Key takeaways
- Ranking and citation are different systems. Ahrefs' 2026 analysis of 863K keyword SERPs found only 38% of pages cited in AI Overviews also rank in the organic top 10 for the same query. The rest rank on page two, beyond position 100, or not at all.
- AI Overviews cite roughly 10 sources per answer, about double ChatGPT's ~5, per Promptwatch's cross-engine data. More citation slots means more ways in, but each one gets less exclusive attention.
- Query fan-out is the main reason the source pool diverges from the SERP. Google splits your query into sub-queries and pulls from those SERPs, not just the one you typed.
- Content format matters more than raw authority. In July 2026, product pages overtook listicles as the most-cited format in AI Overviews, and YouTube gets cited at 4%+ while ChatGPT barely touches video.
- Click behavior changes everything. Pew Research found users click organic results in 8% of visits when an AI summary is present, versus 15% when it isn't.
Summary: The short version
Traditional search gives users a ranked list of links and lets them pick. AI Overviews synthesize an answer from multiple sources and show it above those links. That sounds like a cosmetic difference. It isn't.
The deeper difference is in source selection. Regular Google search ranks pages for a single query using page-level relevance and authority signals. AI Overviews run a retrieval process closer to what other AI engines do: they expand the query into sub-queries, pull passages from a much wider pool of pages, and cite the ones that best support a synthesized answer. A page can be cited without ranking in the top 10. A page ranking #1 can be skipped entirely.
The numbers back this up. Ahrefs' updated analysis (863K keyword SERPs, 4M AI Overview URLs) found that only 38% of AI Overview citations also rank in the traditional top 10 for the same query. That's down sharply from the ~76% overlap their earlier 2025 study found. Google is citing a different pool of sources than the SERP shows, and the gap is widening.
How regular Google search picks sources
Traditional search is, at its core, a ranking problem. Google crawls the web, builds an index, and for each query runs its ranking systems over that index to produce an ordered list. Position matters most: the top result gets the lion's share of clicks, and visibility falls off steeply from there.
The signals that decide position are page-level: relevance to the query, backlink authority, user experience signals, freshness where it matters. A single page competes for a single position on a single SERP. If you rank #1, you get the traffic. If you rank #8, you get crumbs.
One more thing worth remembering: the SERP itself is the product. Google shows ten blue links (plus ads, plus featured snippets, plus whatever else), and the user does the work of choosing. Zero-click searches existed before AI Overviews, but the default behavior was still "scan titles, click something."
How AI Overviews pick sources
AI Overviews sit on top of the same index, but the selection process is different in three important ways.
Query fan-out changes the eligible pool
Google's own documentation confirms that AI Overviews and AI Mode use query fan-out: the original query gets split into multiple related sub-queries, and the system retrieves results for each. Ahrefs attributes the shrinking top-10 overlap largely to this. A page that ranks well for a sub-query of your target query can get cited even if it doesn't rank for the query itself.
This is why "rank #1 and you'll get cited" stopped being reliable advice. The citation pool is built from a fan of SERPs, not one.
Passage-level extraction, not page-level ranking
Regular search ranks pages. AI Overviews extract passages. The system looks for chunks of text that directly answer a piece of the question, cleanly and unambiguously. A 3,000-word page with one perfectly-phrased answer paragraph can get cited for that paragraph while the rest of the page is irrelevant to the retrieval decision.
This favors different content shapes. Promptwatch's July 2026 data on AI Overviews citation types shows listicles at 18.0%, product pages at 16.3%, how-tos at 15.1%, and news articles at 13.5% of citations. Product pages overtook listicles as the single most-cited format in late July 2026, ending the month at 17.9% vs 16.2%. Listicles averaged ~26% in Q1 2026 and have been falling since. What wins is content that's structured for extraction: clear headings, direct answers, tables, and lists.
Consensus and credibility across sources
Because AI Overviews synthesize rather than quote, the system prefers sources whose claims align with each other. Outlier claims get filtered out even if the page making them ranks well. Google's official positioning says AI Overviews "are built to surface information that is backed up by top web results" and "are designed to show a range of sources" — note the plural. A single authoritative page isn't the unit of value; a corroborated fact is.
Side-by-side: AI Overviews vs regular search
| Factor | Regular Google search | AI Overviews |
|---|---|---|
| Result format | Ranked list of links | Synthesized answer with cited sources |
| Click behavior | User clicks through to a page | User often gets the answer without clicking |
| Ranking signal | Page-level relevance and authority | Source credibility and answer clarity across multiple pages |
| Content that wins | Optimized single pages | Clear, well-structured, citable content |
| Visibility measure | Position in results | Inclusion as a cited source |
| Sources per result | 10 blue links | ~10 citations per answer (Promptwatch) |
| Retrieval method | Page-level ranking | Query fan-out + passage-level extraction |
| Self-reference | Minimal | YouTube + google.com = ~6.5% of citations (Promptwatch, June 2026) |
That last row deserves a comment. Two of the top three most-cited domains in AI Overviews are Google-owned (YouTube at 4.16%, google.com at 2.30% in June 2026, per Promptwatch's citation share data). Google citing its own platforms for ~6.5% of all AI Overview sources is a concentration of self-reference with no equivalent in traditional organic results, where Google properties appear as results far less often.
The overlap problem: what the studies actually say
Here's where I have to be honest about the state of the research, because the numbers conflict.
| Study | Dataset | Top-10 overlap finding |
|---|---|---|
| Ahrefs (2026 update) | 863K keyword SERPs, 4M AIO URLs | 38% of cited pages rank in top 10 |
| Ahrefs (2025 original) | Earlier dataset | ~76% overlap |
| BrightEdge (16-month study) | Enterprise dataset | Grew from 32.3% to 54.5% |
| BrightEdge (Feb 2026 analysis) | Separate dataset | ~17% overlap |
| seoClarity | Broader "top 10 web results" match | ~90-94% (different methodology) |
The spread is enormous: 17% to 94%, depending on who's measuring and how. Some of it is methodology (matching exact URLs vs matching domains vs matching against broader result sets). Some of it is time period, since the overlap has been shifting month to month as Google tunes the system. The Ahrefs 2026 figure of 38% is from the largest and most recent dataset, and its direction of travel is clear: declining reliance on the literal top-10 SERP in favor of fan-out sub-query results.
The practical takeaway isn't "ignore rankings." It's that ranking is no longer a sufficient condition for citation. The two systems are related but distinct, and you have to track them separately.
Where AI Overviews sources come from: the data
Promptwatch's cross-engine citation data fills in what the overlap studies can't show: what kinds of sources actually get picked.
AI Overviews cites more sources than ChatGPT
Per Promptwatch's average sources per response data, Google AI Overviews cites roughly 10 sources per answer, about double ChatGPT's ~5, and this count has been notably steady over time. (Perplexity also sits at almost exactly 10.) The practical implication: AI Overviews offers twice the citation slots of ChatGPT, making it the more forgiving engine to break into for newer or mid-authority domains. The flip side is that each individual citation carries less exclusive attention than a scarce ChatGPT slot.
The citation mix is shifting commercial
Promptwatch's July 2026 citation-type data for AI Overviews shows product pages at 16.3% of citations, up from roughly 9% in January. Listicles, the long-dominant format, fell from ~26% in Q1 2026 to ~18% by July. Google is increasingly comfortable citing commercial pages directly in AI answers, which changes what kind of content you should be building if citation visibility is the goal.
Compare this to ChatGPT in the same month: product pages at 32.8%, nearly double AI Overviews' share, and video at just 0.1%. The engines have different appetites.
Video and community content get real weight
YouTube leads AI Overviews' social citations at ~4.1%, ahead of Reddit at ~2.7%. ChatGPT is the mirror image: it heavily favors Reddit (5.19% of citations) and all but ignores video. If your strategy is built entirely around text content, you're leaving the single biggest social citation channel in AI Overviews on the table.
One more engine-specific quirk from Promptwatch's review-platform data: on AI Overviews, G2 is the clear B2B review winner at ~0.2% of citations, 3-4x Capterra or Trustpilot. On ChatGPT, Trustpilot rivals or beats G2. Google's AI leans harder on structured B2B software review data.
What this means for your content strategy
If the source-selection systems differ, your optimization has to differ too. Here's what actually moves the needle for AI Overview citations, based on what the data shows about which pages get picked:
- Structure for extraction. Clear headings that mirror how people phrase questions, direct answer paragraphs, tables and lists over prose walls. The system extracts passages; make passages extractable.
- Cover the fan-out, not just the head query. If Google splits "best CRM for small business" into sub-queries about pricing, integrations, and ease of use, your content should have dedicated, clearly-labeled sections answering each. One page trying to rank for the head term is no longer the unit of value.
- Don't abandon rankings. Traditional authority signals still determine which sources are eligible to be pulled in. The generative layer sits on top of the ranking system, not instead of it. Keep the backlink and E-E-A-T work going.
- Get cited where your buyers verify. G2 for B2B software, YouTube for anything with a visual component, Reddit for community validation. These are citation sources in their own right, not just traffic channels.
- Track citations separately from rankings. Because the systems diverge, you need visibility data that measures citation presence, not just SERP position. Tools like Promptwatch track exactly this: which of your pages get cited in AI Overviews, which prompts trigger citations, and how your citation share trends against competitors.
The click question
All of this matters because visibility without clicks is a different economic proposition. Pew Research Center's March 2025 study (900 U.S. adults, 68,879 real Google searches) found that users clicked a traditional organic result in just 8% of visits when an AI summary was present, versus 15% when it wasn't. Roughly half the click rate. And users clicked a link inside the AI summary itself in only about 1% of visits.
So citations don't deliver traffic the way rankings used to. What they deliver is presence in the answer itself, which is where the user's attention now sits. Whether that's a good trade depends on what happens after the answer: Seer Interactive research reported in SeoProfy's 2026 data found brands cited inside AI Overviews see 35% more organic clicks than non-cited brands on the same SERP. The citation feeds back into traditional click-through. Familiarity converts when users do eventually click.
Where to go from here
The gap between ranking and citation is the defining SEO problem of 2026. The data says it's real, widening, and driven by a structural change in how Google retrieves sources. If your reporting still equates "ranks #1" with "visible," it's measuring the wrong thing for a growing share of searches.
Start by auditing the gap for your own site: which queries do you rank for but not get cited on, and vice versa. Promptwatch's free AI Brand Visibility Report is a reasonable starting point if you don't want to commit to a full platform yet. The point is to see the two systems side by side, because that's the only way to know which one you're actually winning.