Ranking #1 on Google Used to Get You Cited by AI. Three Studies Now Say That's a Coin Flip, and the Odds Are Getting Worse.

The coin flip Ahrefs was honest enough to name out loud

Here's a sentence I want you to sit with for a second, because it's not mine. It's Ahrefs', from a study they published on July 21, 2025, after running the numbers on 1.9 million citations across 1 million AI Overviews.

"If you rank #1 in the SERPs, you're more likely to be cited in an AI Overview than if you were ranked lower. But that chance is a coin flip at best."

Read that again slow. Not "you'll probably get cited." Not "it helps some." A coin flip. At best. At the top of Google's results. Where every plumber, every HVAC tech, every dentist in America has been told for two decades that if you just get here, you've won.

That was already true in July of 2025. It has gotten worse since. And the mechanism behind why it's gotten worse is public, documented, and traces to a patent Google filed back in 2018. None of this is speculation. So let's walk through it the way I'd walk a friend through it, one piece at a time, because the small business owner who hears "we already rank on page one" from their web person deserves to know exactly what that sentence does and doesn't buy them anymore.

Three studies, laid out straight

I'm going to give you dates and sample sizes on purpose, because vague numbers are how this kind of story usually gets told, and vague numbers are easy to wave away. These aren't vague.

Ahrefs, July 21, 2025. 1.9 million citations, 1 million AI Overviews. 76.10% of the cited pages ranked in Google's top 10 for that same query. 9.50% ranked somewhere between 11 and 100. And 14.40% of cited pages didn't rank in the top 100 at all, meaning they weren't really "ranking" in any meaningful sense and still got cited. The median rank of the first-position cited URL was position 2. So at that point in time, ranking well really did correlate with getting cited. Ahrefs called it "a positive yet moderate correlation," which is a careful way of saying what it says in the next sentence: a coin flip at best, even at #1.

Ahrefs, reported March 2, 2026. Same methodology, bigger sample this time, 863,000 keywords, 4 million AI Overview URLs. Only 38% of AIO-cited pages now rank in the top 10 for the same query. That's down from 76% eight months earlier. Roughly 31% rank somewhere in the 11 to 100 range, and roughly 31% don't rank in the top 100 at all. Search Engine Journal, in the piece reporting this (byline Matt G. Southern), attributed part of that drop to "improved parsing methodology and Google's query fan-out process." We'll get to what that phrase actually means in a minute. For now, whatever relationship existed between "ranks well" and "gets cited," it roughly cut in half in eight months.

BrightEdge, a 16-month study running May 2024 through September 2025. Different tool, called Generative Parser, measuring a related but genuinely different thing. BrightEdge found that 54.5% of AI Overview citations now come from pages ranking organically anywhere on Google, up from a 32.3% baseline when they started tracking. But narrow that down to specifically top-10 rankings, and it's only 16.7%. And underneath that overall number is a spread that should stop you cold if you're trying to generalize about "how AI search works" from any one industry: healthcare sits at 75.3% overlap between ranking and citation, education climbed 53.2 percentage points to land at 72.6%, e-commerce is stuck at 22.9% and basically hasn't moved, and restaurants sit at 19.2% after starting from zero.

Three studies. Three different numbers for roughly the same question. And here's where a lot of writing on this topic goes wrong: it picks the number it likes best, drops the other two into a footnote, and calls it a day. I don't want to do that, because the disagreement between these numbers is actually the most useful thing in this whole post.

Why the numbers don't agree, and why that's the finding

Let's be precise about what's actually happening here, because "two studies disagree" can mean a lot of different things, and most of them are boring. This isn't boring.

Ahrefs' 38% and BrightEdge's 16.7% look like they're answering the same question and landing more than 20 points apart. They're not answering the same question. Ahrefs' 38% is measuring top-10-only rank against citation. BrightEdge's 16.7% is measuring that same narrow slice, top-10-specific, but through a different tool built on a different parsing methodology, over a different window of time that ends in September 2025, five months before Ahrefs' March 2026 update. Different instrument, different clock, adjacent question. That's not a contradiction. That's two people measuring a moving target from two different angles and getting two different snapshots.

Now stack BrightEdge's own industry table on top of that, and the "there's one true percentage" idea falls apart entirely. Healthcare at 75.3%. Restaurants at 19.2%. That's a 56-point spread inside a single study, using a single tool, over a single time window. If the rank-to-citation relationship behaved consistently, you wouldn't see that kind of range from one dataset alone, let alone see two separate firms land 20-plus points apart from each other on top of it.

Here's the honest way to read this, and I think it's more useful than picking a side. Ahrefs and BrightEdge are both serious, well-resourced research operations. Neither one is sloppy. Neither one has an obvious reason to cook a number in either direction. And they still can't agree within 20 points on how much your Google rank predicts whether AI cites you. That disagreement isn't a reason to distrust either study. It's evidence, on its own, that the relationship between ranking and citation is not stable enough for anyone, including Ahrefs and BrightEdge, to nail down with a single clean percentage. If the ground were holding still, two careful teams measuring it would land closer together than this. They don't. That tells you something the individual numbers can't tell you by themselves.

So don't walk away from this section with a number memorized. Walk away with the shape: whatever the relationship between rank and citation is, it's moving, it's inconsistent across industries, and even the best people trying to measure it can't pin it down. That instability is worse news for the "I already rank well, so I'm covered" assumption than either single number would have been on its own.

The mechanism: query fan-out

Here's the part that explains why the numbers are moving in the first place, and it isn't a mystery. Google told everyone directly.

At Google I/O on May 20, 2025, Google introduced the retrieval mechanism behind its AI Mode in its own words: "AI Mode uses our query fan-out technique, breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf." That's not an SEO blogger's theory about how Google works. That's Google's own announcement, on its own blog, on the record.

The technique traces back to a real patent, US11663201B2, titled "Generating query variants using a trained generative model," filed in 2018 and granted in 2023. What it describes, in plain terms, is a model trained to take one question and generate a bunch of variants of it, follow-up questions, broader versions, narrower versions, translated versions, and more, and then score each variant on its own. So when someone types one question into Google, that one question gets split into a handful of related sub-questions behind the scenes, and each sub-question goes out and gets its own answer, retrieved and evaluated independently. Search Engine Land's guide to the process describes what happens next: an LLM reviews everything that came back from all those sub-queries and looks for patterns and overlaps before it writes the answer you actually see.

Now here's the detail that actually matters to you, the one that explains everything above it. According to iPullRank's analysis of this mechanism, the selection of sources under fan-out happens at the passage level, not at the level of the whole page. Their own words: "a single paragraph from a 500-word post on a small blog can outrank a 5,000-word ultimate guide from a major publication, if that paragraph is a sharper match for a specific sub-query." Wellows describes the same shift as content getting scored "paragraph-based" instead of "document-based."

Sit with what that actually means for a plumber in a mid-size city. Your competitor across town might out-rank you overall, might have the bigger site, the longer blog posts, the better domain history. None of that is the fight anymore, not entirely. If your page has one clean paragraph that answers one specific sub-question, say, "does this company offer emergency water heater repair on weekends," better than their page does, that paragraph can get pulled into an answer even while your competitor's overall page keeps beating yours in the regular rankings. You don't have to win the whole war. You have to win the specific skirmish that's actually about you.

One more thing worth being straight about here, because it gets repeated a lot and stated with more confidence than it deserves. You'll see people describe "8 types of sub-queries," equivalent, follow-up, generalization, specification, canonicalization, translation, entailment, clarification, as if Google published that exact list somewhere. Google didn't. That taxonomy is the SEO research community's own reading of the patent's language, credited to analyses from iPullRank and Wellows, not an official list Google itself hands out. It's a reasonable and useful breakdown. It's just worth knowing where it actually came from before you repeat it to a client as gospel.

It's bigger than Google

Everything above is a Google story specifically, query fan-out is Google's mechanism, its own patent, its own announcement. But the underlying pattern, that the old competitive picture doesn't hold up the way it used to, shows up outside Google too.

"The State of AI Citations 2026," published by 5W Public Relations in May 2026, pulled together six independent citation datasets to get a wider view: Profound at 680 million citations, Goodie, Surfer's AI Tracker at 46 million citations across 36 million AI Overviews, Semrush tracking 230,000 prompts over 13 weeks, Peec AI at 30 million citations from a March 2026 snapshot, plus supplementary data from Ahrefs, BrightEdge, and WebFX. Two of the findings from that synthesis are worth knowing cold. Wikipedia accounts for 47.9% of ChatGPT's top-10 source share. Reddit accounts for 46.7% of Perplexity's. And only an estimated 11% of domains get cited by both ChatGPT and Perplexity at all.

Think about what that actually means for the fight you thought you were in. You probably assumed your competition, for AI citations, was the same competition you've always had, the other plumber, the other dentist, whoever's been beating you in Google rankings for years. That was never quite the real picture, and now it's even less true. On the broad, general questions, "how does a water heater work," "what does a root canal involve," the citation pool is dominated by Wikipedia and Reddit, not by any local business's website, regardless of how well that business ranks. Your actual shot isn't the broad informational question. It's the narrow one that's specifically about you: your hours, your service area, whether you do a specific job on a specific day. That's the sub-query nobody but you can answer correctly, and it's the one you actually have a chance at winning.

What's actually in your control

So given all of that, what does an audit like ours actually do for you, and where does it stop being able to help? I'd rather tell you the honest boundary than let you assume it covers more ground than it does.

Our AI Crawler Access module checks whether five named bots, GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, and Bingbot, can actually reach your site through robots.txt. There is no Google-specific bot on that list, and there isn't one because there isn't one to check. Google's AI Overviews and AI Mode draw from the same standard Googlebot-fed index that regular search results come from. There's no separate crawler you could accidentally block that would keep you out of an AI Overview specifically. Whether you show up there is governed by the ranking and fan-out mechanism this whole post has been about, not by anything sitting in a robots.txt file. I want to say that plainly rather than let this module sound bigger than it is: it cannot move the needle on Google AI Overview citation. That's not this check's job.

What does generalize, across Google's fan-out and every other engine in that 5W Public Relations synthesis, is the stuff our Content Structure module checks (question-format headings answered directly underneath them, an actual FAQ section, information organized into lists instead of buried in paragraphs) along with what Structured Data and Entity & Trust check (clean schema markup, consistent business facts, a page that clearly states who you are and where you operate). Those checks reward exactly the kind of content that's easy to lift as a clean, self-contained passage, which is precisely what a sub-query grader is looking for, on any engine, independent of whose crawler is doing the looking.

I want to be careful about the claim I'm making there, because it would be easy to oversell it. I'm not telling you our audit gets you into AI Overviews. Nobody can promise you that, and anyone who does is selling you an outcome they don't control. What I am telling you is that the mechanism research above explains why the specific things our audit already checks are the right things to fix, regardless of which engine ends up doing the citing. That's a narrower claim than "buy this and get cited." It's also the true one.

Stop treating page one as the finish line

For years, "rank #1 for plumber plus your city" was the whole game. Get there, and you were done. That was never as safe an assumption as it sounded, and the research above is the receipt. Even at the top of Google, in the best data anyone has, it was a coin flip. Eight months later it looked worse. And the mechanism behind why is a patent from 2018 that Google put a name to in public in 2025.

The narrower, more winnable target isn't outranking the plumber down the street anymore. It's making sure the specific facts about your business, your hours, your service area, the credentials that back you up, direct answers to the actual questions customers ask, sit somewhere on your own site in clean, structured, directly extractable form. That's the sub-query you can actually win, on any engine, no matter where you land in the overall rankings.

If you want to know where your own site stands on exactly that, the free Grade Check on aeocheck.net scans one page and hands you a real score off the same scoring engine behind our paid audits. Takes about a minute, costs nothing, and it's a specific answer instead of a guess.