Why content, not schema, is the real input for AEO

Most AEO conversations start in the wrong place. Teams debate schema, headings and whether they need an llms.txt file, while the question a buyer is actually typing into ChatGPT has no clear answer anywhere on their site. Formatting and technical access matter, but they only help a system find and read information that already exists. If the answer lives in a sales call, a support ticket or a specialist’s head, no amount of optimisation will put it into an AI-generated response.

Google said this plainly when it published its guide to generative AI search in May 2026: creating unique, useful content will likely influence your presence in AI search more than any other recommendation in the guide. OpenAI’s guidance for ChatGPT search starts from the same place: any public website can appear, provided OAI-SearchBot isn’t blocked from reading it. Access is the entry ticket. What earns the citation is having an answer worth using.

How answer engines actually find you

An answer engine doesn’t read the whole web every time someone asks a question. It searches first, then writes. Google describes its AI features as built on retrieval-augmented generation (RAG): the system retrieves relevant pages from its existing Search index, often after breaking one question into several related searches, a technique it calls query fan-out, and then generates an answer grounded in what it retrieved. To be eligible at all, a page has to be indexed and able to show a snippet in regular Search.

The business consequence is simple. If your answer isn’t among the pages the system retrieves, the model has nothing of yours to work with, and it will build the answer from a competitor’s page, a review site or a forum thread instead. That is why AEO is a content problem before it is a technical one. Google’s guide even lists tactics site owners can ignore: special AI text files, dedicated markup, Markdown versions of pages and AI-specific rewrites of existing copy.

The raw material is already inside your company

Most businesses don’t need a bigger blog. They need to publish what they already know. Your sales team answers the same handful of questions every week. Support has years of resolved tickets explaining what actually breaks and why. Onboarding docs, case studies, objection-handling notes and internal data all contain specific, experienced answers that no competitor can copy.

This is exactly what Google calls non-commodity content. Its guide contrasts a generic piece like “7 Tips for First-Time Homebuyers” with something like “Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line”: the second carries expert, first-hand insight that goes beyond common knowledge. A generic article is easy for any AI model to produce on its own, so there is little reason to cite yours. A documented answer from the people who do the work is the opposite.

The practical move is translation, not invention. If prospects keep asking whether your platform integrates with HubSpot, that answer belongs on an integration page, not buried in a sales email. If clients keep confusing two of your service options, write the comparison. Every one of those pages also does a second job: it answers the question before it reaches your team, which shortens sales conversations and cuts repeat support tickets.

There is a useful overlap here for any company already building internal AI. A RAG assistant that answers employee or customer questions needs the same thing an answer engine needs: clear, current, well-scoped knowledge. Teams that have organised a knowledge base for an internal copilot have done half of their AEO groundwork, and the reverse is true as well.

Match the format to the question

There is no single “AEO format”. The Wix Studio AI Search Lab, working with data from Peec AI, analysed 75,000 AI answers containing 1,056,727 citations across ChatGPT, Google AI Mode and Perplexity in 2026. Its central finding was that query intent predicted which type of content got cited better than either the industry or the AI model. In other words, what the user is trying to do matters more than which chatbot they use.

Example prompts are illustrative. Percentages: Wix Studio AI Search Lab / Peec AI (2026).

Two details are worth reading carefully. First, dedicated comparison pages earned only 2.20% of citations overall, so an “us vs. them” page is not the shortcut it looks like. Second, the listicles winning commercial queries are mostly not brands ranking themselves: in professional services, the study found that 80.9% of listicle citations went to third-party lists and only 19.1% to self-promotional ones. The study used non-branded prompts, so read it as a map of category questions, not of searches for your brand name.

What to do with this: tag each question your buyers ask with its intent before deciding what to write. Informational questions need explanatory articles and guides. Commercial questions are largely won through independent coverage, so that budget may work harder in PR and reviews than in another list where you rank yourself first. Navigational and transactional questions need product and service pages that state plainly what you sell, to whom and how to buy it.

Clear beats clever

Brand language has a job, but answering questions isn’t it. Compare “We shape tomorrow through intelligent digital experiences” with “We build AI chatbots that answer customer questions using your existing knowledge base.” The first is positioning. The second tells a buyer, and any system reading the page, exactly what the company does, for whom and how.

Google’s guide notes that its AI systems understand synonyms and general meaning, so there is no need to rewrite pages into some special AI dialect. The fix is more basic: make sure every important page states what it is, who it’s for, what problem it solves, what it includes and what evidence backs it up. Put the direct answer near the top of the section, under a heading that reflects the real question. A well-structured vague page is still vague.

Your website is only part of the picture

Your own site is where you control how your products and expertise are described, but answer engines don’t stop there. The Wix study found citations spread across articles, listicles, product pages, category pages, discussions, profiles and more, and the mix varies by platform. Perplexity drew 17.35% of its citations from discussion pages such as Reddit, LinkedIn and G2, more than double the 7.52% average across models.

This changes where the work goes. Owned content makes your position explicit, while reviews, independent lists, directories and community discussions give answer engines outside confirmation. No single page change guarantees a citation, because different systems retrieve different sources for different questions. The goal is a consistent body of information about your business, on your site and off it, that points to the same clear answers.

Build your AEO backlog from unanswered questions

A useful AEO programme starts with an audit, not a content calendar. The sequence below keeps the effort on the gaps that actually cost you deals or support time. Each step feeds the next, so resist the urge to skip straight to step five.

1. Collect the questions

Pull real questions from sales emails, call notes, support tickets, onboarding sessions and your site’s internal search logs. Add the prompts your buyers are likely to type into ChatGPT or Perplexity about your category. Keep the wording your customers use, not the way your team describes things internally.

2. Tag each one by intent

Label every question as informational, commercial, navigational or transactional. “What is X?” and “Which X should I choose?” need different kinds of answer, and the Wix data shows they get cited from different kinds of page. This tagging decides the format before anyone starts writing.

3. Check what already exists

For each question, ask whether there is a public, clear and current answer on your site. Often the answer exists but is buried halfway down another page, out of date, or only available in an internal doc. Sometimes it doesn’t exist at all, and those are usually the most valuable gaps.

4. Close the gaps that matter

Prioritise the questions that come up in deals or drive repeat tickets. The fix might be a new article, but just as often it’s a rewritten service page, a specific FAQ or a published case study. Write from the experience of the people who handle these questions, because that is the part an AI model can’t generate on its own.

5. Then optimise

Only now move to the technical layer: descriptive headings, internal links, crawlable pages and access for crawlers such as OAI-SearchBot. This layer helps systems reach your answers and read them. It does nothing for answers that don’t exist.

Publish the answer first

AEO doesn’t start with a trick, a file or a markup format. It starts with a simple test: when your buyer asks an AI tool a question about your category, have you published an answer worth citing? If yes, make it clearer and easier to reach. If not, no optimisation will close that gap for you.

The first step fits in one afternoon. Export the last 50 questions from your sales inbox, support tickets and call notes, and check whether a public page on your site answers each one within its first two sentences. Every question that fails goes into your AEO backlog, ranked by how much of your team’s time it costs. Answer engines can only repeat what you’ve actually said, so say it in public.

Sources