84% of AI citations are not your client's website

The most common reaction to a first citation report is “we need to publish more”. It is almost always the wrong conclusion, and the number that should stop it is right at the top of the table.

Where citations actually come from

Roughly 84% of AI citations are earned or third-party media — review platforms, directories, comparison sites, trade press, forums, and the long tail of pages written *about* a brand rather than *by* it. The brand's own domain is a minority contributor in most categories, and in some it barely registers.

Underneath that sits a related finding: off-site brand mentions correlate with AI visibility at around 0.66, while backlinks manage about 0.22. The engines are reading what the web says about you, not what you say about yourself.

How to read the table

1. Sort by frequency, not by ownership. The domains cited most often are the ones shaping the answer, whoever owns them.
2. Find the third-party sources you are absent from. A comparison site cited on eight of your prompts that does not list your client is a concrete, addressable gap with a name and a URL.
3. Check the ones you do own. Which of your pages get cited tells you what the engines think you are authoritative about, which is frequently not what the client thinks.
4. Ignore the singletons. Source churn is high — day-to-day overlap on identical prompts runs 34–42%. A domain cited once is noise. A domain cited across a fortnight is a source.

What this means for the retainer

The work that moves AI citation share looks much more like digital PR, listings management and category-page placement than like publishing cadence. That is a harder sell than a content calendar, and it is what the evidence supports.