Last month I dug into a client’s analytics and something smelled fishy. They were doing great in the top five for 40+ keywords in their niche. Organic traffic from traditional search was solid. But when I ran the same keywords in Google’s AI Mode, their site showed up in just two of the results. Just 2 out of 40. That was weird.

So I spent the next week trying to figure out what was going on. It turned out I wasn’t the only one scratching my head. This is a broader shift in how Google selects sources for their AI responses. Most SEO strategies just aren’t ready for it.

The Numbers That Completely Changed My Mind on Rankings

Ahrefs published a new study back in March 2026 and it stopped me dead in my tracks. They took a look at 863,000 keywords and 4 million AI Overview URLs and found that only 38% of pages cited in AI Overviews actually rank in the top 10 for that query. But then I looked back at a July 2025 study and saw that in just 7 months, that overlap between traditional rankings and AI citations dropped from 76% to just 38%.

Think about it – that’s a roughly 50% drop in 7 months. The remaining 62% of citations split roughly 50/50 between pages that rank 11-100 and pages that don’t even show up in the top 100. Another study by BrightEdge put the top-10 overlap even lower, at around 17%. So yeah, the exact numbers might depend a bit on how you crunch your numbers – but one thing is for sure, ranking well for a keyword just doesn’t mean Google’s AI will cite you anymore.

When I first saw these numbers I have to admit it – I was in denial. We’ve built a bunch of client strategies around the idea that strong organic rankings feed AI overview visibility – but it turns out that just isn’t true anymore.

Why Your Rankings Don’t Translate to AI Citations Anymore (And Why It Matters)

The short answer is something called query fan-out.. Google’s own developer documentation says this is how it works. When someone searches in AI Mode, Google isn’t just looking at the results for that exact query. No, it breaks the search down into 10 to 16 related sub-queries – and then pulls results for those sub-queries separately.

So if someone searches for “best CRM for startups”, for example, Google’s AI is at the same time pulling in results for “CRM pricing for small teams”, “startup CRM integration requirements”, and “CRM reviews from founders”. But here’s the thing – those sub-queries have zero traditional search volume because humans just wouldn’t type them. Yet they’re what AI Mode actually retrieves.

This is why a page that ranks position 40 for a related sub-query might end up getting cited in the AI response, while your page that ranks position 2 for the main keyword gets totally ignored. And it all changed when Google rolled out Gemini 3 as the global default for AI Overviews back in January 2026 – Ahrefs noted that the new model replaced about 42% of previously cited domains. The game just changed – and it changed fast.

The CTR Problem Is Worse Than You Think

Even when AI Overviews do cite your page, the traffic math has changed. A Seer Interactive study found that organic click-through rates for queries with AI Overviews dropped from 1.76% to just 0.61%. That’s a 65% drop in just a year. Ahrefs own data shows that even if you’re the number-one ranking page, the AI answer still reduces your CTR by 58%.

So now you’ve got a double problem on your hands. Your top-ranking pages are less likely to get cited by AI, and even when they do show up in the results, fewer people are clicking through because the AI answer already gave them what they needed.

I’m not trying to freak you out – I’m just trying to make you understand what’s happening here. I’ve watched clients lose real traffic while their rankings stayed exactly the same – and that’s a weird and confusing situation if you don’t know what’s going on underneath.

What We’re Doing Differently Now

We’ve had to make some changes at Shortlist over the past few months – and none of this is just theory. This is what we’re actually doing for clients.

First off, we’ve stopped optimizing for single keywords in isolation. The fan-out model means your content needs to cover not just the primary query, but all the related questions Google’s AI is actually retrieving. We use People Also Ask, Reddit threads – and tools that simulate fan-out queries to map out the whole set of sub-topics a piece of content needs to cover. If your article about content marketing only talks about content marketing, it’s probably not touching on the sub-queries about distribution, measurement, repurposing, and team workflows that Google’s AI is actually after.Second – let’s get to the good stuff right up front. Research from multiple sources shows that a whopping 44% of AI citations snag info from the first 30% of a page’s content – that’s the top third. So we’ve overhauled how we write our articles: we put the most specific, rock-solid, citable claims upfront in the opening sections. Truth is, if your key insight is buried on page 15, a machine’s less likely to find it – and use it.

Third, we’re giving YouTube a serious shot. I went in not knowing what to expect, but it turns out YouTube is now the number one most-cited domain in AI Overviews, accounting for 18.2% of all citations from pages outside the top 100. Ahrefs found out that getting your brand name into YouTube video titles, transcripts and descriptions is the strongest signal for making your AI Overview more visible . For any clients who can knock out some video content, this has become a real priority.

Fourth, structured data markup – we’re taking this super seriously now. Research from SE Ranking found that roughly 65% of pages cited by AI mode all have one thing in common : they have structured data. Think about it – FAQPage, HowTo, Article and Organization schema are basically Google’s way of saying ‘Hey, I can extract this bit of info and make it real for the AI’. This is old advice, I know, but this year its even more important.

Fifth, we’re working on building a citation network – this is our term for it. Brands getting talked about on review sites, Reddit, Quora and other third-party editorial sites get cited by AI a whole lot more. One study found that brands with Trustpilot and G2 presence get 3 times as many ChatGPT citations. Getting a strong presence on Reddit and Quora sees AI citation rates go up by 4 times . Its not just what’s on your own site that matters – AIs have a good look around to see what other credible sources are saying about you

This is far from the end of SEO – it is just a different ball game

And to be clear: traditional SEO still counts. Your pages need to be properly indexed, crawled and technically sound. You still need to put up quality content and get some strong backlinks. But if your entire strategy is built around ranking for keywords and just waiting for the clicks to roll in, you’re playing a game thats rapidly shrinking.

The brands that appear in AI responses get 35% more organic clicks and 91% more paid clicks than the brands that don’t. Getting AI to cite you is not just about getting some extra AI traffic – its a whole game-changer.

I’m honest, – we’re figuring this bit out as we go. AI overview visibility is constantly shifting, and data is moving fast enough that what works in March might need an overhaul by June. But the core idea is dead simple: stop focusing on getting individual keyword rankings and start thinking about whether your content actually answers the whole range of questions a human or AI might have about your topic.

That’s the shift. it is not rocket science, but it does mean you have to rethink a few things that used to work. We’re in that process ourselves – still figuring it all out.