In March 2026, Ahrefs published results from a study of 863,000 keyword SERPs and 4 million AI Overview URLs. The finding: only 37.1-37.9% of pages cited in Google AI Overviews hold a top-10 ranking position for the same query. In July 2025, that figure was 76%. (Ahrefs, "Update: 38% of AI Overview Citations Pull From The Top 10," March 2, 2026)
That is not a refinement of an existing pattern. That is a reversal. The assumption baked into most answer engine optimization (AEO) frameworks, that strong organic ranking is the primary lever for AI Overview citation, is no longer supported by the data.
Gemini 3's Query Fan-Out Changed How Google Selects Sources
Google migrated AI Overviews to Gemini 3 on January 27, 2026, five weeks before the Ahrefs study was published. The timing lines up with the shift. Gemini 3 expanded a process Google calls "query fan-out": a single user query gets split into multiple related sub-queries, each evaluated for the most relevant source independently. Under this model, ranking position on the original query carries less weight than whether a page directly addresses one of the generated sub-queries.
The practical implication: a page ranking #14 for "best running shoes for flat feet" might rank #2 for a Gemini-generated sub-query like "motion control shoes for overpronation," and get cited for the latter without appearing in the top 10 of the former. The Ahrefs data supports this at scale: 31.0-36.7% of AI Overview-cited pages do not appear in the top 100 at all for the triggering query.
To be precise about what is established versus inferred here: the correlation between Gemini 3's deployment date and the citation-rank shift is strong. The fan-out mechanism is documented Google behavior. But Ahrefs' study is observational; it identified the shift, it did not isolate every variable that produced it. Treat the Gemini 3 explanation as a high-confidence hypothesis, not confirmed causation.
Top-10 Pages Are Still the Most-Cited Group. That Is Not the Point.
The counterargument practitioners should make to themselves before adjusting strategy: top-10 pages still have a higher individual probability of citation than lower-ranked pages. If you rank first, you are more likely to be cited than if you rank 110th. That is true.
The issue is coverage, not probability. When 62% of citations go to pages outside the top 10, rank-based optimization leaves most of the citation surface unaddressed. You are optimizing for 38% of the outcome space and ignoring 62%. That is not a minor blind spot; it is a category error in the measurement framework.
The stakes for that error are substantial. Seer Interactive's 15-month study across 3,119 queries and 42 organizations found that brands cited inside AI Overviews achieve +35% higher organic CTR and +91% higher paid CTR compared to uncited competitors on the same queries. If citation drives those outcomes, and citation now tracks poorly to rank, the practitioner question shifts from "how do I rank better" to "how do I get cited regardless of rank."
Most SEO programs are not set up to answer the second question. Rank tracking tools report positions 1 through 10 and call it AEO coverage. They do not. Ranking in the top 10 was a reasonable proxy for AIO citation when the correlation held at 76%. At 38%, it is a lagging indicator pointed at the wrong signal.
"The Great Decoupling" Was the First Signal. The March Data Is the Confirmation.
The citation-rank separation did not appear in one month. Ahrefs' June 2025 study, "The Great Decoupling", tracked the impressions-to-clicks correlation across 300,000 keywords and found it reversed direction: from +0.425 in late 2024 to -0.352 in early 2025. That shift tracked the 116% expansion of AI Overview presence following Google's March 2025 core update. More impressions began to mean fewer clicks; two metrics that once moved together started moving in opposite directions.
The March 2026 citation-rank data is the supply-side version of the same observation. Two separate measurement angles, both pointing at the same structural change: the ranking system and the citation system are now running on different inputs.
YouTube at 18.2% of Non-Ranking Citations Is a Structural Signal, Not an Anomaly
One figure in the Ahrefs study receives less attention than it warrants. YouTube is now the single most-cited domain in AI Overviews, accounting for 18.2% of all citations from pages outside the top 100 and 5.6% of total citations. That is a 34% increase over six months. YouTube videos do not rank in Google's text web index for informational queries in any conventional sense.
Search Engine Journal's coverage of the study framed this as evidence that Google is prioritizing "trustworthy, structured information" regardless of organic position. A more precise interpretation: Google is sourcing from content that directly answers a discrete sub-query in a format Gemini can extract cleanly. Video transcripts, with their natural Q&A structure and temporal segmentation, fit that format. So do FAQ sections, long-form explainers with clearly labeled answer units, and any content where the answer to a question is self-contained rather than distributed across paragraphs.
This is correlation data; the mechanism is inferred from the pattern, not from documentation. But if your AEO content mix is entirely conventional prose and your citation rate is flat, the YouTube citation share is worth examining as a structural diagnostic.
Running Separate Tracking Programs for Rank and Citation
If ranking and citation are now measuring different outcomes, they need separate measurement programs. The starting point is straightforward.
Pull your top 50 informational queries from Search Console. For each, check whether an AI Overview appears; this can be done manually or with a rank tracker that flags AIO presence. Then cross-reference against citation data from tools like Ahrefs Brand Radar or SE Ranking's AIO tracker, or from a structured manual sample. The question you are trying to answer: for queries where you hold a top-3 position, are you being cited? For queries where an AIO appears and you are not cited, which pages are cited instead, and what do they have structurally that yours lack?
That gap analysis is the starting point for a citation strategy running separately from your ranking strategy. It will not be the same page list, and the optimization work will differ.
One note on tooling constraints: citation tracking at scale is still maturing. Most rank trackers were built to report position, not citation. Budget time for manual sampling alongside automated reporting. Automated data will tell you when an AIO appears; it will not reliably tell you which page is cited and why.
The 30-day goal is not a complete AEO tracking system. The goal is to know whether your citation rate moves independently of your ranking position. If it does, the decoupling is present in your data, not just in Ahrefs'.
This content was produced with AI writing assistance.