How brands get cited in ChatGPT, Claude, and Perplexity

Search did not die. It split. Half of the high intent research that used to happen on Google now happens inside ChatGPT, Claude, Perplexity, and Gemini. The job is no longer to rank a page on a results list. The job is to become the source the model quotes when a real person asks a real question.

We call this Generative Engine Optimisation. It is closer to PR than to classical SEO, but with a measurable, repeatable structure underneath it. After nine months of running GEO programs for B2B and D2C clients, this is the working model we use.

The shift from ranking to citing

A ranking is a position on a page. A citation is a sentence inside an answer. The economics are different. A citation gets attributed to your brand inside the response, which means the buyer reads your name before they ever land on your site. The click matters less. The mention matters more.

We do not have industry-wide numbers to sell you, and we distrust the ones being passed around without sources. What we have is a pattern seen across every client program we run: a growing share of high intent research is happening inside LLMs, and the brands that are quotable there get named before the buyer ever reaches a results page. If your brand is not quotable, you are invisible inside this surface. And unlike Google, you cannot buy your way in.

The four-part citation model

Across thousands of prompts we have tested, citations cluster around four properties. A page that has all four gets cited far more than a page that is merely well written.

• 1

  Specificity. Numbers, dates, named methods, named people. Models prefer concrete sources over vague ones.

• 2

  Structure. Clear H2s, short paragraphs, lists. Models extract from scannable text.

• 3

  Originality. A claim, a framework, or data the model has not seen on twenty other pages.

• 4

  Authority signal. A real author, a real organisation, a real publication date that is recent.

Patterns that work across engines

ChatGPT and Gemini lean on training data and tend to cite long-lived domains with strong topical depth. Perplexity and Claude lean on live retrieval and tend to cite recent, well-structured pages even from smaller domains. The implication is practical.

• For ChatGPT and Gemini, build evergreen explainers with strong internal linking around a tight topic cluster.
• For Perplexity and Claude, publish fresh, opinionated takes with clear data points and a visible publish date.
• For all four, write definitions a model can lift in one sentence.

Structuring a page to be quoted

The template we use on every GEO page is boring on purpose. The first 80 words must contain a definition, a number, and an opinion. The H2s must read like questions a buyer would actually ask. Every section must end with a one sentence summary.

"Write the sentence you want the model to quote. Then write the page around it."

The quotable paragraph

A quotable paragraph is one to three sentences, contains a specific claim, and stands alone without context. Models love these because they are safe to lift. Write three to five of them per page, spaced through the article, and you give the engine a buffet to choose from.

How we measure GEO performance

We track three numbers monthly. Citation share inside a fixed prompt set. Branded prompt volume from analytics. Assisted conversions from LLM referrers. None of these come out of Search Console. You have to build the measurement.

A 30 day starter playbook

• 1

  Week 1. Lock 25 prompts that map to real buyer questions. Baseline citation share across the four engines.

• 2

  Week 2. Audit your top 10 pages against the four-part model. Rewrite the first 80 words on each.

• 3

  Week 3. Publish two new opinionated explainers with a clear author and date.

• 4

  Week 4. Re-run the prompt set. Track movement. Brief the next cycle.

GEO is not a new acronym to sell. It is the next phase of how brands get discovered. The brands that take it seriously in 2026 will be the ones quoted in 2027. The rest will wonder why their organic traffic looks fine but no one knows who they are.