Why Top Rankings Lost 50% of AI Citations in 2026

Here is the stat that reset half our planning meetings this year. In July 2025, when Google's AI Overviews cited a page, that page ranked in the top 10 for the same query about three quarters of the time. By early 2026, that overlap had roughly halved. Ranking number one no longer buys you a seat in the AI answer the way it did eighteen months ago.

That is the heart of the AI Overviews citation rate story, and it is genuinely one of the more useful shifts we have tracked across our clients. We are an SEO team, so we read these reports for a living, and most of the panic-flavoured takes floating around miss what the data actually says. Let's go through the real numbers, what they mean, and what we would actually change in your strategy. No horse-and-buggy metaphors, we promise.

What is AI Overviews citation rate?

AI Overviews citation rate is how often your content gets named as a source inside Google's AI-generated answer box, rather than just where you sit in the blue-link list below it. It is a separate question from your ranking. You can rank first and never get cited. You can sit on page three and get cited daily.

That separation is new, and it is the whole point. A traditional ranking measures your position in a list. A citation measures whether Google's models decided your page was the best evidence to build an answer from. Those are not the same job, and in 2026 they have drifted apart.

If you want the wider context on how these answer boxes are eating into traffic, our piece on the AI Overviews click-through rate drop in 2026 covers the downstream damage. This article is about the upstream question: who gets named as a source in the first place.

The 76% to 38% drop, and what it really means

The number everyone is quoting comes from Ahrefs, and it is worth getting exactly right because plenty of blogs have mangled it.

In Ahrefs' original July 2025 study of 1.9 million citations from 1 million AI Overviews, 76.1% of cited pages ranked in the top 10 for the same query, with a median ranking position of 3. So back then, AI Overviews mostly cited the same pages already winning the SERP. Top of the list, top of the answer box. Tidy.

Then Ahrefs ran it again at far bigger scale. Their updated analysis of 863,000 keyword SERPs and 4 million AI Overview URLs found that only 38% of cited pages now also rank in the top 10. The rest split almost evenly: 31.2% rank somewhere in positions 11 to 100, and 31% do not appear in the top 100 at all. Search Engine Journal's write-up of the same data lands on the identical figures.

Here is the bit the scary headlines skip. Ahrefs is openly cautious about treating this as a clean before-and-after. Their citation detection improved between the two studies, so part of the gap is simply that they now catch citations they used to miss. The two datasets are not perfectly comparable, and Ahrefs says so itself.

Our take: the trend is real even if the exact size of the drop is fuzzy. Google is clearly pulling sources from a much wider pool than the top 10. But anyone telling you top rankings "lost half their value" with surgical precision is reading more into the data than the data supports. Treat 76 to 38 as a direction, not a law of physics.

Why is Google citing pages that don't even rank?

Two reasons, and the first one is the interesting one.

Google uses what it calls query fan-out. Instead of answering your exact query, the system quietly breaks it into a cluster of related sub-queries, runs all of them, and then builds the answer from pages that performed well across that wider set. So a page that ranks poorly for your literal search term can still get cited because it nailed one of the hidden sub-questions. You are no longer competing for one keyword. You are competing for a fan of them you never see.

The second reason is format. In the updated Ahrefs data, YouTube is the single most cited domain in AI Overviews, growing 34% over six months, and it shows up heavily in citations that do not rank in normal organic results at all. Video is winning citation slots that text pages never had a shot at. We have started telling clients with strong how-to content that a transcript-rich YouTube version is no longer a nice-to-have.

This is exactly why we keep harping on entities and topical coverage rather than single-keyword obsession. If you want the mechanics, our breakdown of knowledge graphs and entity optimisation for AI search explains how Google connects your content to the concepts behind a query, which is what fan-out is feeding on.

Does ranking still matter? Yes, just differently

Let's not throw the baby out. A 38% top-10 overlap is still the largest single bucket. Top rankings are the most likely place a citation comes from, they are just no longer the only place. If your page can't rank at all, fan-out is not going to rescue you from total obscurity.

What changed is the safety margin. In 2024 you could treat "rank number one" as basically synonymous with "win the AI answer". Now it is more like a strong head start in a race with several other entrances. You still want to be in the top 10. You just can't stop there and assume the citation follows automatically.

We see this play out constantly. A client ranks third, looks healthy in the rank tracker, and yet a competitor sitting at position twelve keeps appearing in the Overview because their page answers the specific sub-question Google decomposed the query into. The ranking report says you are winning. The actual search result says otherwise. If that gap sounds familiar, it is the same problem we unpack in our piece on keyword cannibalisation and AI Overviews.

The traffic side: getting cited is worth less than it used to be

Before you rebuild your entire strategy around citations, a reality check on what a citation is even worth.

The Pew Research Center study published in July 2025 tracked the real browsing behaviour of 900 US adults across 68,879 Google searches, 12,593 of which produced an AI summary. The findings are sobering:

• When an AI summary appeared, users clicked a traditional search result on just 8% of visits.
• When there was no AI summary, they clicked a result on 15% of visits, nearly double.
• And the links inside the AI summary itself? Clicked on just 1% of visits to pages with a summary.

Read that last one twice. Even when you win the citation, only about one in a hundred people click it. So the citation is mostly a branding and trust play, not a traffic firehose. We tell clients this plainly: chasing citations purely for clicks is a losing trade. Chasing them so your brand is the name the AI repeats, that is the actual prize.

Google has disputed how Pew framed the click decline, which is worth knowing, but the direction matches what we see in every Search Console property we manage. Impressions hold up. Clicks on informational queries are getting quietly throttled.

Where AI Overviews actually show up

You can't win a citation in a result that never appears, so it helps to know where these things trigger.

Semrush's study of more than 10 million keywords across 2025 found that prevalence is volatile. AI Overviews fired on a small share of queries in January, spiked through summer, then settled lower again later in the year, so any single "X% of searches" stat is a snapshot, not a constant. Different trackers report wildly different numbers depending on their keyword sets, which is why you see everything from roughly 20% to nearly 50% quoted.

The pattern that holds steady is intent. Semrush found AI Overviews lean heavily toward informational queries, and they appear most on moderate-difficulty keywords rather than the bloodiest commercial terms. We have watched the same thing: our clients' "how", "what", and "why" pages get summarised constantly, while their "buy", "near me", and "pricing" pages mostly keep their normal results intact.

Our practical read: protect your transactional and bottom-of-funnel pages as normal SEO, because the AI box rarely touches them. Treat your informational content as the battleground for citations, because that is where Google is summarising hardest.

What we'd actually change: a working playbook

Enough diagnosis. Here is the approach we run on real campaigns when a client wants more AI Overview citations. None of it is magic. Most of it is just good SEO with the dial turned toward clarity.

1. Audit the gap, not the ranking. Pull your top informational keywords, run each search, and note who Google actually cites versus who ranks. The pages where you rank well but never get cited are your highest-value targets.
2. Answer the question in the opening lines. Lead a section with a direct, self-contained answer, then expand. Google's models lift clean, quotable statements far more easily than answers buried under three paragraphs of throat-clearing.
3. Map the fan-out. For each target query, brainstorm the obvious sub-questions a person would also ask, and make sure your page genuinely covers them. You are feeding the cluster, not one keyword.
4. Build entity coverage. Mention the related people, products, and concepts a topic touches, so Google connects your page to the wider subject. This is where our AI search visibility work spends most of its time.
5. Add a video where it fits. Given YouTube's citation dominance, a short, well-transcribed video version of your best guides can win slots your text page never could.
6. Track citations like a metric, not a vibe. Until tooling matures, this is partly manual. Document which queries cite you, which don't, and what the cited pages have in common.

If this sounds like a lot to keep on top of alongside everything else, that is fair, and it is roughly the reason teams hand it to us. Our broader SEO programme bakes citation tracking into the same reporting as rankings and traffic, so you are not staring at three disconnected dashboards trying to guess which one is telling the truth.

The thing the "GEO gurus" keep getting wrong

There is a cottage industry right now selling "answer engine optimisation" as if it were a brand-new discipline with secret levers. We are sceptical, and so, helpfully, is Google.

In its own guide to optimising for generative AI features, Google states plainly that "the best practices for SEO continue to be relevant because our generative AI features on Google Search are rooted in our core Search ranking and quality systems." It goes further: there is no special schema.org markup you need to add, no requirement to chop your content into tiny chunks, and no need to create llms.txt files or other "AI text files" to appear in AI search.

Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add.

So when someone tells you they have a proprietary GEO formula that bypasses normal SEO, hold onto your wallet. The honest version is less sexy and more reliable: write genuinely useful, expert-led content, structure it so the key answers are easy to lift, cover the topic properly, and earn the authority that makes Google trust you as a source. We have a longer, opinionated take on this in our guide to getting your brand into AI answers, and on why so many teams fumble it in GEO best practices and why brands fail.

That said, we don't think you should ignore the new behaviour either. Google saying "it's still SEO" is true, but it quietly raises the bar. When the model can pull from anywhere in the top 100, mediocre-but-well-linked pages have fewer places to hide. Clarity and genuine usefulness matter more now, not less.

So, did top rankings really lose 50% of AI citations?

Roughly, directionally, yes, the overlap between top-10 rankings and AI Overview citations did fall by about half between mid-2025 and early 2026. But the clickbait framing oversells it. Part of that drop is better measurement, top rankings are still the single biggest source of citations, and a citation is worth far less in clicks than it sounds, given that only about 1% of people click the links inside an AI summary.

The useful conclusion is calmer than the headline. Keep doing the SEO that gets you ranking, because that is still your strongest path into the answer box. Then add the citation layer on top: answer questions cleanly, cover the fan-out, build entity depth, and consider video. Stop treating "we rank number one" as the finish line, because Google stopped treating it that way first.

If you want a second pair of eyes on where your content is ranking but not getting cited, that is the kind of audit our team genuinely enjoys. Tell us the keywords that matter to you and we will show you the gap, and what we would do about it.