Written by: Khalid SEO | AI Search + SEO Specialist in Malaysia
If you’re still watching your Google rankings and assuming that tells you how you’re doing in AI search too, that assumption stopped holding up sometime in 2026.
For most of 2025, ranking in Google’s top 10 was a pretty reliable sign you’d also show up in Google’s AI Overviews. Ahrefs’ research from mid-2025 found that 76% of AI Overview citations came from pages already sitting in the top 10 organic results. Rank well, get cited. Simple.
That number has fallen off a cliff. Ahrefs’ updated analysis, based on 863,000 keywords and roughly 4 million AI Overview URLs pulled from its Brand Radar tool, found that only 38% of AI Overview citations now come from top-10 pages. Some other datasets put the overlap even lower, closer to 17%. Depending on which source you look at, the relationship between “ranking well” and “getting cited” has gone from a strong bet to basically a coin flip, and for some query types, worse odds than that.
If you’ve been treating rankings as your main measure of search performance, this matters. You can be doing everything right by traditional SEO standards and still be invisible in the actual answers AI systems are giving people.
What Actually Changed
Ahrefs points to two things happening at once.
First, their own citation detection got better between the two studies, so part of the drop is just Ahrefs seeing more of what was already happening, not a sudden shift in how Google behaves. That’s a fair caveat, worth keeping in mind before assuming the whole gap is new.
Second, and this is the part that actually matters for strategy: Google’s AI Overviews increasingly rely on something called query fan-out. When someone searches and an AI Overview kicks in, Google isn’t just answering the exact words typed into the box. It splits that query into several related sub-questions behind the scenes, runs its own searches against those, and pulls citations from whatever pages show up consistently across that wider cluster. Google has confirmed fan-out is part of how this works, though it hasn’t said exactly how aggressively it’s being used or whether that’s changed recently.
Google also switched AI Overviews over to Gemini 3 as the default model, globally, on January 27, 2026. Ahrefs flags the timing as worth noting, but stops short of saying the model swap directly caused the citation shift.
Put simply, Google’s AI isn’t just checking who ranks first for your exact keyword anymore. It’s checking who covers the topic well enough to answer the question and the reasonable follow-ups around it. That’s a different job than traditional ranking, and it rewards different things. We broke down the mechanics of the SEO to GEO gap in more detail if you want the fuller picture.
Why a Page Can Rank #1 and Still Get Zero Citations
This is the part that trips people up, so it’s worth spelling out.
Traditional Google ranking runs on signals like backlink profiles, historical authority, click behavior, and relevance to one specific query. AI citation selection is doing something closer to retrieval: pulling out passages that answer a question clearly, that cover the sub-questions a reader would naturally have next, and that the AI system can actually reach and read cleanly.
A page can dominate its target keyword and still get skipped for citation if:
- It answers the headline question but doesn’t touch the related sub-questions fan-out queries are pulling in
- The content isn’t written in a way that lets a clean, self-contained answer get lifted out of it
- Something else, a video, a forum thread, a competitor page, covers the same ground more completely, even if it ranks lower or doesn’t rank at all for the original term
That last point is where it gets uncomfortable for a lot of sites. YouTube is now the single most-cited domain in AI Overviews. Ahrefs’ data shows it’s grown 34% over six months, and it makes up roughly 18% of citations that come from outside the top 100 organic results entirely. Reddit, Wikipedia, and Quora show up disproportionately too. None of these are showing up as your competitors in the SERP. You’d never see them coming by checking rankings. They’re winning the citation anyway.
Perplexity plays by different rules again. Ahrefs’ cross-platform work found Perplexity’s citations track much closer to Google’s own top 10, around 29% overlap, compared to roughly 8% for ChatGPT and Gemini. Perplexity was built as a citation-first answer engine, so that tracks. ChatGPT, on the other hand, pulled around 80% of its citations from pages that don’t rank at all for the original query. If you’re only watching one platform’s citation behavior and assuming it applies everywhere else, you’re probably missing most of the picture. Our piece on how ChatGPT’s fan-out queries actually work goes deeper on that specific platform.
What Actually Correlates With Getting Cited
Cyrus Shepard published a meta-analysis in May 2026 that scored 23 different AI citation factors across 54 studies, weighted by how strong the evidence actually was behind each one. The pattern that fell out of it: the things that predict a citation aren’t the same things that predict a Google ranking.
A few of the better-supported patterns worth knowing:
- Freshness matters more for citations than for rankings. Ahrefs looked at nearly 17 million citations and found AI-cited content runs about 25.7% fresher, on average, than typical top-10 organic content.
- Brand mentions across the web correlate more strongly with AI visibility than backlinks do. Shepard’s synthesis put that at roughly a 3x difference, even though backlinks are still a core traditional ranking signal.
- Accessibility isn’t optional, it’s a gate. If an AI crawler can’t reach or parse your page cleanly, quickly, and without JavaScript getting in the way, nothing else on this list matters. Both Shepard’s meta-analysis and SE Ranking’s own research point to page speed and crawlability as genuinely strong, well-evidenced factors.
- Overlap across platforms is rare. Only about 2% of cited URLs show up across AI Overviews, ChatGPT, and Perplexity all at once, per Shepard’s analysis. Around 91% of citations show up in just one AI engine. Winning on one platform doesn’t mean you’re winning on the others.
None of this is a promise that doing any single thing guarantees a citation. Every figure here is a correlation from an observational study, not a controlled experiment, and the people running these studies are upfront about that. Treat it as a map of where the evidence points, not a formula you can plug numbers into.
How to Check If You Have This Gap
You don’t have to guess whether this is happening to your own site. It’s checkable.
- Pull your top-ranking keywords from Google Search Console or Ahrefs. Focus on the ones you’re already in the top 10 for.
- Ask ChatGPT, Perplexity, and Gemini those same topics directly, in natural language. Not the exact keyword, the way a real person would actually phrase it.
- Check whether your page comes back as a cited source. Not just whether the topic gets mentioned, whether you’re the one getting credit.
- Look at your Google AI Overview presence separately from that, since Overviews and the standalone AI chat platforms clearly don’t behave the same way.
- See what’s getting cited instead, where you’re ranking well but not showing up. A video? A forum thread? A competitor covering more ground? That gap tells you exactly what to build next.
Google Search Console gives you some AI Overview impression data now, but it’s limited, and it won’t tell you anything about ChatGPT, Perplexity, or Gemini. We’ve written before about how to actually track your AI Overview presence if you want a more detailed walkthrough. Tools like Ahrefs Brand Radar and SE Ranking’s AI Overview tracker exist specifically to fill that blind spot.
Common Mistakes Worth Avoiding
Assuming AI Overviews and standalone AI chat platforms work the same way. They don’t. Perplexity leans heavily on traditional rankings. ChatGPT and Gemini largely don’t. Optimizing for one won’t automatically carry over to the others.
Treating this as a text-only problem. With YouTube pulling this much citation share, and Reddit and Quora close behind, ignoring video and community platforms means ignoring where a real chunk of citations are actually going.
Assuming this means SEO stopped mattering. It didn’t. Ranking in the top 20 still meaningfully improves your citation odds across most studies. It’s just not enough on its own anymore. Rankings are the floor now, not the whole building.
Chasing schema markup as the fix. One Ahrefs test tracking pages that added JSON-LD schema found no real change in AI citation behavior. It’s not harmful, but it’s not the lever a lot of GEO advice makes it out to be.
Giving up on tracking because it’s harder now. It genuinely is harder to track than a rank position. That’s a reason to bring in tools built for it, not a reason to stop measuring altogether.
What to Actually Do About It
The response here isn’t one clever tactic. It’s matching what the evidence actually shows these systems doing.
Build real depth on a topic, not just one optimized page. Since fan-out queries pull from a whole cluster of related sub-questions, a page that only answers the headline question and skips the reasonable follow-ups is easy to pass over for something more complete.
Write so a clean answer can stand on its own. A good test: would this paragraph still make sense to someone who only read that one paragraph, with zero surrounding context?
Push visibility work beyond your own domain. Brand mentions, a real YouTube presence, showing up in Reddit and Quora threads, these are turning up in the data as genuine citation drivers, not side projects.
Keep content actually current. The freshness gap between cited content and merely-ranking content isn’t small, and it’s one of the more consistently backed findings across several separate studies. For more on how to get cited in AI Overviews specifically, we’ve laid out a fuller playbook.
Track citations on their own, separate from rankings, on whichever platforms actually matter for your business. Not just Google AI Overviews by default.
This is the kind of diagnostic and strategy work I do for clients through Khalid SEO, closing the gap between where a site ranks and where it’s actually getting picked up by AI search. If you want a second set of eyes on where your own site stands, get in touch.
Sources: Ahrefs Brand Radar research (2025-2026), Cyrus Shepard’s AI Citation Ranking Factors meta-analysis (Zyppy Signal, May 2026), SE Ranking studies (2025-2026). Figures reflect the most recent available data as of publication and are attributed to their original sources throughout.