BlogPublished 5 min read

AI citations explained: how assistants choose their sources

What AI citations are, how assistants choose which sources to cite, why cited is not the same as ranked, and how to become citable.

By Kamil Kępiński, founder of EntityRise

AI citations are the sources an assistant references when composing a live answer - the linked footnotes in Perplexity, the source chips in ChatGPT search, the references in Gemini. Citations decide who can enter a live AI answer at all: if no cited source mentions your company, the answer cannot recommend you, no matter how good your website is. That makes citations the supply chain of AI visibility - and like any supply chain, it can be mapped, measured and improved.

How engines pick what to cite

The selection happens in two steps. First retrieval: the engine runs searches and fetches a small set of pages it considers relevant and trustworthy for the question - often comparison articles, review platforms, documentation, industry databases and news. Then synthesis: the model composes an answer grounded in those pages and attributes claims to them. Two properties follow. The cited set is small - a handful of sources carries the whole answer. And it is question-specific - the sources cited for "best X for enterprises" differ from "X pricing", so citation presence must be checked per question, not per brand.

Cited is not the same as ranked

Engines do not simply cite the top of Google. Measured answers regularly cite pages that rank modestly while skipping pages that rank high, because retrieval optimises for answerability: pages that directly, verifiably answer the question in machine-readable form. Dense comparison tables on mid-authority blogs often beat a high-authority homepage that answers nothing. This is good news for challengers: the citation game rewards content quality and verifiability more than domain size.

The $49 audit records which sources AI engines cite for your buyers' questions, so your team knows where to earn presence →

How to become citable

Three moves, in order. Map the citations: run your buyers' questions across providers repeatedly and record which domains and pages get cited - that list is your real competitive terrain. Earn presence in third-party sources: accurate listings and reviews on the platforms that appear in your map, and coverage in the articles engines keep returning to. Publish answerable pages yourself: direct, structured, verifiable answers to the questions - because your own site can be a cited source, and for methodology, pricing and product facts it should be the cited source.

Frequently asked questions

Do all assistants cite the same sources?

No - overlap exists but each engine has its own retrieval habits, which is why citation measurement runs across ChatGPT, Claude, Gemini, Grok and Perplexity rather than one of them.

Can I pay to be cited?

Not directly in the engines. You can earn presence in cited third-party sources; some of those platforms have commercial listings, which is a business decision - but fabricated placements and fake reviews get filtered and are radioactive for trust.

How stable are citations?

They drift as models, indexes and sources change - a reason to remeasure the same questions on a schedule instead of trusting a one-time snapshot.

Kamil Kępiński · Founder of EntityRise and Smophy Labs Inc.

Kamil Kępiński is the founder of EntityRise and Smophy Labs Inc. He writes about how models choose the companies they recommend and how to measure a company's visibility in AI answers.

Every live AI answer is built from a handful of sources.

EntityRise finds yours and shows where your presence is missing, so your own team or agency can earn it.

Run the audit · $49 · For agencies: packs from $28 per audit with white-label reports