AEO / AI Search

Answer Engine Optimization (AEO): A Practical Guide

Published August 27, 2026

Search is splitting into two behaviors. People still type queries and click links. Increasingly they also ask a question and read a synthesized answer that names a few sources — or names none at all.

Answer Engine Optimization is the work of being one of the sources that gets named.

This is measurable, not theoretical. The first systematic study of it — GEO: Generative Engine Optimization (Aggarwal et al., KDD 2024) — built a benchmark of generative-engine queries and tested which content changes actually increased a source’s chance of being cited. Their headline finding: the right changes boosted visibility in generative engine responses by up to 40%. Notably, the changes that worked were things like adding citations, quotations and statistics. Keyword-style optimisation was not among them.

What changes when an answer engine replaces a list of links?

Traditional SEO competes for a position in a list — a list that is now harder to observe from the outside than it used to be, since Google began routing result clicks through a redirect of its own. The user sees ten links and picks one. Your job is to be high in that list and compelling enough to click.

An answer engine does something different. It reads many sources, synthesizes a response, and presents it as prose. There may be citations. There may be one. The competition is not for a rank — it is for inclusion in the synthesis.

That changes three things:

1. Being findable is not enough. The model has to be able to state a clear fact about you, attributable to a source.
2. Ambiguity is fatal. If the system cannot tell whether two mentions refer to the same business, it will use neither confidently.
3. Answer-shaped content wins. Content that states a conclusion plainly is easier to quote than content that circles the topic.

What actually moves the needle?

1. Entity clarity

An answer engine needs to resolve “who is this” before it can say anything about you. That means one consistent identity across everything it can see.

Practically:

• Structured data with stable identifiers. Use Organization and Person schema with a consistent @id across every page, so all your pages describe one entity rather than several.
• sameAs links to verified profiles. LinkedIn, X, industry directories, Crunchbase, Wikidata where applicable. This is the single most-skipped item and the most valuable — an empty sameAs gives the system nothing to anchor to.
• Identical name, address, and phone everywhere. The same NAP consistency that local SEO demands, for the same underlying reason.

2. Structured data that describes meaning

Schema markup stopped being a rich-snippet trick and became a machine-readable description of what your content means.

The types that carry weight for most businesses:

• Organization / LocalBusiness — who you are, where, what you offer
• Person — for founders and authors, linked to the organization
• FAQPage — direct question-and-answer pairs, which map almost exactly onto how people query answer engines
• Article with author, datePublished, and dateModified
• speakable — marks the passages safe to read aloud verbatim

A caveat on that last one, because it is widely oversold. Google’s own speakable documentation still labels it “in beta and subject to change”, and states it works “for users in the U.S. that have Google Home devices set to English, and publishers that publish content in English.” It costs nothing to add and it positions you for voice answers, but it is not what gets you cited by ChatGPT or Perplexity. Anyone selling speakable as an AI-search unlock is overstating it.

The critical rule: the schema must match the visible page. Markup describing content a human cannot see is a manipulation signal, and it is the fastest way to lose trust with both Google and AI systems.

3. Content written as answers

Look at how people phrase things to an assistant. They ask complete questions: “how much should I pay for local SEO in Austin,” not “local seo pricing austin.”

Content that gets quoted tends to:

• State the answer in the first paragraph, then explain. Burying the conclusion under 800 words of preamble makes the passage unquotable.
• Use a question as the heading, matching how it is actually asked.
• Give specific, checkable facts — numbers, ranges, named steps. Vague claims do not survive synthesis.
• Include a short summary block near the top that stands alone as a complete answer.

This is why every guide on this site opens with a “Short answer” block. It is not a stylistic choice.

4. Crawler access

You cannot be cited from a corpus you are excluded from. Check your robots.txt for the AI user agents:

User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

The exact agent names matter, and they change. OpenAI publishes its crawler list at platform.openai.com/docs/bots, and it is worth knowing that they do different jobs: GPTBot collects training data, while OAI-SearchBot and ChatGPT-User are what fetch pages to answer a live question. Blocking the first does not block the second. If your goal is citation rather than corpus inclusion, that distinction is the whole ballgame.

This is a genuine decision, not a formality. Allowing them means your content may be used in training and retrieval without a click back. Blocking them protects that but removes you from the answers your customers are increasingly reading. For a local service business trying to be discovered, visibility almost always wins — but make the choice knowingly.

One trap worth checking for. Some hosts and CDNs now inject their own managed block into robots.txt above yours, which can leave two contradictory groups for the same crawler — a Disallow: / from the platform and an Allow: / from you. RFC 9309, the robots.txt standard Google implements, says rules for the same user agent get merged and the least restrictive equivalent rule wins, so Allow should take precedence. But that relies on every parser doing it correctly. Fetch your live robots.txt — not the file in your repo — and read what is actually being served.

An llms.txt file at your root is also emerging as a convention (llmstxt.org): a plain-text summary of what your site is and where the important pages are, written for machine consumption.

Does AEO replace SEO?

AEO sits on top of SEO, not beside it. If your pages are slow, uncrawlable, thin, or duplicated, no amount of schema fixes that — answer engines draw heavily on the same signals search engines do.

The sequence that works is: get the technical foundation right, build genuinely useful content, then add the entity and structure layer that makes it machine-legible. Doing the third without the first two produces well-marked-up pages nobody cites.

Where should you start?

If you do only three things:

1. Add Organization and Person schema with consistent @ids and a populated sameAs array.
2. Rewrite your top five pages to answer their core question in the first paragraph.
3. Decide deliberately about AI crawler access and set robots.txt accordingly.

That covers most of the gap for most businesses. The rest is the ordinary discipline of publishing things worth citing.