What does an AEO-native website actually look like?

Most websites are built for a person who has already arrived. An AEO-native website is built for a machine that arrives first, extracts a passage, and decides whether to name you in an answer the person reads instead of your page.

Those are different design problems, and the second is almost entirely invisible. A site can be beautiful, fast, accessible and completely unretrievable — because the answer to the question a buyer asked sits in paragraph four, the figure supporting it has no source, and the organization behind it cannot be resolved to a single entity with confidence.

Rather than describe the pattern abstractly, this page examines four live properties. Three were built from nothing; one was an optimization engagement on a site that already existed. All four are open to inspection, which is the point of publishing them.

The Wellness Agency is a talent, media and capital firm whose roster includes Health-Ade, which sold for $500M in 2025 according to The Wall Street Journal. The build had to carry the credibility of that roster while addressing an APAC market entry.

The decision that matters most is language parity. Nine page types exist in English and Simplified Chinese with full parity — not a translated homepage with English interiors, which is the usual compromise. Retrieval operates per language. A Chinese-language query resolves against Chinese-language content or it does not resolve at all.

The second decision was a dated third-party press wall rather than a testimonial section. A testimonial is a first-party claim about yourself, which retrieval treats as marketing. A dated citation from a named publication is a corroborating source a model can weigh. The same area of page does structurally different work depending on which one occupies it.

Each of the ten roster companies is named, categorized and linked to its live site. That is entity work disguised as a client list: ten resolvable organizations, each associated with the agency in machine-readable form.

Mammoth China Outbound addresses roughly $60B in annual China outbound advertising spend, a figure sourced to the ResearchAndMarkets China Digital Ad Spend Databook. Its technology partner, ReachMax / AddNewer, is the largest third-party ad server in China. The site makes an investment-grade argument to two audiences in two languages.

The most transferable artifact here is a dedicated market-data page where every figure carries a named source, a publication date and a last-reviewed stamp. This inverts normal practice. Most sites treat statistics as decoration — a large number in a colored circle, unattributed. Retrieval systems treat unattributed figures as unusable, because a model repeating an uncited number inherits the risk of being wrong.

Attribution is therefore not an academic courtesy. It is the mechanism by which a figure becomes quotable. Controlled experiments by Aggarwal and colleagues at Princeton, presented at KDD 2024, measured up to 40% visibility improvement from adding quotations, statistics and source citations to otherwise identical content. A page built entirely from sourced figures applies that finding at the level of architecture.

The last-reviewed stamp does separate work. Freshness is a documented retrieval signal, and a market figure carrying a review date tells a model the number has been checked recently rather than published once and abandoned.

Golfinity is a 20,000 square foot indoor golf facility on N FM 620 in Austin with eighteen or more simulator bays running Trackman 4, Foresight GCQuad and E6 Connect, plus Club Champion fitting on site. Membership runs $99, $199 and $299 per month.

This engagement is included precisely because no rebuild happened. It is a local AEO retainer on an existing site, which makes it the cleanest available demonstration that AEO is not a synonym for redesign.

The work was alignment across three surfaces most businesses maintain independently: the website, the Google Business Profile and the local citation set. When those disagree — different hours, a different phone number, a different service description — a model attempting to resolve the business encounters competing claims and hedges. Consistency is not a tidiness exercise. It is what permits confident resolution.

Schema was limited to LocalBusiness, Service and FAQPage. Tracking uses a fixed set of conversational Austin-local queries re-run on a schedule, because between 40% and 60% of AI citations change month to month. A one-time measurement of local visibility has a short shelf life.

FabriCare of Greenville is a family-owned dry cleaner with three stores in Greenville and Greer, South Carolina, its own cleaning plant, and free twice-weekly pickup and delivery. The rebuild went live on the client's domain in August 2026. It is the least glamorous property in this set and the most instructive, because the buyer questions are entirely ordinary.

Nobody asks an answer engine for a dry cleaner in the abstract. They ask what dry cleaning costs in Greenville, whether there is a delivery fee, whether same-day service exists, whether there is a drive-thru. The build answers those questions in those words, because the retrieval unit is the question, not the topic.

The second decision is geographic granularity. Eight service pages cover dry cleaning, wash-dry-fold, pickup and delivery, wedding gown preservation, household textiles, leather and suede, same-day service and commercial uniform accounts. Six city pages cover Greenville, Greer, Simpsonville, Mauldin, Five Forks and Taylors separately.

That separation is deliberate. One page attempting to serve six municipalities competes weakly in all six, because it cannot answer any of them specifically. Six pages each answering one place can each be retrieved for that place. The same logic runs per store: each location carries its own phone number, its own hours and its own neighborhood list.

Operational specifics also carry more retrieval weight than adjectives. "In by 10:00 AM, back by 5:00 PM, flat $8.95 rush fee" is extractable. "Fast, friendly service" is not, because there is nothing in it a model can quote to somebody who asked a question.

The pricing page is worth studying on its own, because it publishes a refusal rather than a number. FabriCare has no price list — cost follows the garment, not the category — so the page explains what drives the price and commits to a quote before work begins. Inventing a list would have been inaccurate and omitting the page would have left the category's highest-volume question unanswered. An honest answer is retrievable; a fabricated one is a liability.

Across four properties in four unrelated sectors, five decisions recur. They are the transferable part of this page.

First, the answer precedes the argument. Every page opens with a direct response to the question it exists to answer, placed above the supporting reasoning rather than concluded at the end.

Second, headings are phrased as questions in the buyer's own language. "How much does dry cleaning cost in Greenville, SC?" rather than "Pricing." The heading is the retrieval handle, so it should match the query.

Third, every figure carries a source. Where a number has no attributable origin it is removed rather than published bare.

Fourth, specificity replaces adjectives. Square footage, equipment names, hours, rush fees, membership prices, page counts. A model cannot quote an adjective to a person who asked a question.

Fifth, corroboration is built rather than claimed. Dated press pages, named partners, linked roster companies, aligned third-party profiles. A model recommending a business is reproducing a consensus it found in public, and consensus cannot be marked up into existence.

None of these are visual choices, which is why AEO-native sites do not look like a genre. They can look like anything. What they share is that the answer to a real question is present, attributed and reachable in the first HTML response.