What is Answer Engine Optimization (AEO)?

Why a new acronym was needed

For twenty years the unit of search was the page, and the prize was a position in a list. You optimised a URL to rank, and a human decided whether to click it.

Answer engines changed the unit. When someone asks Perplexity, ChatGPT search, Claude or Google's AI Overviews a question, the system does not hand back ten links. It retrieves a handful of passages, judges which of them actually resolve the question, writes an answer, and attaches citations to the passages it used. The page is no longer the thing being ranked — the passage is. And the reward is no longer a click but a citation, which may or may not produce a click.

Answer Engine Optimization is the name for adapting to that unit. It asks a narrower question than SEO: is there a passage on this page that a machine could lift out and use, on its own, to answer a specific question — and can it verify that the passage is trustworthy?

How an answer engine decides what to cite

The details differ between systems, but the pipeline is broadly the same, and each stage rewards different properties of a page.

1. Discovery and crawling. The engine has to know the page exists and be allowed to fetch it. This is classic technical SEO: a crawlable URL, a sitemap, no noindex, and a robots.txt that does not block the engine's user agent. OpenAI, Anthropic and Perplexity each publish their crawler names; Google uses Googlebot for AI features and Google-Extended only as a training opt-out.
2. Chunking. The page is split into passages — typically by heading and paragraph, sometimes by a fixed token window. A passage that starts mid-thought, or that depends on a definition three sections earlier, loses most of its value here.
3. Retrieval. For a given question, the engine scores passages by relevance, usually with a mix of embeddings and keyword signals. Passages whose wording resembles the question, and that contain the entities in the question, score higher.
4. Selection and grounding. From the top candidates, the model picks the ones it will actually use. Research on generative engines found that passages containing citations, quotations from authorities, and concrete statistics were used substantially more often than the same content without them — up to 40% more visibility in the Princeton/IIT Delhi GEO study.
5. Generation and attribution. The model writes the answer and attaches sources. A passage that is the answer tends to be cited; a passage that merely discusses the answer tends to be paraphrased without attribution.

Every AEO tactic maps onto one of those stages.

What AEO actually changes

Write the answer first

The single highest-leverage change is structural: open every page, and every major section, with a self-contained answer. Forty to seventy words, in plain declarative sentences, with the key entity named explicitly rather than referred to as "it". Then explain, qualify and expand.

This is the opposite of the "hook, context, reveal" structure that human editors like, and it is the reason AEO sometimes feels like writing an encyclopedia rather than a magazine. The compromise that works: a short-answer block at the top (this page's Short answer box), an ordinary article underneath, and a heading for every question you expect someone to ask.

One question per URL

Answer engines match questions to passages, and passages inherit trust from their page. A page that answers "what is AEO" and "how do I measure AEO" and "what tools exist for AEO" has diluted signals for all three. Give each question its own URL, title and canonical, and link them.

Make entities unambiguous

Language models resolve entities — people, products, concepts — by name. Use the full name on first mention, add the abbreviation in parentheses, and be consistent afterwards. A glossary with one page per term, marked up as DefinedTerm, is one of the cheapest ways to make a whole site easier to understand.

Show who, when and why

Retrieval systems cannot read your page for quality the way an editor can, so they lean on signals they can check: a named author with a profile, visible published and updated dates, and outbound links to primary sources. Google's own documentation for AI features points at the same E-E-A-T guidance it uses for search.

Mark it up

Structured data does not make a bad passage good, but it removes ambiguity about what a passage is. Article (with author, datePublished, dateModified, citation), FAQPage, DefinedTerm, Person and SpeakableSpecification are the types that matter for answers. Generate them from your CMS fields so the markup can never disagree with the visible text.

Ship a machine-readable version

Rendering a page with JavaScript, then expecting a crawler with a small fetch budget to execute it, is a bet against yourself. Pre-render HTML on the server, keep the DOM semantic, and offer the same content as Markdown — a .md suffix, or content negotiation on Accept: text/markdown. An /llms.txt file gives models a curated index of what you have.

What AEO does not change

It is worth being clear about the limits, because the acronym attracts hype.

• It does not bypass quality. Every engine that cites sources still filters on the same trust signals as search. Thin, derivative pages with perfect markup are still thin, derivative pages.
• It does not guarantee traffic. A citation is a mention, not a visit. Some answer engines send meaningful referral traffic, others almost none. Measure both.
• It is not a trick. The tactics above are the same things that make a page easier for a hurried human to use: say the answer, say who wrote it, show your sources.