The Posts Keep Going Up And The Models Keep Saying Somebody Else’s Name
Your monthly publishing chart looks great. Ask ChatGPT who the players in your category are, though, and your name doesn’t come out of it.
I spent about half a year assuming volume was the lever — publish enough and something eventually gets picked up. If you’ve stalled in the same spot, you have company.
Publishing more isn’t the mistake. But line up the pages that AI assistants actually cite and the shape of the thing you’re supposed to be publishing turns out to sit somewhere else entirely.
Someone counted 25,000 of those URLs, so let’s start there.
What Kind Of Pages AI Search Actually Cites
Roundups. Across about 400 million citations analyzed by Evertune, 63% pointed at a listicle — a “best 5 project management tools” sort of page R.
A listicle (ie an article that lines candidates up and compares them) isn’t a format anyone would call sophisticated.
It’s the format that wins anyway.
Split by model, the share lands between 40% and 65%, so no assistant dips under four in ten. Copilot is the low end, Gemini the high end; the six surfaces measured were ChatGPT, Copilot, Gemini, Google AI Mode, Google AI Overview and Perplexity, over March and April.
Evertune sells AI-visibility measurement. I’ll come back to what that does to how you should read this.
Ranked Lists Are Where The Open Slots Are
Cut the listicles open and ranked lists — pages that number their picks 1, 2, 3 — take 71% to 86% of them. Unranked parallel lists come second.
Rankings published by institutions and industry bodies? 1.4% to 4.7%.
If your plan for getting cited runs through an industry award or an official league table, that’s a slot with a queue and an entry bar, and it accounts for under a twentieth of the cited roundups. Almost all of the rest are comparison pieces that publishers, review sites and individuals wrote on their own initiative.
Which means the doors are a lot wider than the awards-and-accolades route makes them look. Pitching an editor who runs a “best tools for X” page is a Tuesday afternoon; getting onto an industry ranking is a fiscal year.
The Cited Pages Weren’t Written For Machines
Word counts on the pages that got cited sat between about 1,000 and 2,000 words, at an average sentence length of 18 words.
Model by model: Copilot’s cited pages averaged 964 words across 24 paragraphs, Gemini’s averaged 1,977 words across 53 paragraphs.
Those aren’t targets. They’re the average of what happened to get cited, and writing to hit 1,977 words won’t summon Gemini.
What the spread does tell you is that ordinary comparison articles, written for humans to read, are being picked up as-is. Since none of the numbers point at a machine-readable format, the “special AI formatting” you keep hearing about doesn’t show up in the data at all.
I’d braced for the opposite. I turned up in formal wear and the whole room was in jeans.
How Far To Lean On A Vendor’s Own Numbers
Evertune sells an AI-visibility platform. “You should be in the listicles” is a conclusion that happens to point straight at their product, and you should read it with that in view.
None of it is peer-reviewed either, so if you’re pulling these figures into an internal deck, name the party that ran the study on the same slide.
Second, this is a breakdown of URLs that got cited — not evidence that making a listicle causes citation. Commercial evaluation queries (“best X”, “X vs Y”) are already thick with listicles before any model touches them, and the study can’t rule out that the population it sampled was skewed that way to begin with.
Third, don’t line up 63% and 50% as though one became the other. The 63% comes from 25,000 URLs across six models in March and April; the 50% comes from a separate study of 40,000-plus URLs across about 50 categories over 60 days in April 2026 R. Different denominators, different windows. Put them on one line and you’ve invented a trend.
Those three checks work on the next vendor study you read too, which is most of the value of writing them down.
Count The Roundups You’re In Before You Count Your Posts
Here’s the small version of this. Search “best [your category]” and “[your category] comparison”, open the ten roundups that come up, and count how many of them name you. The ones that don’t are your list of things to do next week, in priority order, for free.
The second thing is where you check the result. AI assistants don’t pull from the same pile: asked the same question, Perplexity drew on 838 unique URLs while Google AI Mode drew on 6,687 (Evertune, 2026). Judging your progress from the Google surface alone tells you about one of those piles.
Maybe I’m wrong about how far the listicle finding generalizes — one vendor, two months, one snapshot of a fast-moving thing. What I’d still bet on is the counting exercise, because it costs an hour and tells you where you actually stand.
Once you’re on a list, where your name sits inside the page is its own question, and I went at that one in the shape of a mention that models pick up. Whether being cited means being used in the answer is a different question again, which is selection versus absorption.
Today’s is the step before both: which shelves you’re on in the first place. Go count the shelves.
