Citations Are Downstream

Tim Soulo shared the key findings of Ahrefs' large research looking at AI citations here.

Most people focused on the rankings.

We got stuck on something else.

About 28% of the pages ChatGPT cites most frequently have essentially no Google visibility. Not poor visibility. Not page five. Effectively none.

That's a strange result if you believe AI systems are simply replaying Google. It suggests there's another layer operating underneath the one most of us can see.

Then there was another finding.

ChatGPT retrieves more pages than it ultimately cites. It pulls information from a much larger pool of sources than the handful that end up attached to the answer. Some sources get cited. Some sources get used. Some sources get used repeatedly and never receive credit at all.

That's a very different picture from the one most AI visibility tools present.

Most of the industry is focused on the visible layer. The citation. The mention. The link. The screenshot. The dashboard notification. The thing you can point at and count.

Nothing wrong with that. Citations matter. If your brand is showing up in AI answers, that's useful information.

The problem is that citations are an outcome. They're not the process.

Before an AI system can cite something, it has to find it. Then it has to access it. Then it has to make sense of it. Then it has to decide whether it trusts it enough to use. Only after all of that does a citation become possible.

Most of that chain is invisible.

And if the Ahrefs data tells us anything, it's that the last step is far noisier than many people assume. Citation sets change constantly. Sources appear and disappear. One answer cites five websites today and five different websites next week while reaching essentially the same conclusion.

The attribution moves around. The underlying information often doesn't.

That's why we've spent so much time thinking about participation instead of citations. Not because citations don't matter. Because they're downstream.

At aater.ai, we're interested in the conditions that exist before a citation ever happens. Can an AI system discover your content? Can it access it without friction? Can it understand what it's looking at? Can it connect that content to a real entity it recognizes? Can it use what it finds without doing unnecessary work?

Those questions don't tell you whether you'll get cited tomorrow. Nobody can honestly promise that.

But they do tell you something important: how difficult are you for AI systems to use?

That's what Participation State measures. Reachability. Legibility. Authority. Three gates. Each one shapes how much effort an AI system has to expend before it can use what it finds. The more friction you introduce, the harder that becomes. The less friction you introduce, the easier it becomes.

We're not sure we have all of this right yet. But this is where we've landed — and the Ahrefs data made us more confident we're at least asking the right question.

The web spent the last twenty years optimizing for rankings. The next decade may look different.

The question might not be whether an AI system mentioned you. The more interesting question is whether it could use you in the first place.

Everything else sits downstream of that.