The Most-Mentioned Wins: AI Counts Citations, Not Rankings

Ranking number one stopped guaranteeing the AI answer. It did not stop mattering.

A third of AI Overview citations still come from pages in Google’s top ten, and only nine other pages compete for that bucket. Ahrefs, studying 863,000 keywords and 4 million AI Overview URLs, found 38% of cited pages ranked in the top ten, down from 76% in mid-2025. The rest split almost evenly: 31.2% from positions 11 to 100, and 31.0% from beyond rank 100. BrightEdge, using a different method, put the top-ten overlap closer to 17%.

The field flattened. It did not level. So the panic going around, that rank is dead, reads the data backwards. Number one still holds the best odds in the room. What changed is that you can no longer keep the spot. So stop optimizing for a rank you cannot hold, and start being the answer the model keeps finding.

Did ranking number one stop mattering? No, the field flattened

Number one still has the best odds. It just stopped being a guarantee.

Run the math on the three buckets. A third of citations come out of the top ten, fought over by nine other pages. The next third comes from positions 11 to 100, fought over by ninety. The last third comes from everything past rank 100, fought over by everyone alive. Same prize, wildly different field sizes. The top ten is still the smallest room with the best seat.

So you still want to rank. You just make peace with ranking seven and getting cited anyway. The odds favor the top. They no longer promise it.

Why you can no longer hold a position

You cannot hold a spot because query fan-out reshuffles the answer on its own clock. You are in the overview Monday and gone by Thursday, and nothing about your page changed.

Fan-out is the mechanism. The model breaks one question into a dozen sub-queries, runs each, and pulls from deeper in the results for every one. That is why YouTube is now the single most-cited domain in AI Overviews, accounting for 5.6% of all cited URLs in the Ahrefs dataset. Fan-out reaches past the top ten into video, forums, and pages that never ranked for the original query.

You cannot defend a flag on ground that moves. Chasing the position is chasing a target the system rerolls without telling you. Build the thing that survives the reroll instead.

What the model actually does: it counts mentions

An LLM does not crown a winner the way a ranking does. It aggregates. It looks for trend lines, repeat mentions, the same figure showing up across sources, and it presents what it sees most.

Mentioned twice loses to mentioned seven times. That is the whole mechanism in one sentence. One mention on your own site is a single data point. The same numbers said again on YouTube, again in a Reddit thread, again on TikTok, is a pattern the model reads as consensus. Rank is a moment. Mention is a pattern.

This is Share of Brand Voice wearing a GEO coat. I have measured share of voice for years as the count of times your brand and your claims surface across the places buyers look. AI search moved that count inside the answer box.

Show up in more places with the same real numbers

Repeat the same true figure across modalities. The model counts you, so give it more to count.

The content that wins is valuable, evergreen, and loaded with real data. Unique data if you have it. The figures only you can serve are the ones that cannot be averaged out, because they exist nowhere else for the model to pull. I grew a TikTok account from 400 followers to over 110,000, posting twice a day, with the breakout landing on day 34. That number lives in my business and nowhere else. A model cannot fabricate it, and it cannot blend it into the crowd.

This is not spam. Spam is volume with no substance. This is one real fact, stated in more rooms. Google’s March 2026 spam update flattened the people who confused the two. The model is not counting words. It is counting consistent mentions of a claim only you can make.

Turn the blog into a YouTube video

Read the post to a camera. It is the cheapest second modality you have, and YouTube is the most-cited domain in the overviews.

The blog already exists. The data is already written. Saying it on video is a second mention of the same numbers, in the format fan-out reaches for first. The page and the video reinforce each other inside the answer. One pass, two citations. One structural detail matters more than your subscriber count: a full description and real chapters correlate with repeat citation far more than views or likes. Write the description like it is the post.

Where this leaves you

Stop asking how to hold your rank. Ask how many places say your number.

Count your own footprint. One page, or a page plus a video plus a thread plus a podcast, all carrying the same real figure. That count is your share of the answer. It ties straight back to the Eight Dominoes: you build the memory before the buying window opens, in every place the buyer might look. Position is a moment you cannot keep. Presence is a pattern the model keeps finding. The question quietly changed from did your content rank to does your content answer.