, GoFishDigital), AI Visibility Fan Out (WordLift), plus Ahrefs Parent Topics, AI Content Helper, and Brand Radar for citation analysis. These are the surfaces that approximate Google’s internal fan-out behavior.

Study Design Details

  • Sample size. 863K keyword SERPs analyzed. 4M AI Overview URLs in the citation set. Roughly 2× the July 2025 sample.
  • Data source. Ahrefs Brand Radar (their AI visibility product) — same data infrastructure as the May 2026 schema causal study but a different analytical cut.
  • Two-pass methodology.
    • Pass 1 (all SERP blocks): tracked ads, featured snippets, PAA boxes, video packs, and organic listings as separate blocks. 37.9% of AIO-cited URLs appear in the first 10 blocks.
    • Pass 2 (blue links only): ignored ads and SERP features. 37.10% of AIO-cited URLs in the top 10 organic positions.
  • Distribution (blue-link-only view).
    • Top 10: 37.10%
    • Positions 11-100: 26.20%
    • Outside top 100: 36.70%
  • Parsing methodology improvement. Ahrefs notes they improved their citation parser since the July 2025 study, so they catch more of the citations rendered inside AIOs. This is a non-trivial confound — some of the apparent shift could be measurement-side rather than model-side. ^[inferred]
  • What the study cannot isolate. Whether the drop is purely Gemini 3 vs. accumulated changes to fan-out behavior across the seven months. Ahrefs proposes Gemini 3 as the driver but cannot run a clean A/B against the prior Gemini 2.5 baseline.
  • What the study explicitly does NOT measure. Effect of schema, content format, or other on-page factors on citation rates. That’s the scope of the May 2026 causal study. This study is purely about top-10-overlap geometry.

What This Changes in the Cluster Playbook

The AI SEO hub currently leads with two recommendations distilled across the 8-article cluster. Both need adjustment in light of this update:

Recommendation #1: “Win classical SEO first.” The cluster’s prior framing was “rank in the top 10 because that’s where ~76% of AI citations come from” — the July 2025 number was the load-bearing stat behind this advice. With top-10 overlap dropped to 38%, the framing weakens but doesn’t break. Top-10 is still the single highest-yield citation source per individual position (37.10% of citations from 10 URLs vs. 26.20% spread across 90 URLs in positions 11-100), so rank-first economics still hold. But the implicit promise that top-10 ranking gives you majority citation coverage is no longer true. The recalibrated framing: win classical SEO because it gives you the densest citation yield per page, not because it captures the majority of citations.

Recommendation #2: “Optimize for query fan-out.” Previously listed as the second-priority play. This update promotes it from “second-priority” to “co-equal with classical SEO.” Roughly two-thirds of AIO citations now originate from URLs that don’t rank in the top 10 of the user’s actual query — those URLs are coming from fan-out sub-queries. The cluster’s existing fan-out coverage in AirOps (88.6% of queries trigger exactly 2 fan-out sub-queries; retrieval rank dominates 4.1× over on-page signals) and the GEO-16 framework (12+ pillar hits → 78% cross-engine citation) gives the methodology. The recalibrated framing: fan-out coverage is no longer optional. Parent-topic cluster depth is now load-bearing because Gemini 3 is pulling from sub-query SERPs at scale.

New cluster sub-recommendation. YouTube presence is now its own line item. 18.2% of non-ranking citations / 5.6% of total citations / 34% YoY growth as the most-cited domain. The cluster previously treated YouTube as a “see Similarweb data” footnote — this update promotes it to a first-class AI-visibility surface. For dental marketing (the seo-content topic’s applied lens), this maps cleanly onto the existing OmniPresence video pipeline; for general practitioners, it means transcript-rich video content is no longer optional infrastructure.

Open Questions

  • Is the drop driven by Gemini 3 alone, or by gradual fan-out expansion across the seven-month window? Ahrefs attributes it to Gemini 3 but cannot run a clean before/after against the Jan 27, 2026 launch date. A study tracking the same SERPs week-over-week through January would settle this. ^[inferred]
  • How much of the shift is measurement-side rather than model-side? Ahrefs explicitly notes parser improvements between July 2025 and March 2026. If the new parser catches more citations from positions 11-100 and 100+ that the old parser missed, some of the apparent top-10 erosion is artifact. The magnitude of this confound isn’t reported. ^[inferred]
  • Does the 38% number hold across non-English markets and across query-intent classes? The Ahrefs sample is presumably US-English-heavy ^[ambiguous] and doesn’t break down by informational vs. commercial vs. transactional intent. Fan-out behavior likely varies meaningfully across those axes.
  • Ahrefs Schema → AI Citations Causal Study — Companion Ahrefs study from the same authors, two months later, on whether adding schema causes citation lift (null result). Same Brand Radar data infrastructure, different analytical question. The two studies bound the cluster: top-10-overlap geometry (this article) + schema-as-causal-lever (companion article).
  • AirOps Fan-Out Effect ChatGPT Study — The single deepest fan-out-mechanics study in the cluster. 16,851 queries, 50,553 responses, 88.6% of queries trigger exactly 2 fan-out sub-queries, retrieval rank dominates 4.1× over on-page signals. Methodology for replicating fan-out analysis at scale.
  • Digital Applied 1,000 AIO Citation Pattern Study — 1,000-AIO correlational study covering domain authority concentration (top 1% capture 47% of citations) and schema-type lift breakdowns. Cross-validation surface for the YouTube and domain-concentration findings here.
  • Zyppy AI Citation Ranking Factors — Cyrus Shepard’s 54-study meta-analysis ranking 23 citation factors. Search Rank is #1 at 8.5/10; this Ahrefs update is the recalibration of how heavily Search Rank dominates (still #1, but the absolute share dropped from 76% to 38%).
  • GEO-16 Framework — Academic AEO/GEO study covering 1,702 citations across Brave / AIO / Perplexity. Pillar-hit coverage threshold (≥12 hits → 78% cross-engine citation) is the methodological complement to Ahrefs’s top-10-overlap geometry.
  • Google’s Generative AI Search Optimization Guide — Google’s official position: AIO and AI Mode rely on the same Search index, so AI search optimization is core SEO. Ahrefs’s longitudinal data tests how that core-SEO advice holds up under model upgrades — the answer is “weakens, doesn’t break.”

Try It

  1. Run a fan-out audit for your top 10 head terms. Use Qforia, the Hinkley Gemini-API workflow, or WordLift’s AI Visibility Fan Out to simulate Google’s fan-out expansions. For each head term, identify the 3-5 most common sub-queries and check whether your existing content ranks for them. Gaps in sub-query coverage are now load-bearing citation gaps.
  2. Audit your YouTube footprint. If you’re not on YouTube, you’re missing 5.6% of AI Overview citations on average — and 18.2% of non-ranking citations. For service-business clients, this is the year to ship a transcript-rich, topic-authoritative channel. The existing ai-video-content topic covers the pipeline tools.
  3. Re-run your citation reporting. If you previously told stakeholders “we rank in the top 10, therefore we’ll be cited,” the framing is now wrong. Pull your AIO citation data (Brand Radar, BrightEdge, Similarweb) and report on cited-but-not-ranking and ranking-but-not-cited separately. The two failure modes need different fixes.
  4. Shift one quarter of your AEO budget from schema to cluster depth. Combined with the schema causal null result, the cluster now has explicit cost-of-effort guidance: same-page schema additions are infrastructure, parent-topic cluster expansion is the lever. Make the budget reflect it.
  5. Watch the next Ahrefs update. This study landed five weeks after Gemini 3. The number will move again with Gemini 4. Treat 38% as a 2026-Q1 snapshot, not a stable equilibrium.

Graph View

Backlinks