Who AI cites when people ask
Access and readiness decide whether engines can use your pages. Citation is the result on the other side: what ChatGPT, Claude, and Perplexity actually say when someone asks about your category — the questions where they link to you, and the ones where they cite a rival instead.
I keep a set of tracked questions — the category prompts your customers really ask. Each pass, for every question and every engine I ask the question with web search on, then record the answer and the source URLs the engine relied on. Rather than parse each provider's own citation metadata, I let the model self-report its answer and sources as structured output — one code path across every engine.
One probe per question per engine. Cited means a returned source's host is your domain — or a subdomain of it.
The engines are configurable — by default, the same model that powers your chat — each an agent with a web-search tool. I re-check them on a schedule, every 168 hours by default.
Every run keeps its evidence. On any tracked question you can see, per engine, whether you were cited, the exact page it cited instead, and an excerpt of what it said — so a gap shows not just that a rival won the citation, but which page won it, and why.
Your visibility score (0–100) is the share of a pass's runs that cited you. If half of this week's question×engine runs returned your site as a source, you score 50. Each pass's score is one point on the trend — the sparkline that shows whether you are winning more citations over time.
Illustrative only — the per-pass score climbing from mostly-uncited toward cited as pages improve. Your real trend lives on your domain's page.
A score tells you how often you are cited; share of voice tells you against whom. Across the latest pass, every cited host gets one slot per citation — you, when cited, plus each other host the engines cited. A host's share is its slots over all slots, ranked highest first. This is the picture of who AI names instead of you.
Illustrative shares — your slots versus the rivals the engines cited in the same pass. The real numbers live on your domain's page.
An engine can name your brand in its answer without linking you. That is a mention, not a citation — and the difference is the whole opportunity.
I surface your named rate (the percent of runs that named you) against your cited rate (the percent that cited you). When you are named far more often than you are cited, engines already trust your name but cannot find a link worth making — earn one with structured data and original data, and you convert mentions into citations.
A tracked question you are not cited for is ranked HIGH when a rival is already cited for it — a citation you could win — or it mirrors a real question from your chat (proven demand). Every other not-cited question, where the engines cited no one, is MED.
The tracked questions are not guessed: discovery generates them automatically from your pages, your competitors, and the demand visitors show on your chat — or you pin and mute them yourself. And citation rests on the stages before it: engines can only cite a page they can fetch and find ready to quote.
access → readiness → demand/discovery → citation