Search engines rank pages and hand the reader a list of links. Answer engines read those pages, synthesise a reply, and cite a handful of sources. The same content can win in one and be invisible in the other, which is why “should we do AEO or SEO” is the wrong question. You need both, and one of them depends entirely on the other.
What AEO actually is
Answer engine optimisation is the practice of making your content the source an AI system draws on and cites when it answers a question. Search engine optimisation is the practice of ranking a page in a list of results. The difference is the unit of success: SEO competes for a position, AEO competes for inclusion in an answer somebody else composed.
That distinction changes what “winning” looks like. A page can rank fourth and still be the source an answer engine quotes. A page can rank first and be skipped because its useful claim is buried three scrolls down in a paragraph that only makes sense after reading the two above it.
AEO does not replace SEO. It sits on top of it
This is the part most strategy decks get wrong, and it has budget consequences.
Answer engines do not have their own private view of the web. They retrieve from an index, and for most of them that index is either a search index or something built the same way — a crawler that has to find your page, fetch it, render it and store it. Google’s AI answers are built on the same Search index that Googlebot fills.
So every technical fundamental still gates everything downstream. If your pages are not crawlable, not indexed, slow to render, or duplicated across five near-identical URLs, you are not in the candidate pool that any answer gets assembled from. The technical layer has not become less important — it has become the entry requirement for two channels instead of one.
You cannot be cited by a system that never retrieved your page. AEO is a layer on top of SEO, not an alternative to it.
Answer engines quote passages, not pages
The single most useful mental shift is this: your page is not the unit being evaluated. A passage is.
Retrieval systems break documents into chunks, score those chunks against the question, and assemble an answer from the best ones. A brilliant article whose key insight only makes sense in sequence produces poor chunks. A plainer article where each section stands on its own produces good ones.
That has a concrete writing consequence. Every section should be comprehensible if someone lands on it cold, with no memory of the paragraphs above. Practically, that means restating the subject rather than leaning on “it”, “this” and “the above” — the pronouns that make prose flow are exactly what makes a passage useless once it is lifted out of context.
It also means one page should answer one clear question. Pages that try to cover four related topics produce chunks that are ambiguous about which question they answer, which is the same structural problem that causes keyword cannibalisation, showing up in a new place.
What actually changes in how you write
Answer first, reasoning second
State the direct answer in the first sentence or two under a heading, then explain. This is the inverted pyramid that journalists have used for a century, and it happens to be exactly what an extractive system needs. It is also better for the human reader, who is scanning.
Say the definition out loud
If your page never contains a plain sentence of the form “X is …”, it cannot be used to answer “what is X”. A surprising number of expert pages assume the definition and skip straight to nuance. Write the boring sentence.
Write claims that survive being lifted
Specific, checkable statements get quoted. Vague ones do not. “Structured testing and clean tracking” is unquotable; “run one variable per test and hold the audience constant, or you cannot attribute the result” is quotable. A good test: would this sentence still mean something pasted into an answer with no surrounding context and your name next to it?
Use the formats that extract cleanly
Comparison tables, numbered steps, and short definition lists are disproportionately useful to a system assembling an answer, because the structure carries meaning that survives extraction. Headings phrased as the question a person would actually type help for the same reason.
Add schema, but understand what it does
Structured data does not make you cited. It confirms, in a machine-readable way, what the page is and which entity published it. Organization, Article and FAQPage are the practical set for most content. Treat it as disambiguation, not as a ranking lever, and never describe something in schema that is not visible on the page.
The crawler decision most teams make by accident
You cannot be quoted by a system you have blocked, and a lot of sites have blocked them without a decision ever being made — usually by pasting in a robots.txt template.
The AI crawlers worth knowing by name are GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, and Google-Extended. The last one causes the most confusion, so it is worth stating plainly: Google-Extended is not Googlebot. It governs whether your content is used for Gemini training and grounding. Blocking it does not remove you from Google Search, and it does not by itself remove you from Google’s AI answers, which are built on the ordinary Search index.
There is a genuine trade-off here and it is not the same for every business. If your content is the product — a paid research library, a subscription publication — restricting AI crawlers is defensible. If your content exists to make buyers aware of you, blocking the systems that could recommend you to them is self-harm. Make it a decision, write down the reasoning, and re-check the file after any hosting or plugin migration, because these directives get overwritten quietly.
How to measure something that produces no click
Rank tracking cannot see AEO. If a system reads your page and tells the user the answer, there is no position and frequently no visit. Four measurements are worth more than any rank report here.
- Prompt testing. Keep a fixed list of twenty to thirty questions your buyers actually ask, run them across the major answer engines on a schedule, and log whether you were cited and what was said about you. It is manual and unglamorous. It is also the only direct measurement of the thing you are trying to influence.
- Referral traffic from AI surfaces. Visits from ChatGPT, Perplexity and similar arrive as ordinary referrers in analytics. Segment them. The volume is usually small; the intent is usually high, because the person has already read a summary and chosen to come anyway.
- Impressions against clicks in Search Console. Rising impressions with flat clicks on informational queries is the signature of being surfaced inside an answer rather than clicked. That pattern is information, not failure — but only if you were reading it deliberately.
- Self-reported attribution. One open text field on your enquiry form asking how someone heard about you will surface AI-assisted discovery that no analytics tool can attribute, for the same reason it surfaces the rest of your unattributable demand creation.
What to do first, in order
Do not rewrite the site. The sequence that gets results fastest starts with what is already working.
- Start with pages that already rank. They are already in the retrieval pool, which is the expensive part. Restructuring them for extraction is cheap by comparison.
- Fix structure before adding words. Answer-first openings, self-contained sections, an explicit definition, headings phrased as real questions.
- Check the crawler file. Confirm you are allowing the systems you want to be recommended by, deliberately.
- Add the schema that describes what is genuinely on the page.
- Baseline your prompts before the changes, so the follow-up test means something.
- Then write new content — against the questions the prompt test showed you losing, and mapped one intent to one page.
Where this goes wrong
- Treating it as a new trick. There is no keyword-stuffing equivalent for answer engines that survives contact with a model built to judge whether a source is worth citing. The durable strategy is being genuinely worth quoting.
- Writing for machines and losing the reader. Pages stripped down into robotic question-and-answer blocks get extracted and then convert nobody, because the human who does click finds nothing with a point of view. A position worth holding is what makes content quotable in the first place.
- Abandoning SEO on the assumption that search is finished. The retrieval layer that feeds answer engines is the search index. Disinvesting from it removes you from both.
- Chasing citations that cannot become revenue. Being quoted on a broad definitional question is pleasant and often commercially worthless. Prioritise the questions a buyer asks close to a decision.
- Measuring it once. Answer engines change their sourcing behaviour frequently. A single audit is a snapshot; the value is in the repeated test.
The strategic answer
AEO and SEO are not competing disciplines to choose between. They are two outputs of the same underlying work: being findable, being clear, and being worth citing. Technical health and topical authority put you in the retrieval pool. Structure and specificity decide whether you get used once you are there.
What genuinely changes is the reward. A share of your best content will now be read by people who never visit your site, and you will not be able to prove it happened in a dashboard built for clicks. That is uncomfortable, and it is the same discomfort as every brand-building activity that works. The businesses that handle it well are the ones that decided in advance what they were measuring and why, rather than judging a new channel with an old report.
