Shopify SEO consultant or agency: what the platform already does, and what it can't

What it leaves to you

Every Shopify ranking problem I have seen lives in one of six places, and the platform is innocent of all of them.

Which collections exist, and what links to them

A collection is a page. Shopify will make one for every tag, every vendor, every product type and every filter someone clicks, and the theme decides which of those pages get linked, which get indexed, and which are left for a crawler with time on its hands to discover. The policy has to be written before anything else: which collections deserve to be pages, what their canonical is, what the sitemap says, what the theme links to. On a WooCommerce tile retailer, 196 pages that should never have been public were sitting in the sitemap as a shadow site of duplicates. Shopify builds that kind of surface differently, through tags and filters rather than reusable blocks, and it builds it just as readily.

Filters and facets

Stack two filters and the platform's robots file blocks the result, which is sensible. It also means the single-filter pages, the ones that could rank for a colour or a size, are left to your theme and your judgement. Faceted navigation needs the same written policy it needs on every catalogue: which combinations are pages, which are parameters, which are noindexed, and what each canonical points to. Then the templates, the sitemap and the internal links are made to agree with it. Urban Outfitters' estate, where I led the technical SEO campaign for architecture, crawl and indexation, is where I learned to write that policy down before touching a template.

Apps, and the weight they add

Every app is a script, and most of them load on every page whether the page needs them or not. Speed problems on Shopify are rarely the theme and usually the app list. The fix is subtraction, which no app will sell you. I audit what loads, what each script is for, and what breaks when it is removed. Then it is removed.

What the machines actually read

The product data has to be in the HTML the server sends. A theme or an app that renders the price, the reviews or the variant table after the page loads is invisible to most of the AI crawlers and late for Google. Two structured-data graphs that disagree, one from the theme and one from a reviews app, are worse than one. The Agent-Ready Grader on this site tests extractability as one of its three gates, and it is free, so run your best product page through it before you pay anyone.

Traffic that isn't traffic

Analytics on a storefront can be wrong in ways that cost real money. Mowers Online's August showed roughly 45% of sessions coming from a headless-Chrome farm that never bought anything; subtract it and the store's real visitors were converting at close to double the blended figure. A forty-minute probe at the origin counted 1,596 farm hits coming through the edge against 28 arriving direct, which is the kind of reading a platform's own dashboard cannot give you. On Shopify the edge is Shopify's, so the levers are different, and the first step is the same: find out whether the traffic is human before you spend anything chasing conversion.

Migrations, in and out

Shopify's URL prefixes are fixed, so a move onto the platform changes almost every URL you have, and a move off it changes them back. The redirect table takes one old path to one new path, uploaded in bulk, with no pattern rules; the map has to be generated and tested somewhere else and then loaded. The method is the one in my migration files: freeze the inventory first, rank the map by the URLs that earn links and clicks, test every row. Trade2Sync's 1,687-URL map was fetched row by row through the tested config, 1,687 of 1,687 landing single-hop; a UK locksmith group's rebuild cut over on 238 of 238. Trade2Sync is also the file that keeps me honest about blame. Its collapse began seven months before the migration everyone was blaming, and the Search Console exports said so once someone read them side by side.

International

Markets gives you localised paths and generates the hreflang for them, which is more than most platforms do. It does not decide whether a second market needs its own content, whether the currency switch is minting duplicate pages, or whether the localised collections are thin. Those are architecture decisions, and they come before the toggle.

The agent layer Shopify just handed you

In May I wrote that the agentic commerce protocols would be abstracted away by the big platforms, and that a merchant on Shopify would inherit the agent layer without doing the integration. That has now happened at the storefront level. On a large Shopify store I fetched this week, the robots file opens with instructions for AI agents, points them to an agents.md file and to a Universal Commerce Protocol endpoint on the store's own domain, and tells them that checkouts are for humans and payment needs the buyer's approval. The sitemap index carries an agentic-discovery sitemap beside the products and collections. An older store I checked still had the previous default, so this is arriving store by store rather than everywhere at once.

What that buys you is eligibility: an agent shopping on a buyer's behalf can read your catalogue, build a basket and start a checkout through a protocol instead of scraping your theme. What it does not buy you is being chosen. Discovery is still earned on the page, by the product data in the HTML, the collection that answers the query, and the structured data that agrees with the copy. The platform has done the plumbing. What comes through it is still yours to decide.

Consultant, agency or app

Three different things are for sale under "Shopify SEO", and they are not substitutes.

A Shopify SEO agency sells capacity: content every month, link outreach, reporting, and a team to run it. If the store's architecture is right and what it lacks is volume, that is a sensible buy. If the architecture is wrong, the agency's content lands on pages the policy should have retired.

An app sells rules at scale: titles and descriptions across thousands of products, alt text, broken links, a score. I know the shape well because I built one for a rival hosted platform. SEOflare runs on the ShopWired app store. Its first production-scale install lost 1,647 of 1,752 pages to the platform's CDN firewall in about thirty seconds, and the pacer that came out of that incident now crawls the same store in 13.6 minutes with zero 403s, against a 31.9-minute baseline. Its generated metadata passes two gates or does not ship: deterministic validators first, then claim-grounding. That is what an app is good at. It cannot decide which collections exist, it cannot remove the other apps, and it cannot tell you what your traffic really is.

A consultant sells judgement and, in my case, the implementation as well. The diagnosis from the instruments, the architecture written down and agreed, the fixes shipped in the theme, the settings, the redirect table and the app list, and the readouts afterwards from the same instruments. One person. If what you need is capacity or a score, I will tell you so and point you at the right kind of supplier.

Four checks before you hire anyone

1. Search Google for site: and your store's domain, and look at what comes back after the first page of results. Tag pages, vendor pages, filtered collections and the all-products collection are the usual surplus, and every one of them competes with a page you meant to rank.
2. Fetch your best-selling product page with JavaScript switched off, or run it through the grader on this site. If the price, the variants or the reviews are missing from what comes back, the machines are missing them too.
3. Open the network panel on any page and count the scripts that are not your theme. Then ask what each one is for.
4. Read your robots file. If it opens with instructions for agents, you have the new layer. If it does not, you are on the previous default, and the platform will get to you.

If the first two fail, the problem is structural, and no monthly retainer will reach it. Shopify has done the half that can be automated. The other half is a set of decisions about your own catalogue, and that is the work.