Generative engine optimisation, measured rather than described

Generative engine optimisation (GEO) is the common term in the research literature. It describes optimising for engines that generate an answer rather than return a list.

Answer engine optimisation (AEO) is the same idea named after the output instead of the model.

AI SEO is what most buyers actually type into a search box, so it is the term we lead with.

All three are also written the American way, optimization. If an agency tells you these are fundamentally different services, be careful.

• An audit scoring AI crawler access, structured data, page structure and metadata.
• Technical fixes that decide whether you are eligible to be cited at all.
• Content built to be quoted: specifics, ranges, dates, definitions and comparisons rather than positioning copy.
• Pages answering the questions buyers ask an engine, in their phrasing.
• Daily measurement across six AI surfaces against a verified baseline.

We sample six AI surfaces every day: ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and Google AI Mode. For every buyer question we track, we record whether you were named, whether you were cited with a link, and which competitors were named alongside you.

That matters because AI visibility is not a stable ranking. Independent measurement puts day to day overlap in the sources AI engines cite at roughly 0.34 to 0.42, so one check on one day tells you very little. A screenshot of a flattering ChatGPT answer is not evidence. A daily series across six engines is.

The published research is less flattering than the marketing around it, and it shapes how we work.

Being retrievable matters more than being clever. A 2026 survey of generative engine optimisation research found that moving a source higher in the retrieved set has a greater effect than most rewrites of the page itself. Ordinary search fundamentals are still the largest lever on whether an engine can use you at all.

There is no universal recipe. One benchmark found only three of fifty four optimisation method and industry combinations produced a significant gain. Tactics that work in one market fail in another, which is why we test per engine and per sector.

Engines disagree with each other. Audits put the overlap between the sources different engines cite at roughly 11% to 26%. Being named by Perplexity tells you almost nothing about whether ChatGPT will name you.

Optimising the wrong way can cost you. There is published evidence that tuning page bodies purely for citation can reduce ordinary search presence. We treat retrieval and citation as one problem rather than trading one for the other.

We work across the UK events industry: stand designers and contractors, furniture and equipment hire, lighting, AV and production, organisers, venues, caterers and planners. That is where our measured results are.

One exhibition client went from a verified zero to being named in 17.5% of answers across six engines in eight weeks. Verified means we confirmed the absence was real at the start, not a measurement failure, by checking that competitors were correctly detected in the same answers.

We are deliberately specific about the sector because it is where we can show you numbers rather than describe a process. If you are outside events we will tell you honestly on the call whether we are the right fit.

Two things we will not claim, because the evidence does not support them.

AI engines do not send much traffic yet. Independent panel data puts referrals from AI assistants at under 1% of all web traffic. The value of being named is that a buyer arrives already knowing who you are, or searches for you directly later. It is rarely a click you can watch happen.

Attribution is early. We install first-party tracking that records AI crawler visits, AI referrals and enquiries, so the picture sharpens over time. Anyone promising you a tidy lead count from ChatGPT today is guessing.

What we can show you is measured: how often each engine names you, whether it links to you, and who it names instead.