AEO: how companies appear in AI answers
More and more searches end without a click: the answer is already there. Anyone who does not appear in it does not exist for those users – even if they rank first in the classic results. This article shows why that happens and what can be done about it.
In short
- AEO does not replace SEO. Almost everything that makes a page findable for search engines also makes it usable for answer engines – the difference is in the form of the content, not the technology.
- What gets cited is what can be answered and evidenced in a single paragraph. Anyone who only gets to the point in the fourth section is skipped.
- The single biggest lever is structure: question as the heading, answer in the first sentence below it, reasoning after that.
- Measurement is still incomplete. If an agency promises exact AEO rankings, ask to see the data source.
Search has split in a few short years. Some users still enter terms and click results. A growing share asks a question in full sentences – in ChatGPT, in Perplexity, in Google's AI answers – and receives a finished answer with a few source references.
For companies that means an uncomfortable shift: visibility no longer depends only on being found, but on being cited. And that follows partly different rules.
What changes compared with SEO
The good news first: most of the work is the same. A page that loads fast, is cleanly structured, is delivered without technical faults and has something to say is a good basis for both routes. Anyone starting from zero on SEO does not start with AEO – they start with SEO.
The four actual differences
- The query is longer
- Instead of "marketing automation cost" the question is: "What does marketing automation cost for a company with ten employees?" Content optimised for single terms fits that worse than content that answers a question.
- It is read out, not skimmed
- An answer engine reads the text, extracts statements and reassembles them. Anything that cannot be lifted out as a standalone statement is lost – including everything that only makes sense because of the three paragraphs before it.
- Evidence beats phrasing
- A concrete, checkable statement is picked up sooner than an elegant one. "Four weeks is the minimum" is quotable, "it takes some time" is not.
- The click is not guaranteed
- Even when you are cited, the user may never reach the page. The gain is then the mention itself – the brand stands in the answer. That is less than a visit, but considerably more than nothing.
Worth knowing
The different answer engines work differently. Some search the web at runtime and cite what they find. Others answer largely from what was learned during training and only search when needed.
In practice: with the former, fresh and well-structured content takes effect quickly. With the latter, what counts is how often and how consistently a brand appears across the web – that takes longer and cannot be influenced in the short term.
Six measures that actually work
Question as heading, answer in the first sentence
Write subheadings as the question someone would actually ask. The first sentence below answers it completely. Reasoning, qualification and example come afterwards.
Why it works: that pairing of question and concise answer is exactly the shape that can be cleanly lifted out.
Effort: low · Effect: highConcrete numbers instead of orders of magnitude
"Between four and six weeks" is quotable, "reasonably quickly" is not. Where a number cannot be evidenced, the qualification belongs with it – that is a statement too.
Why it works: answer engines prefer statements that can be checked, because they lower the risk of a wrong answer.
Effort: medium · Effect: highMark up questions and answers
An FAQ section at the end, machine-readable as FAQPage. Google barely shows its own results from these any more – answer engines still read them.
Why it works: it is the cleanest way to link a question unambiguously to its answer.
Effort: low · Effect: mediumAuthor and company as an entity
Who writes, in what role, at which company – stored as structured data, not only as text. Plus a consistent company name everywhere on the web.
Why it works: answer engines prefer sources they can place. A nameless page is harder to assess than a named company with named authors.
Effort: medium · Effect: high, but slowPermit access
Check in robots.txt whether the answer engines' crawlers are allowed. Many websites block them without anyone having decided to.
Why it works: what may not be read will not be cited. The flip side belongs to the decision: the same access usually also permits use for training.
Effort: minimal · Effect: decisiveProvide an llms.txt
A short Markdown file in the root that describes the site and its most important pages. The convention is young and not yet universally supported.
Why it works: when it is read, it saves the machine the crawling. Costs ten minutes – with uncertain but possible effect.
Effort: minimal · Effect: openThe structure that gets cited
The pattern below has proven robust. It sounds simple but contradicts how most specialist texts are written – leading up to the point rather than answering it.
We built the same content twice. Once classically: introduction, lead-in, argument, conclusion at the end. Once inverted: answer first, then reasoning, then qualification.
When put to several answer engines, the second version was consistently picked up – and specifically its first paragraph, almost verbatim. From the first version, often nothing was taken at all, although the same content was in it.
The lesson: it is not about more content but about its order. The closing sentence of a paragraph is too late for this kind of search.
How to measure it
Honesty belongs here: there is as yet no reliable measurement comparable to classic rankings. What does work is the following.
Fixed questions, asked regularly. Define twenty questions your audience would actually ask. Put them to the relevant systems once a month and record whether your brand appears and whether it is linked. That is laborious and repeatable – and therefore more dependable than any tool that outputs a number without a traceable basis.
Separate referrals in analytics. Visits from answer engines arrive with their own referrer. Set up a dedicated segment for them. The numbers are usually small, but the trend over months is informative.
What does not work: determining a position. There is no third place in an AI answer. Either you appear or you do not, and that can differ from request to request for the same question.
Prompt for self-assessment
This lets you check an existing text for its usability in answer engines without walking through every paragraph yourself.
Here is a specialist text from our website. [TEXT] Assess it solely for usability by answer engines: 1. List every subheading and state which question it answers. If it answers none, write "none". 2. For each section, name the sentence that could be cited as a standalone answer. If there is none, say so. 3. List every statement that contains a number or a concrete value. 4. Name the three places where the text takes longest to get to the point. Do not rewrite the text. Assess only.
In closing
AEO is not a new discipline but a shift of priorities within a familiar one. Anyone already working cleanly – fast site, clear structure, evidenced statements, named authors – has to change little. Anyone producing content that only gets to the point at the end has to reverse the order.
The strategic point is timing. The field is young, competition is thin, and demand is building. Authority established here over the next months is disproportionately more expensive to catch up on later – much as it was with classic search fifteen years ago.
Common questions
What is the difference between SEO and AEO?
SEO aims at ranking high in a results list. AEO aims at being cited as a source inside a finished answer. The technical foundation is largely the same; the difference lies in the form of the content – answer first, evidenced statements, clear question-and-answer structure.
How does my company get into ChatGPT's answers?
Through three things: access must be permitted in robots.txt, the content must directly answer the questions people actually ask, and the company must be consistently named and placeable across the web. There is no lever that forces this in the short term.
Should you block or allow AI crawlers?
That is a business decision, not a technical one. Allowing means visibility in answers and at the same time that the content may be used for training. Blocking protects the content and makes you invisible. Whoever seeks reach allows it – whoever treats content as a core asset weighs it up.
What is an llms.txt and do you need one?
A short Markdown file in the root describing the site and its most important pages for AI systems. The convention is young and not yet universally supported. It costs a few minutes – so little speaks against it, but do not expect a guaranteed effect.
How do you measure visibility in AI answers?
Only incompletely so far. What is practical is a fixed set of around twenty questions, put to the relevant systems monthly and logged, plus a dedicated analytics segment for visits from answer engines. A position figure like a classic ranking does not exist.
Is AEO worth it for small companies?
Especially there. Competition for citations is lower than competition for first place, and a specific, well-structured specialist text has good prospects on a concrete question – even without the domain authority of a large provider.
Marketing that sets itself up
The Studio Engine beta is live. Claim your spot and help shape it from the start.
Join the beta →