AI Citations: 62% of Sources Aren't in Google's Top 10

On March 2, 2026, Ahrefs published an analysis of 863,000 queries and 4 million URLs cited by Google's AI Overviews, authored by Louise Linehan and data scientist Xibeijia Guan. The headline finding is stark: just 38% of the pages cited in those AI answers also rank in the organic top 10 for the same query. A year earlier, in July 2025, that share stood at 76%. In eight months, the link between "ranking well on Google" and "being cited by AI" has roughly halved.

The rest of the citations split almost evenly between pages sitting from positions 11 to 100 (31.2%) and pages beyond position 100 (31.0%), i.e. invisible in traditional search. In other words, nearly a third of the sources reused by AI Overviews don't even rank in the top 100. The finding corroborates a BrightEdge observation, which measures over sixteen months a top-10 citation share holding steady around 16–17%, with most of the overlap coming from pages ranked between positions 21 and 100.

Why Google cites pages it doesn't rank first

The cause is a mechanism documented by Ahrefs: query fan-out. When a query triggers a generative answer, Google doesn't just pull from the main SERP. It breaks the question into several sub-queries, runs a search for each, then aggregates the pages that appear most often across those secondary SERPs. A page can therefore be cited because it perfectly answers a specific sub-question, even if it's neither first nor second on the original query.

The direct consequence: AI citation rewards granular coverage and passage-level relevance, not just a page's overall authority on a keyword. It's a shift in logic we describe in our analysis of the formats AI Overviews reuse and recommend: the model looks for the best answer to each fragment of intent, not the page that ranks best on the surface.

What this concretely changes for a business

For twenty years, a leader's SEO goal fit in one phrase: "rank on Google's first page." That target remains useful, but it no longer guarantees visibility where more and more buying decisions now start: inside AI-generated answers. Content can now be cited without ranking first, and stay first without ever being cited. Here's how to adapt.

1. Cover intent by fragments, not just by keyword

List the real sub-questions around your target query ("how much does it cost," "who is it for," "alternatives," "how to choose") and address each with a self-contained, quantified, dated passage. That's exactly the logic of our approach to trust signals and content built for AI: you don't write a page for a keyword, you write answers that fan-out can isolate and cite.

2. Make every page citable, not just rankable

A direct answer up top, clear definitions, sourced data, explicit headings, machine-readable structure. Those are the signals that make a model reuse your sentence instead of a competitor's. It's the heart of Generative Engine Optimization (GEO), distinct from classic SEO even though it builds on the same technical foundations, and it's reshaping how visibility is fragmenting across ChatGPT, Gemini and Claude.

3. Stop neglecting your "poorly ranked" pages

A page in position 40 on the main query can be a citation goldmine if it answers a sub-question very well. Before unpublishing or merging it, check whether AI already reuses it. The ranking/citation decoupling turns some pages deemed "weak" into underrated GEO assets.

4. Measure the traffic AI already sends you

Traffic from AI assistants is often mis-attributed, and therefore invisible in the usual dashboards. This matters more as the AI answer layer becomes a durable distribution channel. Set up dedicated tracking so you can see, segment by segment, which qualified visitors arrive from ChatGPT, Claude and Gemini.

What this study does not say

• The main scope is Google's AI Overviews. A separate Ahrefs study, published August 11, 2025 across 15,000 prompts, shows the top-10 overlap is even lower for conversational assistants (around 12%), with Perplexity an outlier at 28.6%. Each engine has its own citation logic.
• Ranking still holds value. The decoupling doesn't mean SEO is useless. It means ranking first no longer guarantees citation, and that citation becomes a goal in its own right, to be measured and worked separately.
• Numbers vary by method. BrightEdge, Ahrefs and others use different samples and definitions; the orders of magnitude converge, the exact values don't. Read this as a strong trend, not a measurement to the decimal.
• The data is a snapshot. Citation systems evolve fast. A decline measured in March 2026 describes a dynamic, not a carved-in-stone law: track it quarter by quarter, sector by sector.

Frequently asked questions

The Cicéro take

The useful message for a business leader is simple: visibility has split in two. There are now two contests to win, the Google ranking on one side, the citation by AI engines on the other, and they follow increasingly distinct rules. Steering only the first means optimizing for a shrinking field while another one opens. The good news: both feed on content that is clear, sourced and built to answer, not to fill space.

That's exactly the Cicero Studio methodology: GEO audit + editorial production + automated semantic internal linking, with a minimum quality score of 90/100 on every piece. Agency-quality work, software-grade productivity. If you want to know whether your pages are ready to be cited by AI, and not just ranked by Google, we'll diagnose it for free. We show you where you're cited, where you're not, and the three priority actions to change that.

Alexis Dollé

CEO & Founder

Growth and SEO content strategist, I founded Cicéro to help businesses build lasting organic visibility, on Google and in AI-generated answers alike. Every piece of content we produce is designed to convert, not just to exist.

LinkedIn