SEO & AEO Optimization

Something new is reading your store

For twenty years the job was to be found. A person typed a query, a list of links appeared, and the work was to be high enough on that list to be clicked.

That is still most of the traffic and it will be for a while. But a growing share of buying research now never produces a list of links at all. Someone asks ChatGPT which agency builds Shopify headless storefronts. Someone asks Perplexity to compare two platforms. Someone reads Google’s AI overview and never scrolls. In each case a machine has read a set of sources, decided which ones to trust, and produced an answer that names two or three companies. There is no second page of results to be on. You are named, or you are not in the conversation.

The useful thing about this shift is that the work it rewards is largely work that was already worth doing. Systems that summarise cite sources they can parse and entities they can resolve. That means structured data, clear identity, self-contained passages and server-rendered HTML — the same list that has been sitting at the bottom of technical SEO audits for a decade, ignored because it was not urgent.

It is urgent now, and almost nobody in ecommerce services is treating it that way. That is the whole reason this service exists.

Start with whether SEO is your problem

Before any of that, the diagnosis, because it changes the engagement more often than any other step.

A site can have real organic traffic, healthy engagement, and effectively no leads. The instinct is always that the site fails to convert. Frequently it does not. The query set is simply wrong.

Here is the pattern, and we see it constantly. A business publishes articles for two years. The articles rank. The rankings sit on informational queries — “is dropshipping profitable”, “which platform should I use”, “how do I start a store”. The people searching those things are researchers, students and people who do not yet have a business. Average time on page is a minute, which reads like failure and is actually success: the reader got their answer and left, exactly as intended. There was never a purchase in that session to lose.

Meanwhile the entire commercial surface of the site is one page, listing what the company does, with no page beneath it that any buying query could possibly land on. There is no URL that can compete for “shopify app development agency” because no such URL exists.

No amount of engagement optimisation fixes that. The fix is to build the pages that buying queries can land on, link them properly, and stop measuring the blog against a conversion goal it was never going to serve. That is an information architecture job wearing an SEO job’s clothes, and diagnosing it correctly is worth more than a year of retainer.

The technical layer

Nothing else works until a page can be reached, rendered and understood.

Crawl and index coverage. What is indexed that should not be, what is not indexed that should be, and why. On ecommerce the usual culprits are faceted navigation generating thousands of near-duplicate URLs, and canonical tags that point somewhere well-intentioned and wrong.

Faceted navigation and pagination. Which filter combinations deserve an indexable URL, which should be canonicalised away, and which should not be crawlable at all. Getting this wrong is the single most common way an ecommerce site wastes its own crawl budget.

Internal linking. Which pages receive links, from where, with what anchor text. A page nothing links to cannot rank however good it is, and on most stores the service and category pages that matter commercially are the ones nothing links to.

Core Web Vitals. Measured on field data, not lab scores. LCP under 2.5 seconds, INP under 200 milliseconds, CLS under 0.1. Speed is not a large ranking factor and it is a very large conversion factor, which is a better reason to fix it anyway.

Entities, and why one identifier matters

Most sites that have structured data have it as scattered blobs — a Product node here, an Organization node there, each written by a different plugin, none of them referring to each other. Technically valid, and close to useless for anything trying to build a model of who you are.

The version that works is a graph. One Organization node with a stable identifier, emitted once site-wide. Every Service, Article, Product and BreadcrumbList on the site points its provider or publisher at that identifier instead of repeating an inline copy. sameAs links it to every real profile the business has — the social accounts, the Shopify Partner listing, the directory entries.

That single resolvable entity is what lets a model connect the name of your company to what it does with any confidence. sameAs corroboration is the main signal available to a brand that is not large enough to have a Wikipedia entry, and it is free.

The rule underneath all of it: schema is generated from the content that renders. Never hand-written. Hand-written JSON-LD drifts away from the page within about two content updates, and a mismatch between markup and visible content is both a maintenance trap and a manual action waiting to happen.

Answer units

Language models quote passages, not pages. That has a direct implication for how a page should be written.

Every commercially important page needs a handful of real questions with answers that stand alone — that make complete sense when lifted out and pasted into someone else’s answer, with no “as we mentioned above” and no dependency on the paragraph before them. Specific answers with numbers, dates and named constraints get quoted. Hedged answers do not, because there is nothing in them to quote.

Those same questions render as visible content on the page and emit as FAQPage structured data from the identical source. One set of facts, two consumers.

The questions themselves come from your buyers, not from a keyword tool. What they ask on the first call. What they ask twice because the first answer was not clear. What they ask when they are about to say no. Those are the passages worth writing, and they are usually already sitting in someone’s sent folder.

Measuring something that has no console

Search has Search Console. Answer engines have nothing, and pretending otherwise is how this service gets sold dishonestly.

Two measurements are available and both are real. First, referral traffic from chatgpt.com, perplexity.ai, claude.ai and gemini.google.com in your analytics — small numbers today, growing, and disproportionately qualified when they arrive. Second, a fixed set of prompts re-run on a schedule: the questions a buyer would actually ask, run monthly against the major systems, recording whether you were named and what was said about you.

That log is not a vanity metric. It is the most persuasive artefact we produce, and it is also the honest one — it will show you months where nothing improved.

And this site

You should ask anyone selling this whether their own site passes. This one carries the markup we would write for you: one Organization entity referenced site-wide, Service and FAQPage data generated from the same content that renders on the page, breadcrumbs on every nested route, and server-rendered HTML on every route so that everything reading it sees a complete document.

Check it. Selling answer-engine optimization from a page with no structured data on it is self-refuting, and the sort of buyer worth having will look.