Generative engine optimisation: GEO, AEO and AI SEO – measured

The acronyms, defined

Four names, one job

Generative engine optimisationGEO · the umbrella

The umbrella term. It covers any deliberate effort to influence how AI assistants – ChatGPT, Gemini, Perplexity, Claude and their kin – describe, cite or recommend a brand when answering questions.

Answer engine optimisationAEO · the sibling

The near-identical sibling, coined for engines that answer directly rather than listing links. In practice AEO and GEO describe the same work, and the industry uses them interchangeably more often than it admits.

AI SEOAI SEO · the adapted one

Usually means traditional search optimisation adapted for AI-generated results – Google’s AI Overviews especially. Sometimes it means “SEO done with AI tools”, which is a different thing entirely. When you meet the term, check which one is being sold.

LLM SEOLLM SEO · the fourth name

The same idea again, named for the technology (large language models) rather than the surface. Same goal, fourth name.

The terms overlap because the field is young and hasn’t settled its vocabulary. Southlume measures the outcome all four chase – who actually gets recommended – and practises none of them. We’re the scoreboard, not a team.

The honest test

How you’d know if it worked

Here’s the uncomfortable centre of this whole space. A business can buy GEO services, follow every AEO checklist, and still have no idea whether anything changed – because a single check of an AI answer proves nothing. We measured this: the same buying question, asked four times, returned 22 different businesses from one assistant – and only one of them turned up in all four answers.

So the only honest test of GEO, AEO or any of the acronyms is repeated, dated measurement. The same questions, on a schedule, with the randomness accounted for.

According to Southlume’s July 2026 measurements, that randomness is large enough that most single-check “AI visibility reports” can’t distinguish a win from a coin flip.

That measurement, run across Australian categories, is the job of The Southlume Index. The full method – questions, assistants, bands, dates – is public.

The playbook

What GEO practitioners actually do

Described neutrally, because we measure this work rather than sell it. The common playbook: making a site’s content structured and quotable, so an assistant can lift a clean answer from it. Clarifying what a business actually is, in plain machine-readable terms. Earning mentions in the sources assistants read – press, directories, reviews, industry lists – since assistants recommend what their sources talk about. And monitoring the answers to see whether any of it moved.

Some of that is old-fashioned good practice wearing a new acronym. Some of it is genuinely new. Whether any given version of it works, for a given business, in a given category – that’s not a matter of opinion. It’s a reading.

The history

Where the terms came from

A short history, for orientation. Search optimisation is decades old. Answer boxes and featured snippets pushed it toward “position zero” in the 2010s. Then conversational assistants arrived in late 2022, and by 2023 researchers and marketers were coining terms for optimising toward generated answers – GEO spread from academic work around then, AEO alongside it, with AI SEO and LLM SEO following as the industry reached for names buyers might search.

The vocabulary is still settling, and it will keep mutating. The underlying shift is stable: some buyer questions that used to end in a list of links now end in an answer that names names.

A different discipline

GEO vs SEO

SEO earns a page a position in the links; GEO shapes what the answer says.

Related – strong SEO fundamentals often help both – but they’re different surfaces with different rules, and neither replaces the other. Anyone declaring one of them dead is selling the other one.