Answer Engine Optimisation, Explained Without the Hype

A growing share of buyers now ask a model instead of running a search. They type a question into ChatGPT, Claude or Perplexity, get a short answer naming two or three companies, and act on it. If you are not in that answer, you were never in the consideration set, and no amount of ranking fourth on Google fixes it.

That is the whole of it. Everything else is implementation detail.

The acronyms are mostly the same thing

You will see AEO (answer engine optimisation), GEO (generative engine optimisation), LLMO (large language model optimisation) and AI SEO used more or less interchangeably. There are people drawing careful distinctions between them. Those distinctions do not currently change what you would actually do on a Tuesday.

I use whichever term the client already uses and get on with the work.

What actually changed

Three things, and only three.

The result set got shorter. A results page shows ten options and invites comparison. An AI answer names two or three and the buyer often stops there. That is a harsher failure mode than ranking fourth: fourth still gets seen, absent does not exist.

Retrieval replaced ranking. Search ranks documents. Models retrieve passages and synthesise. What gets lifted is a self-contained chunk that answers a specific question, not a page that ranks well overall. A page can rank first and never be quoted, because nothing in it is quotable.

Identity started mattering more than position. Models cite sources they can identify and corroborate. If your organisation is ambiguous across the web, has inconsistent naming, or has no resolvable entity behind it, a model has a reason not to name you even when your content is good.

What you can actually influence

This is where most of the advice gets vague, so here is the concrete version.

Being readable at all. Google executes JavaScript on a delayed second pass. GPTBot, ClaudeBot and PerplexityBot largely do not. A client-rendered single-page app is an empty document to them. This is the single most common reason a site with decent Google rankings gets no AI citations whatsoever, and it is usually invisible in every report the marketing team is looking at. Check what the crawler receives, not what your browser renders.

Being identifiable. Organisation and Person markup, consistent naming, and third-party signals that agree with each other. This is unglamorous and it is where most of the durable gain sits. A model needs to resolve “who is this” with confidence before it will attach your name to a claim.

Being extractable. Answer engines lift passages. A definitive answer sitting in a self-contained block, with enough specificity that quoting it is safe, gets lifted. Three paragraphs of throat-clearing before the answer does not.

Being specific. Models are noticeably more willing to quote a concrete, falsifiable statement than a vague one. “Most firms see improvement” is not quotable. “Local and technical work typically moves the map pack within two months; competitive practice-area terms take longer” is.

What nobody can promise

You cannot control what a model says. Anyone telling you otherwise is overselling, and you should treat it the way you would treat someone guaranteeing a Google ranking.

There is no submission process. There is no AEO equivalent of a sitemap ping that makes you appear. llms.txt exists and is worth having because it costs nothing, but it is not a ranking factor and I would be suspicious of anyone selling it as one.

Model behaviour also shifts. A citation you hold this month can vanish next month because the underlying model was updated, not because you did anything wrong. That is why this is a monitoring discipline rather than a project with an end date.

How to know whether it is working

The honest answer is that you need a baseline, and almost nobody captures one before starting.

Build the set of prompts your buyers actually use. Not keyword phrases — the questions people type into a chat box, which are longer and more conversational. Run them across the major engines on a schedule. Record whether you are cited, and critically, who is cited instead.

That gives you a citation share you can track over time and a competitor set you can benchmark against. Without it, every claim of improvement is unfalsifiable.

If someone is selling you AI visibility work without a baseline and a repeatable measurement, there is nothing to hold them to.

Is it worth doing yet

For most businesses, it is early enough that competition is thin, which is exactly why it is cheap to win now. The work also overlaps heavily with good technical SEO, so a lot of it you should be doing regardless.

But it is still a small channel for most industries. I would not fund an AI visibility programme ahead of fixing a site that cannot convert the traffic it already has, or an intake process that drops leads. Sequence matters more than enthusiasm.

If you want to know where you actually stand, the free AI visibility scan runs a version of the baseline described above. The AI visibility service is the full version, and SEO covers the crawler accessibility and schema work that underpins both.