AI Search / GEO Build for luxury ecommerce

Your next customer might not be a person.

ChatGPT, Perplexity, Claude, and Google AI Mode are already recommending products to buyers who never see a search results page. Most luxury stores give them nothing to work with — not because the products are wrong, but because nothing on the site is readable by a machine that is trying to recommend you.

  • FormatFixed-scope build
  • ThenOptional monitoring
  • RunsStandalone or inside a build
Book a discovery call

What is actually changing

Search is shifting from a list of links you compete for to a single answer you are either part of or absent from. The mechanics of being found are different, and most ecommerce sites were never built for them.

  • The answer replaces the results page

    When an AI assistant recommends three products, there is no page two. You are cited or you are not.

  • Machines read differently than crawlers

    Traditional SEO optimises for ranking signals. AI systems parse for structure, entities, and unambiguous factual claims.

  • Your reviews and specs matter more

    Assistants synthesise from what they can verify. Thin product data and scattered claims give them nothing to work with.

  • The answer comes before the citation

    The best available research suggests assistants choose which brands to name from what they already know, then look for sources to support the choice. Your site decides whether you are the source they can use — not whether you get considered.

  • It is early, which is the point

    The brands establishing entity clarity now are the ones being cited later. This is a compounding position, not a switch.

What this work does not do

The single strongest signal in the published research is not something that lives on your website, and we would rather say so on this page than discover it with you in month four.

  • It does not make you famous

    Being named in an answer tracks most closely with being independently talked about — reviews, video, press, community. That is earned across the web, not built in your theme.

  • We do not run PR or outreach

    We have no in-house media relations team and will not pretend this engagement buys one. We will tell you it is the bigger lever and point you at people who do it properly.

  • It is necessary, not sufficient

    When an assistant does reach for a source on a product question, it needs verifiable specifications. Thin data means it cites a competitor instead. That part is ours, and it is worth doing.

  • Beware anyone claiming otherwise

    The tactics most loudly sold in this category — llms.txt, piling on schema — have been tested and did not move citations. Ask any agency for the evidence before you pay for it.

How we scope engagements

What we actually build

This is engineering work with an end date, not a content retainer. Most of it happens in your theme, your structured data, and your product data model — and most of it is done once, properly.

  • Product data depth

    Specifications, materials, sizing, and use cases written to be parsed and cited rather than skimmed. This is where the evidence says citations are actually won.

  • Entity architecture

    Schema wired properly for rich results, and the harder work of making your brand resolve as a known entity rather than a guess. We do not expand markup for its own sake — more schema does not buy citations.

  • Crawlability for AI agents

    Robots directives, render paths, and performance work so agents can actually reach and read your pages.

  • Answer-shaped content

    The questions buyers ask assistants, answered on your own domain in a form that can be quoted.

  • Citation monitoring, if you want it

    Tracking where and how you surface across assistants after the build ships. Priced as monitoring, because that is what it is — not dressed up as an ongoing growth programme.

  • Hygiene, priced as hygiene

    Things like llms.txt cost nothing and we ship them — but the evidence says they do not move citations, so we will never bill hours for them.

How we measure it

AI search has no rank tracker and anyone selling you one is overstating what exists. Here is what we actually track, and what we do not claim to.

  • Citation presence

    Whether your brand and products appear in assistant answers for the prompts your buyers actually use.

  • Share of answer

    How often you appear relative to the competitors showing up in the same responses.

  • Referral traffic from assistants

    Sessions arriving from AI surfaces, instrumented properly so they are not misattributed to direct.

  • What we will not promise

    We cannot guarantee placement in any assistant. These are third-party systems with undisclosed, changing ranking behaviour. What we control is whether your site gives them something to work with.

Who this is for

  • A good fit

    Ecommerce brands with real product depth and a category where buyers ask questions before purchasing.

  • Also a good fit

    Brands mid-replatform or mid-rebuild, where structured data and entity work can be built in rather than retrofitted.

  • Probably too early

    Stores with thin catalogues or no organic search presence at all — foundational SEO comes first, and we will say so.

AI Search / GEO Build

A fixed-scope engagement, quoted against catalogue size and how much structural work your site needs. It has a defined end, and monitoring afterwards is optional rather than assumed. Runs standalone, or as a workstream inside a wider build.

Book a discovery call

Not sure whether this applies to you? The audit sprint includes an AI-search readiness review, so you find out before committing to anything.

Questions
  • There is overlap — clean structure and fast pages help both. But the objective is different. SEO competes for a position on a results page. This work makes your catalogue legible to systems that synthesise an answer and cite a handful of sources. In practice most of the effort goes into product-data depth and entity clarity rather than link building or keyword targeting.
  • No, and the overlap is shrinking. Even a first-position ranking appears in the AI answer only around half the time, and the share of AI-answer citations drawn from top-ten results has fallen from roughly three-quarters to under forty percent in about a year. Meanwhile, when an AI answer shows up, clicks to the ordinary results below it drop by more than half — unless you are one of the sources the answer cites, which measurably recovers them. Ranking and being cited are different games, and only one of them is growing.
  • The best evidence says no. The largest study to date — around 300,000 domains — found no correlation between llms.txt and AI citations. We still ship one, because it costs nothing and forces the useful exercise of stating clearly what your site is, but we will not bill hours for it. The work that moves citations is product-data depth, entity clarity, and crawlability. Anyone telling you a single file fixes AI visibility is selling you something.
  • Structural work lands immediately; citation behaviour moves slowly and unevenly across assistants. Expect to be reading directional signal at around three months, not week two. This is a compounding position rather than a campaign.
  • No, and we will not pretend otherwise. These are third-party systems with undisclosed ranking behaviour that changes without notice. We control what your site makes available to them, and we measure what happens. Anyone guaranteeing placement is guessing.
  • No, though it is where we are deepest — we have built in the Shopify ecosystem for seven years and ship our own app on the App Store. The structural work applies to any platform we can access the templates of.
  • Yes. It runs standalone as its own fixed-scope build. It tends to work better alongside a wider build, because a lot of the fixes live in the theme and product data — but it does not require one.
  • Because the work is mostly finite and we would rather say so. The product data model, the schema, the render paths, the answer content — those get built once and then they are built. Charging monthly for a job that finishes creates pressure to invent work, which is exactly the dynamic we think is wrong with this market. If you want ongoing citation monitoring afterwards, that is available and priced as monitoring.
  • Because the two are not alternatives. The research suggests assistants pick which brands to name largely from what they already know, then retrieve sources to support it. Nothing on your site fixes the first half — that is earned through being talked about, and we will say so plainly. But the second half decides whether the source they cite is you or the competitor with better product data. That half is engineering, it is genuinely underdone on most stores, and it is ours.