“We’ve never been able to crack the top of Google. So once everyone’s searching through ChatGPT and AI Overviews, we’re just going to get buried even deeper.” I’ve been hearing some version of this in marketing meetings a lot lately.
The logic feels airtight. The big incumbents who already win at search get cited by the AI too, and everyone else falls further behind. If you’re losing on the one scoreboard you understand — search rank — why would the new game go any differently?
I get the instinct. When you’re behind on a clean, legible metric, it’s natural to assume you’ll lose on the next one too.
But when you actually look at which pages the AI is pulling from, that pessimism turns out to be jumping the gun. So let’s dig into that.
Why You Can Rank #1 And Still Go Unmentioned By AI
Start with one thing: ranking well in search and getting cited by an AI are no longer the same game.
“AI visibility” is basically how often the AI’s answer names, quotes, or links to you. And that’s drifting away from your old search rank in a way that catches people off guard. A page sitting at #1 in search goes completely unmentioned in the AI’s answer. A page that was buried halfway down the results quietly shows up inside the AI’s reply. That mismatch is turning up all over the place.
So your search rank — the one number you’ve been judging yourself by — doesn’t cleanly map to where the exposure lives now.
More Than Half Of AI Citations Come From Page Two And Beyond
So how big is the gap, really? A few firms went and measured it, each on their own data.
First, a quick definition. “AI Overviews” is the AI-written summary that sits at the top of a Google results page. Several analysts have tracked which pages those summaries actually cite — and the answer is not “the top ten.”
- Ahrefs found that only about 38% of AI Overviews citations come from the top-10 search results. The other ~62% are pulled from rank 11 and below.
- 5W Research reports that the overlap between top Google rankings and AI-cited sources has collapsed from 70% to under 20%.
- BrightEdge, tracking the same thing across 16 months, saw citation/rank overlap rise from 32% to 54% — but most of that growth was in pages ranked 21-100, and the top-10 share was only about 16.7%.
The numbers swing around a lot. But every one of them points the same direction: the AI isn’t just reading the page-one winners. It’ll happily quote a page ranked 11th, or 50th, if it decides the content is useful for the answer.
One honest caveat before anyone gets too excited. These are all commercial analyses, each built on a different panel, a different time window, and a different definition of what counts as a “citation.” So you can’t line up that 38% against the 16.7% and call one right. But when several studies with mismatched methods all lean the same way, that convergence is worth taking seriously.
Why The Gap Exists: The AI Grabs What’s Easy To Quote
So why doesn’t the AI just copy the search rankings? Understanding this makes your own next move a lot clearer.
The short version: the AI is picking pages by “is there something here I can drop straight into the answer,” not by “is this page winning at search.” Those two things used to travel together. They don’t anymore.
There’s a whole research thread on making content easier for generative engines to cite — it’s called GEO (generative engine optimization), basically SEO aimed at the AI’s answer instead of the ranked list of links. And the research on GEO keeps landing on the same finding: what reliably works isn’t search-side tricks, it’s substantive, grounded content the model can actually stand on.
So the techniques that win you a ranking and the substance that gets you quoted aren’t the same lever. A page polished purely for the search engine can get skimmed right past by the AI. Which, when you think about it, makes complete sense.
But Don’t Rip Out Your SEO In A Panic
Read this far and you might conclude rank no longer matters at all. It’s not that simple.
Right now, today, most of the traffic to your site is still coming from plain old search. The AI citations drifting away from rank doesn’t mean your search traffic vanishes overnight. Yanking out your SEO foundation in a panic is like tearing out a load-bearing wall mid-renovation because you got excited — a little reckless.
The move isn’t switching, it’s re-centering. Keep search rank as the base that protects the traffic you already have, and grow “designed to get cited by AI” as a separate effort, on its own budget line. Two muscles, side by side.
Conclusion: There’s Exposure Left Out Past The Rankings
So here’s the one thing I’d have you take away. Even if you’ve never won at search, there’s real room to claw back exposure in the AI era.
The AI cites from outside the top 10, so missing the lead pack on rank doesn’t lock you out. Put grounded, citable substance on the page and you can climb back onto the field as a citation, regardless of rank. If anything, this might matter most to the latecomers who could never quite win the ranking race.
With that in mind, the work on the ground comes down to two things, and only two.
First, build the habit of measuring how often you’re cited by the AI — as its own number, separate from rank. Stare only at the rankings table and you’ll miss both the citations you’re winning and the ones you’re leaking.
Second, put evidence on your pages that the AI can lift straight into an answer: concrete numbers, primary sources, plain clear explanations. Treat it as a different job from climbing the rankings, and that’s the shortcut.
It’s early, and I won’t pretend any of this is settled. But “we can’t rank, so AI’s hopeless for us” looks, on the data, like giving up a beat too soon. If it nags at you, start by just counting how your main pages are getting cited.
Sources
- [R] Ahrefs (2026), “Only 38% of Google AI Overview Citations Come From Top 10 Pages” (via DesignRush), DesignRush
- [R] BrightEdge (2026), “AI Overview Citations and 16 Months of AIO Rank Overlap”, BrightEdge
- [R] 5W Research (2026), “Overlap Between Top Google Rankings and AI-Cited Sources Has Collapsed From 70% to Under 20%”, PR Newswire
- [R] Aggarwal, P. et al. (2024), “GEO: Generative Engine Optimization”, KDD 2024, arXiv
