Getting Cited by AI: How Generative Engine Optimisation Works

A decade of SEO trained businesses to chase one thing: a position on a page of ten blue links. That page is no longer where most answers get delivered. ChatGPT, Gemini, Copilot and Google's AI Overviews now compose an answer first and cite a handful of sources beneath it — and for a growing share of searches, the user never scrolls past that answer at all.

This is the shift generative engine optimisation addresses. The objective is no longer to rank above a competitor. It is to be the source the model reaches for when it assembles the response. Ranking is a position; citation is a decision the model makes about which pages it trusts enough to quote.

Why Ranking and Citation Are Not the Same Problem

Answer engines do not read the web the way a classic crawler ranks it. They retrieve passages, weigh them against each other, and stitch the surviving fragments into prose. Several things follow from that mechanic, and each one breaks a familiar SEO habit:

• Passages get cited, not pages. A model quotes the paragraph that answers the question cleanly. A page that buries its answer under 600 words of preamble supplies nothing worth extracting, regardless of how well it ranks.
• Specificity beats volume. Concrete figures, named processes, dates and prices survive retrieval because they are verifiable. Generic marketing copy is discarded — it adds no information the model does not already hold.
• Consistency across the web is the trust signal. When your services, location, pricing and claims match across your site, your Google Business Profile, directories and third-party mentions, the model reads a coherent entity. When they contradict, it reaches for a safer source instead.
• Rendering still decides everything. Many AI crawlers execute little or no JavaScript. Content that only exists after hydration is, to them, content that does not exist.

Building Pages Models Can Actually Quote

Practical generative engine optimisation is less exotic than the terminology suggests. It is disciplined information architecture applied with retrieval in mind.

Lead with the answer. Put the direct response in the first paragraph under each heading and use the elaboration that follows to earn the reader's confidence, rather than making them work for the point. Write headings as the questions people actually ask, then answer them in self-contained blocks that make sense lifted out of context — because that is exactly how they will be used. Publish the specifics competitors hedge on: real turnaround times, real coverage areas, real constraints. And make the machine-readable layer agree with the visible one through accurate Organization, Service, FAQPage and Article schema, so structured data confirms the copy instead of describing a different business.

None of this works on a slow, script-heavy site. Clean semantic HTML, server-rendered content and fast, stable delivery remain the precondition — the same architectural discipline that entity authority depends on. An answer engine that cannot cheaply parse your page will simply cite one it can.

Measuring Something That Does Not Report Itself

AI referrals arrive with sparse, inconsistent attribution, so the old dashboard will understate them. Watch the signals that do surface: referral traffic from assistant domains, branded search volume rising without a matching campaign, and enquiries that arrive already informed — prospects quoting your process or pricing back to you before the first call. Test directly, too. Ask the assistants the questions your customers ask, note who gets cited, and treat every competitor citation as a specific, addressable content gap rather than a mystery.

The Underlying Bet

Generative engine optimisation rewards the same thing good publishing always has: clear, specific, verifiable information delivered on infrastructure that does not get in the way. The difference is the reader. A site written to impress a human skimmer and a site written to be quoted by a machine now have to be the same site — and the businesses building that way are the ones showing up inside the answer instead of underneath it.