What is answer engine optimisation?

A search results page shows ten links and the buyer chooses. An AI answer names two, three or four businesses and the buyer chooses from those. Everything below that cut is not ranked lower — it is absent. That changes what winning looks like: not a position, but inclusion.

The mechanics differ too. An engine composes its answer from what it has read during training and what it retrieves at question time, then paraphrases. It has to be able to state what you do, where you do it, and why you are a reasonable suggestion — in a sentence, without hedging. If your site never says that plainly, and nothing else on the web corroborates it, the engine has nothing to lift and names someone whose story is easier to tell.

• Your own site — but only the parts a crawler can reach and a model can parse. Text in images, text rendered only after a click, and copy that never states your category or city are all invisible.
• Your robots.txt. Blocking relevant crawlers — OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended — can reduce an engine's ability to retrieve or cite content directly from your site. Engines can still learn about you from search indexes and third-party pages, so this is a limit on direct retrieval rather than a total opt-out.
• Structured data — business type, address, opening hours, services, service area, reviews — which makes key information machine-readable and reduces ambiguity for systems that use it.
• Third-party corroboration: directories, review sites, industry associations, local press, forum threads and comparison articles. A single unsupported claim on your own site carries less weight than the same claim repeated somewhere independent.
• Question-shaped content. Answers get assembled from text that already reads like an answer, which is why a page that states a question and answers it directly is quoted more often than a brochure paragraph.

Each provider uses separate user-agents for training, for search indexing and for fetching a page while answering a live question. Blocking one is not the same as blocking another, so it is worth naming them individually in robots.txt.

Allowing a crawler does not oblige an engine to name you, and provider agent names change over time — re-check your robots.txt against each provider's published list rather than treating a table as permanent.

• Measure first. Find out how often you are named now, across a fixed question set and several engines, and record who is named instead.
• Unblock the assistant crawlers and publish an llms.txt so a reader arriving at your root gets a plain summary of who you are.
• State the boring facts in structured data: what you are, where you are, when you open, what you charge for, which suburbs you serve.
• Write the pages your buyers' questions imply, and answer each question in its first paragraph rather than three screens down.
• Earn corroboration off-site. This is the slowest lever and the strongest one.
• Re-measure the same question set later and compare rates, not screenshots.

The first and last steps are the ones people skip, and they are the only ones that tell you whether any of the middle worked.

Nobody controls model output. Two identical questions can produce different names, engines disagree with each other, and a change in the underlying model can move every result overnight. AEO cannot promise a position, and any offer that does is selling certainty it does not have. What it can do is remove the reasons an engine has for not naming you, and measure the difference in rates over time.