How to become the source an AI quotes.

There is no single AI search to optimise for

The first thing to absorb is that the systems disagree with each other. Analysis of citation patterns found that only 11% of the domains cited by ChatGPT overlap with those cited by Perplexity.

Their sourcing habits diverge accordingly. Gemini leans on brand-owned websites, which supply around 52% of its citations. ChatGPT leans on internet consensus, with close to half its citations drawn from third-party sites and directories. Perplexity weights industry expertise and customer reviews more heavily than either.

The practical consequence is that a company can be highly visible in one system and effectively absent from another, and that a single 'AI visibility score' from a monitoring tool conceals more than it reveals. The systems have to be checked separately because they are answering from different material.

Earned media does most of the work

Across the aggregated research, 84% of AI citations trace back to earned media rather than owned content, paid placement or optimised landing pages. Earned placements outperformed owned content for citation rates by a wide margin.

This is an uncomfortable finding for the search industry, because earned coverage cannot be packaged as a monthly retainer deliverable. It is produced by doing something worth writing about, and by being useful to the publications, directories, forums and communities that cover a sector.

It also revives a discipline that digital marketing had quietly subcontracted to public relations, and argues that the two should stop being budgeted as separate line items with separate reporting.

The corollary matters as much: Google's guidance warns explicitly that pursuing manufactured mentions is unproductive, because the systems weigh quality above volume. Paid link networks and syndicated press releases are not a shortcut here, and the research on citation quality suggests they are increasingly discounted.

Structure helps at the margin

Content characteristics do matter, though less than the tooling market implies. Selection favours clear heading structure, a direct answer to the question posed, and claims that carry their evidence within the text rather than gesturing at it.

One structural detail is worth acting on. Roughly 44% of citations are drawn from the first 30% of a page's content. Burying the answer beneath several paragraphs of context measurably reduces the chance of being quoted.

Google's guidance cautions against taking this too far — writing specifically for machines, or fragmenting pages for them, is explicitly discouraged. The two positions reconcile easily. Answer the question early because it serves the reader who wanted the answer, and accept the citation benefit as a by-product rather than the purpose.

The variable formatting cannot fix

Underneath the tactics sits a finding no amount of structure overcomes. Global household names appear in around 73% of relevant AI answers. Established mid-market brands appear in 44%. Niche brands appear in 11%.

Fame is an input to machine visibility. Language models assemble answers from patterns across an enormous corpus, and a company described consistently in thousands of independent places is a strong pattern where a company described in forty is not.

That does not make tactical work pointless, but it does set realistic expectations. A smaller company's achievable aim is to be cited on the specific topics where it holds genuine, documented expertise — not on the broad category terms its largest competitor has owned for a decade.

A programme that fits a small budget

The research supports a narrow approach, because AI visibility behaves as a topic-level game rather than a single brand score.

Choose two or three questions the company can answer better than anyone else in its market, ideally ones where it holds data, cases or operational experience that competitors lack. Publish the evidence properly — numbers, method, what did not work — rather than a summary of received wisdom.

Then work on corroboration. Get that expertise quoted, referenced and argued with by other people writing about the same questions: trade publications, podcasts, industry associations, community answers, partner sites. This is the 84% at work, and it is slower than publishing.

Audit quarterly across at least ChatGPT, Gemini and Perplexity, recording not just whether the company is named but whether the description is accurate. An incorrect description is a more urgent problem than an absent one, and it is fixed by correcting the underlying sources rather than the website.