Most Google AI Overview Citations Come From Outside the Top 10 — And What to Do About It

According to BrightEdge (February 2026), only about 17% of the sources Google AI Overviews cite also rank in the organic top 10 for the same query. According to Ahrefs (March 2026), the figure is 37.9%. The estimates differ, but both say most links point to pages outside the top 10. If so, a position in the results and a place in an AI answer are different things. Below: where the numbers come from, how they might be explained, and what you can do on your site.

Where the numbers come from

In its review of February 12, 2026 (analysis period February 2025 to February 2026), BrightEdge writes: “Only about 17% of sources cited in AIOs also rank in the organic top 10,” and says the share barely moved. The complement to 100% (about 83%) is our arithmetic, not a BrightEdge figure. This is the vendor’s own data and its method is not detailed (source).

In its update of March 2, 2026, Ahrefs analysed 863K keyword results and about 4M URLs from AI Overviews: 37.9% of the cited URLs also appeared within the first 10 blocks of the results page. In July 2025 the same Ahrefs found 76.1% across 1.9M citations (the top three citations of each answer). Ahrefs notes that its method has improved since, so the two numbers cannot be compared directly (2026, 2025).

Google does not publicly describe how it picks links for AI Overviews, so any explanation below remains a hypothesis.

Why AI systems may not just use the top 10

One possible explanation is retrieval-augmented generation (RAG), a scheme many AI systems use. The system first searches a much broader set of documents than the top 10 for texts close to the question, then hands them to a language model that writes the answer. Whether Google uses exactly this scheme in AI Overviews, we do not know in detail.

Such schemes usually rely on vector similarity: a numeric measure of how close a document’s meaning is to the question. A page at #3 for the broad query “project management software” may be far from the question “what suits a remote engineering team using Jira?” A page at #31 that answers exactly that may come out closer.

If so, the gap between position and a link in an answer is explainable: they may use different criteria.

What may influence which sources are chosen

These are hypotheses, not results of our own measurements. Three properties make a page clearer to both people and machines.

1. Semantic Completeness

The page must directly, comprehensively answer the specific question being asked — not the broad topic, but the exact question variant. A page about "email marketing" has low semantic completeness for the query "how often should SaaS companies send onboarding emails." A page with a dedicated section titled "Onboarding Email Frequency for SaaS Products" has high completeness for that query.

Practical implication: structure your content around specific, precise questions rather than broad topics. Each section should be answerable independently. AI systems, as RAG is described, work with passages rather than whole pages, so each section is worth making self-contained.

2. Content Structure and Scannability

Structured text is easier for both people and machines. In particular:

• FAQPage JSON-LD — marks up your content as question-and-answer pairs. An effect on citations is not proven; see the FAQ schema article for details.
• Numbered lists with clear items — each item becomes an extractable data point. "The top 5 reasons X happens" is easier to cite than a paragraph explaining the same 5 reasons.
• Definition-style headings — headings that contain the answer, not just the topic. "Email open rates are declining because of Apple MPP" vs. "Why Email Open Rates Are Declining."
• Concrete numbers and statistics — a page with concrete facts is more useful to the reader and easier to quote.

3. Freshness

People asking AI questions expect current information, and a visible date helps both the reader and the machine. How much freshness weighs in choosing sources is not publicly known.

A page’s date can be stated in several ways: dateModified in JSON-LD, the last-modified HTTP header and a visible publication date. If a good article has not been updated for a long time, check that the date on the page is present and true.

What you can do even when you rank #47

A reasonable approach is to think less about position and more about how precisely a page answers a question. It does not guarantee a result. Here are five steps.

Step 1: Find your audience’s questions

Find the specific questions your audience asks AI services. They are longer and more conversational than ordinary keywords. Customer questions to support, Perplexity’s suggestions and Google’s “People also ask” box can give ideas.

Step 2: Give each question its own section

For each question, create a section on the most relevant page (or a new page) with the question as the heading and an answer that reads on its own. Put the answer right after the heading so readers do not have to hunt for it.

This is the answer-first structure journalists use: the conclusion first, then the detail. It suits a person and any program that takes the first sentences of a section.

Step 3: Add FAQPage markup

Once the answer sections are ready, add FAQPage JSON-LD that mirrors them. It gives a machine a clean version of the answers without parsing the layout. Whether it increases citations is unknown: there is no public data, and we have no measurements of our own.

How to add the markup is in FAQ Schema and AI Citations: What the Markup Does, What We Know, How to Add It.

Step 4: Update date signals

Add or update datePublished and dateModified in your Article JSON-LD. Make your publication date visible on the page (not buried in a footer). If you update content, change the dateModified value. Both the reader and a machine can see at once how fresh the material is.

Step 5: Include specific statistics

Replace vague claims with figures and a source. Instead of “many companies use email marketing,” give a number with the source and year. Such text can be checked, which helps any reader.

Pages beyond the top 10

BrightEdge writes that the share of AI Overview sources ranking somewhere in the top 100 for the query rose slowly from about 49% to 53%, with the growth coming from pages 2 through 10 of the results. It gives no separate figures for positions 11–50. One explanation: such pages are often narrower in topic than the top-10 results and answer specific questions. That is an assumption we cannot verify.

If your pages sit beyond the top 10 for important queries, it makes sense to make them more precise and easier for a machine to read. Whether they end up in AI answers cannot be said in advance.

What this means for SEO

Even if most links point beyond the top 10, it does not mean SEO is worthless: Google says its AI features work over the regular search index, so crawler access and page quality still matter.

The old model: reach page one and traffic follows.

A possible new model: be the most precise answer to specific questions, whatever the position for the broad query.

Frequently Asked Questions

Related reading