85% of AI citations don't come from your website. Here's what to do about it

There's a number that should sober up anyone who still thinks AI search is just "another Google." 85% of B2B brand mentions in AI answers come not from the brands' own websites, but from third-party sources. You can polish your site until it shines, and the model will still quote some random article about you — without linking back to you at all. That's a fact. You have to accept it and start working differently.

Search Engine Land broke down the path to citability into three phases. Let me walk you through what this actually means in practice.

Phase 1: make yourself readable to the machine

An LLM doesn't read your site the way you do. It needs logical navigation, machine-readable markup, a sane H1/H2 hierarchy — no chaos where a third-level heading sits above the first one just because it "looks nicer." Sounds like basic technical SEO, and it is. The difference is that you used to optimize for a crawler that indexes the whole page. Now you're optimizing for a model that pulls out meaningful chunks from it.

Here's the practical takeaway: your top content should include explicit key-takeaway blocks — short, standalone summaries of the core point. Not because it's trendy, but because it's easier for a model to quote a ready-made thought than to piece one together from three paragraphs.

For product pages, this works just as bluntly: H1 with the product name, H2 with the category, bullet points with specs, a short description, a comparison with alternatives. Boring structure — the kind both humans and models love. There's no room here for creativity for creativity's sake. Nothing to be done about that.

Phase 2: stop chewing on what everyone else already said

Phase two is about content with your own original angle. Your own data, your own research, your own stats, your own position. LLMs are trained on massive volumes of text, and if you're writing the same thing that's already been written fifty times before you, there's simply nothing to quote you for — you blend into the noise.

This is where the hub-and-spoke approach comes in: you build one strong piece with a genuinely original angle, then spread it across formats — infographics, video, short posts, a newsletter. This isn't about "squeezing content to the max." It's about getting the same idea in front of the model from multiple angles and in multiple forms. The more places your original thesis lives, the higher the chance it gets picked up as the source.

Phase 3: authority isn't built at home alone

And this is exactly where that 85% figure clicks into place. Owned media is your site and your social channels. Earned media is mentions of you on someone else's platform. A model forms its answer based on the entire mass of text on the web — not just what you've published yourself. If third-party sources stay silent about you, you barely exist for an LLM, no matter how flawless your own site is.

So the job isn't just to publish on your own channels — it's to get other people to write about you and link to you: industry outlets, experts, partners. And here's the catch: your messaging has to be consistent everywhere. If you position yourself one way on your site and say something different in an interview with a third-party outlet, the model picks up on the contradiction and gets confused about which version to trust.

It's also worth tracking specifically where you're already being cited. This isn't a one-time check — it's ongoing monitoring: which sources actually credit you, and which ones use your information without linking back. The second case is a signal that you need to work on that relationship directly, possibly with the platform itself.

What to do right now

None of this works as a one-off campaign. It's a long game: first you fix technical accessibility, then you start producing genuinely original content, and only after that — gradually — you build up third-party citations. Skip the first step, and the second and third won't work — models simply won't be able to parse your page. Skip the second, and there's nothing worth citing. Everything in order, no skipping steps.

If you don't have the time to figure this out yourself — this is exactly the kind of work someone who's already tuned site visibility for both Google and AI search should be doing, and that's where a conversation about a technical and content strategy for your project can start.