What is answer engine optimisation

The short answer: Answer engine optimisation is the work of making your brand and your pages legible to the systems that write answers instead of listing links. It covers three things: whether an engine can fetch and parse your pages at all, whether it can tell who you are and what you sell, and whether anything it already trusts says so too. The output is not a rank. It is whether your name appears in the answer, and whether the sentence next to it is correct.

Most of the confusion around the term comes from people assuming it is a renamed version of search optimisation. Some of it carries over. A good deal does not, because the surface has changed shape.

The surface changed, so the work changed

A search results page gave you a position and a click. An answer gives you a sentence. If your brand is in that sentence, you are in the consideration set before the buyer has visited anything. If it is not, you are not in the running, and no amount of position eleven will save you, because there is no position eleven.

That has two consequences worth sitting with. The first is that the unit of success is a mention with a correct claim attached, not a ranking. The second is that the thing being read is often not your page. It is a passage from your page, pulled out of context, next to passages from three other sites. Answer-first writing exists because of that second point.

What an engine actually does with your site

The engines that describe their own retrieval describe a version of the same loop. Google's documentation on AI features sets out how its systems fetch, use and cite pages, and OpenAI publishes which of its agents request a live page and why. In that shape, the system turns the user's question into one or more queries, fetches candidate documents from a search index, a cache, or the live page, cuts those documents into passages, keeps the few that look most relevant, writes an answer from what it kept, and attaches citations to some of it. Engines that publish nothing may work differently, and we cannot see inside them.

Each step is a place to lose. A page behind a bot rule is not fetched by a crawler that honours that rule. A page that renders its content in the browser can arrive empty at any fetcher that does not run JavaScript, and the engines differ on whether they do. A page that buries the answer under four hundred words of preamble gets cut into passages that do not contain the answer. A brand with inconsistent facts across its own site gets summarised as whichever version the engine happened to keep.

Access is the step you can check rather than guess at, because both of those documents name the user agents involved and the rules they say they follow.

Three layers, in the order that pays

The work sorts into layers, and the order matters more than the list.

Access. Can the systems reach the page, render it, and read it inside their time budget. This is unglamorous and it is where most of the recoverable loss sits. A firewall rule, a client-side render, a redirect chain, a slow origin. None of it is interesting. All of it is fixable in days rather than quarters.

Identity. Can an engine tell that the company in the page, the company in the schema, the company in the directory listing and the company in the news article are one entity. This is where structured data, consistent naming and the links between your own profiles do their work. When identity is loose, an engine hedges, and a hedge reads as absence.

Corroboration. Does anything the engine already trusts repeat your claim. This is the part you do not own, and the part that takes longest. It is also the reason the rented and owned split is worth drawing on a whiteboard before anyone spends money.

What it is not

It is not a keyword exercise with a new name. Prompts are longer, messier and more specific than keywords, and they vary by engine.

It is not a dashboard. A visibility score with no evidence underneath it tells you a number moved, not why, and not what to do on Tuesday.

It is not a one-time fix, because the answer is regenerated every time somebody asks. In our own audit runs we have watched two identical prompts, ten minutes apart, return different sources. We have not measured how often that happens, or on which engines it happens most, so read it as something we have seen rather than a rate. If that surprises you, read why an answer changes between runs before you read any report that quotes a single run as evidence.

And it is not separate from the thing you already do. The naming argument between AEO, GEO and SEO matters far less than whether anyone in your company owns the three layers above.

How to tell whether it is working

Pick the questions a buyer actually asks, not the ones that flatter you. Run each of them several times, on more than one engine, because a single run is an anecdote. Record three things each time: whether you were named, whether what was said was true, and which sources the engine leaned on.

Then change one thing and run it again. Fix the access problem, wait, re-run. Correct the contradictory pricing page, wait, re-run. Slow, boring and checkable beats a score that moves for reasons nobody can name.

An audit is the compressed version of that: a prompt set of around 45 prompts, scoped to the client, across five engines, run with retrieval on and off, with the evidence kept so a finding can be argued with. The sample audit report shows the shape of it. If you would rather do the first pass yourself, the self-check is ten things you can test this afternoon with no tooling at all.

What we are still measuring

This post does not establish how any specific engine ranks or selects passages, because none of them publish that, and the loop described above is drawn from the documentation Google and OpenAI do publish rather than from inside the systems. We have no clean attribution model connecting a mention in an answer to revenue, because many of these surfaces send little or no referrer, and absence in our logs is not proof that nobody read the page. We have seen identical prompts return different sources minutes apart in our own runs, but we have not published a sample size, a date range or a method for that, so it is an observation and not a measurement you can reproduce. We also do not know how durable a fix is: we re-run client prompt sets monthly and results move, and we cannot separate our own changes from changes the engines made in the same month.

FAQ

Is answer engine optimisation different from SEO? It overlaps and it is not identical. Crawlability, clean markup and real content serve both. The difference is the goal: a passage that can be lifted and a brand identity that resolves, rather than a position in a list.

How long before anything changes? Access fixes can show up as soon as a crawler returns, which varies by site and by engine and is not a figure anyone publishes. Identity work is slower. Corroboration from third parties is the slowest of the three and the least predictable, so treat anyone offering a date with suspicion.

Do I need a tool to start? No. You need a question list, a spreadsheet and the discipline to run the same prompts repeatedly. Tools become worth it when the manual version gets too expensive to repeat.

Does blocking AI crawlers protect my content? It also removes you from the answers those systems write. That may be the right call for a paywalled archive and the wrong call for a product site. It is a business decision, not a technical default.

Sources

• Google Search Central, documentation on AI features and your website, checked 12 October 2026: developers.google.com/search/docs/appearance/ai-features
• OpenAI, bots and crawler user agents, naming GPTBot, OAI-SearchBot and ChatGPT-User, checked 12 October 2026: developers.openai.com/api/docs/bots