How AI citations work

What is an AI citation?

An AI citation is a mention of your business inside a generated answer, usually with a link back to a source that supports it. It is the AI-era equivalent of being the recommendation - except the recommendation arrives without a results page.

Two distinct things earn it: what the model already associates with your name, and what it can retrieve and verify while writing the answer. Work that fixes only one of the two rarely changes the outcome.

How does an assistant decide who to name?

It expands the question into searches, retrieves sources, keeps the claims that agree across independent sources, and synthesizes a shortlist of 2–3 names. Anything it cannot corroborate gets dropped rather than risked.

1. Retrieval. The buyer's phrasing becomes several machine queries - the category, the city, the qualifiers. Pages that are slow, blocked or thin never enter the candidate set.
2. Corroboration. Claims that repeat across independent sources survive; claims that appear only on your own site are treated as marketing.
3. Synthesis. The assistant writes a short answer and names the few businesses it is most confident about. Confidence, not eloquence, decides the list.

Why consensus beats claims

A business describing itself is one source. Independent directories, review platforms, press and third-party roundups are what turn a claim into a fact the model will repeat.

This makes the gap concrete and fixable: find the roundups and listicles your competitors appear in for your category and city, and note where you are absent. Every one of those absences is a source the assistant reads without ever encountering your name.

What makes a page easy to cite

A citable page gives the model one URL to point at and one unambiguous answer to quote. Five properties do most of that work.

• A self-referencing canonical. One declared address for the page, so the citation resolves somewhere real instead of splitting across duplicates.
• Structured data identifying the entity. Markup that says which business this page belongs to and how it connects to the rest of your footprint.
• The answer stated first. Under a heading that matches the question as a buyer asks it, with the answer in the opening sentences.
• Stable facts. Name, address, phone, services and locations stated identically everywhere, so nothing contradicts.
• No content behind JavaScript. Agents frequently do not execute it. If the answer only appears after hydration, assume it was never read.

The discipline that ties these together is answer engine optimization.

Why your rankings can be fine while your citations are not

Ranking rewards a page that competes well for a keyword. Citation rewards an entity the model can confirm. A site can do the first and fail the second.

The usual causes are unglamorous: an entity that reads as two different businesses, a phone number that varies by directory, service pages that bury the answer under a paragraph of positioning, and no independent source that says what you say. Rankings tolerate all of that. Assistants do not.

How to check whether AI cites you

Ask the assistants the buying question your clients actually ask - not your brand name - and write down who gets named. A brand-name query almost always finds you and tells you nothing.

Run the same question across ChatGPT, Perplexity, Claude and Google AI Overviews, because the shortlists differ. Then read the sources each one cites: those pages are the shortest path onto the list. If you would rather have that mapped for you, the Phoenix engagement starts with exactly this audit.

Common questions

Why doesn't AI mention my business?

Usually because the assistant cannot confirm you. Either your entity is unclear, your facts are stated differently across the web, or you are absent from the independent sources it retrieves - so a competitor that is easier to verify gets named instead.

How do I get cited by ChatGPT or Perplexity?

Be the clearly-identified, corroborated, easily-retrievable answer to the question your buyer actually asks: structured data that identifies the entity, the answer stated in the first sentences under a matching heading, identical facts everywhere, and presence in the independent sources that already rank for that question.

Do AI citations affect revenue?

For high-ticket businesses, yes. The assistant returns 2–3 names, so being omitted removes you from the consideration set entirely - and where one client is worth $3,000–$50,000, that is a lost deal rather than a lost click.