Your Top 10 Ranking Doesn't Matter the Way You Think It Does
Two Kinds of AI Risk
Are you assuming AI is only a threat to your marketing strategy? Microsoft just shut down infrastructure for EvilTokens, an AI-assisted platform that had compromised 12,000 Microsoft accounts across 10,000 organizations before it got taken offline on September 22, 2026.
That story matters for two reasons. First, it is a reminder that the same generative capabilities publishers use to write faster and rank better are also being used to break into accounts at scale. AI cuts in both directions, and Microsoft's takedown is proof that the offensive side is not theoretical.
But there is a quieter risk happening in AI right now, and it has nothing to do with security. It is happening inside search itself, and most publishers have not caught up to it.
For twenty years, marketers built entire content strategies around one number: where does this page rank. Top 10 meant traffic. Page 2 meant you basically did not exist. That mental model is breaking down fast, and it is breaking down specifically inside the tool that is supposed to be the next evolution of search.
Google's AI Overviews do not pull citations the way organic rankings used to reward pages. The overlap between "ranks well" and "gets cited" has been collapsing all year, and the data on that collapse is not subtle.
The Data Nobody Ranking-Obsessed Expected
Ahrefs ran the numbers in March 2026, looking at 863,000 keywords and roughly 4 million AI Overview URLs. In July 2025, 76% of citations came from Google's top 10 organic results. By March 2026, that number had dropped to 37.9%. Less than half.
Conductor ran a separate analysis around the same window, looking at nearly 168 million AI Overview citations. Their finding lines up with Ahrefs: 57% of citations came from pages ranking outside the organic top 10.
Two firms, two different datasets, the same conclusion. Ranking well used to be a reasonable proxy for getting cited. That proxy is broken now.
Part of this traces back to a specific change. Google made Gemini 3 the default model behind AI Overviews on January 27, 2026. Once that happened, the average number of sources cited per answer jumped by 31.8%. More sources per query means more room for pages that never touched page one.
The mechanism behind all of this is query fan-out. When you type a question into Google, the AI Overview does not just look at what ranks for that exact phrase. It breaks your query into several sub-queries, runs each one separately, and pulls citations from whatever ranks well for those fragments, even if none of those fragments match your original search.
Why Page 11 Can Beat Page 1
you need to crack the top 10 before AI Overviews will ever notice you. The Ahrefs and Conductor numbers say otherwise. A page sitting at position 34 can get cited. A page that never breaks the top 100 for the main query can still surface through one of the fan-out sub-queries. Ranking well helps, but it is no longer a gate you have to pass through first.
The stakes for getting this wrong are not small. Cited brands inside AI Overviews see roughly 35% higher organic click-through rate. Brands that get left out are losing up to 61% of the clicks they used to count on, based on mid-2026 data. That is not a gradual erosion. That is a cliff, and which side you land on has almost nothing to do with your position on the old blue-link SERP anymore.
So the question worth asking is not "how do I rank higher." It is "what makes a page extractable regardless of rank." Those are different problems with different solutions. One is about backlinks and domain authority built over years. The other is about how clearly a page answers a specific question, right now, in a format a model can lift cleanly.
What to Publish Instead
Extractability comes down to structure, not polish. Answer the question in the first sentence of a section, before you explain your reasoning, before you set up context, before any of that. A model doing query fan-out is scanning for the sentence that resolves a sub-query cleanly. Bury that sentence under three paragraphs of throat-clearing and it never gets pulled, no matter how good the page ranks.
Name your sources inside the content itself, not just as a link at the bottom. If you're citing a study, say who ran it and when. Ahrefs, Conductor, whoever it is. Models weigh attributed claims differently than unattributed ones, and a named source gives the AI Overview something concrete to cite back to.
Put a visible freshness date on the page. Not just in the metadata, on the page itself, where a reader and a crawler both see it. Content that reads as current gets preferred over content that reads as evergreen filler from three years ago.
Format for lifting. Short paragraphs, direct sentences, one idea per sentence. A model extracting a chunk needs that chunk to make sense with the surrounding words stripped away. If your point only lands after four sentences of setup, it will not survive the extraction.