What is Generative Engine Optimisation (GEO)?

How AI answer engines choose who to cite

There are two paths by which a business ends up cited in an AI-generated answer. The first is live retrieval: when a user submits a query, the engine searches a real index and synthesises an answer with citations drawn from the freshest results it finds. ChatGPT and Microsoft Copilot draw primarily from Bing, Google AI Overviews and Gemini from Google Search, and Perplexity from its own crawler. Live retrieval is the primary mechanism for most cited answers today, and it is the faster lever to pull.

The second path is trained-in knowledge: information baked into the model during its training run. This knowledge is slow to change, updates only at the next training cut, and reflects broad presence across a wide range of sources on the web rather than any specific optimisation. A business that is well-covered by independent third-party sources will benefit from trained-in knowledge over time, but it is not something you can engineer quickly.

Because live retrieval dominates, the most effective GEO work focuses on making your site retrievable and citable in real time rather than waiting for model retraining cycles.

The levers that make a site citable

Six factors determine whether an AI engine can retrieve and cite a given page. They apply regardless of which engine you are targeting, because the underlying mechanics of retrieval and trust are broadly shared.

• Retrievability: the page must be crawlable by AI bots, served as real HTML rather than JavaScript-only content, and present in the search indexes the engines draw from.
• Answer-shaped content: self-contained passages that answer one question up front, with definitions and comparisons, are what models lift and quote. Marketing copy that never directly answers a question is invisible to them.
• Atomic citable URLs: a page dedicated to a single topic is far more citable than a homepage covering ten. Each question, service, or concept deserves its own URL.
• Entity authority: structured data combined with corroboration across independent sources (company directories, professional profiles, named individuals) gives a model the confidence that your business is real and identifiable.
• Freshness: recently updated, clearly dated content is weighted more heavily than stale pages with no visible date signal.
• Honest signals: fabricated reviews, ratings, or testimonials are among the fastest routes to being filtered out. Genuine trust signals from real sources are the only kind that hold.

How you know it is working

Being indexed is not the same as being cited. The measurement question in GEO is share of voice: across the prompts your buyers are actually typing into these engines, how often does your business appear, and in what form?

For brand queries and highly specific questions, citations can appear within a few weeks of getting indexed, assuming the site is well-structured and the content is answer-shaped. For competitive, non-brand queries, the timeline is longer. Authority is cumulative, and a site with limited off-site presence will take months to compete with established sources in a given topic area.

The practical way to start measuring is to query the engines yourself. Ask ChatGPT, Perplexity, and Gemini about your own business, then about the category you operate in. That gives you a baseline for where you appear, how accurately you are described, and what sources the engines are trusting instead of you.

Because AI answers vary between runs on the same query, a single snapshot is not reliable. Meaningful measurement means sampling across multiple engines, multiple runs of the same prompt, and tracking the trend over time rather than treating any one result as definitive.

Can you do it yourself, or should you hire

The on-page foundations of GEO are achievable in-house by a team that has already handled technical SEO. Serving real HTML, adding structured data, restructuring content to answer specific questions directly, and publishing an llms.txt knowledge file are all well within reach for a competent developer and a content lead working together.

The harder work falls into two areas. The first is entity corroboration: building the off-site presence across independent sources that gives AI engines the confidence to cite you. This requires a coordinated effort across directories, professional profiles, and earned coverage that most in-house teams deprioritise because the payoff is not immediate. The second is measurement: tracking share of voice across multiple engines at scale, sampling enough to see a trend rather than noise. Both areas are where most teams find the effort-to-signal ratio too high to sustain without dedicated support.