642,000 SKUs later, every catalog has the same seven problems

Field notes from the number on our homepage, and what it actually bought us.

Every store believes its catalog is a unique mess. For our first few projects, we believed it too.

Each one arrived with a backstory. The CSV export nobody dared touch. The Magento migration that "mostly worked". The VA who left and took the naming convention with them. We took notes like detectives, convinced every scene was a new crime.

642,000+ SKUs later, the notes read like one page, photocopied. The same seven problems, in every catalog we opened, regardless of vertical, team size, or how confident the founder sounded on the kickoff call.

The seven

The empty category. Shopify maintains an open source Standard Product Taxonomy, versioned like the API and spanning 25+ verticals. The category field feeds tax calculation, sales channel eligibility, category metafields, and what Shopify Magic can suggest. In the wild it sits blank, or wrong. We have met lipstick filed under Home & Garden.

The description that is secretly a database. Fabric, dimensions, care instructions, compatibility, all buried in one HTML paragraph. Shopify has metafields for exactly this: typed key-value fields attached to any product. The data exists. Machines just can't reach it.

Variants doing someone else's job. Sizes published as separate products so an old app stops complaining. Colors split apart so a collection page looks fuller. The API will accept 2,048 variants in a single bulk operation; what it cannot do is guess which duplicate of "Blue Denim Jacket" is the real one.

The silent images. Alt text on product media: 512 characters available, 125 recommended, zero written. Screen readers and image search read the same field. Empty is empty for both audiences.

Titles written for one reader. Google's product data spec gives a title 150 characters and disapproves listings with missing variant details or data that contradicts the landing page. Most titles we open were written for the founder's own memory, and the founder already knows what the product is.

The twins. The same physical product living twice: once from the old import, once rebuilt by hand, a different price on each. Neither gets deleted, because nobody remembers which one the bestseller collection points at.

Nobody's job. The seventh problem is the one that manufactures the other six. Ask who owns product data and the room goes quiet. Marketing owns the words. Ops owns the stock. The fields in between belong to whoever touched them last.

The day the notebook became a checklist

Our working theory was that we were in the forensics business: every catalog a fresh crime scene, every defect a local story. We were wrong in a useful way (we billed ourselves as detectives and turned out to be a checklist).

When every patient presents the same seven symptoms, the interesting fact is no longer the patient. Something structural is producing them.

Our read: a catalog is a database that everyone edits and nobody administers.

Look at what Shopify has quietly built. A standardized taxonomy on a quarterly release cycle, currently at version 2026-05. Typed metafields and metaobjects for custom schema. Category attributes that map the right fields to each product type. That is database tooling. The platform has been treating product data as a data model for a while now.

Teams still staff it as housekeeping. A cleanup sprint before Black Friday, a VA with admin access and good intentions. The tools assume an administrator; the org chart doesn't have one. Every symptom on our list, schema drift, orphaned records, untyped data, duplicate keys, is what any database does when it has users but no administrator.

What the repetition is worth

The consistency turned out to be the product. We stopped auditing catalogs for surprises and started auditing them against the seven, which is faster and considerably humbler. We run 147 production workflows, and an uncomfortable share of them are the same seven fixes wearing different SKUs.

To be fair about the line: if one careful person can still read every product in your store, attention plus Shopify's native tooling will hold, and you should not hire us. The seven turn structural at volume, not before.

What we'd tape to your wall:

• If a fact about a product matters, it needs a field, not a paragraph.
• The category field is load-bearing. Set it like it decides your tax treatment, because it does.
• A catalog without an owner degrades by default. Appoint one, human or system, before the twins arrive.

Your catalog is a special mess. So was the last one we opened, and the several hundred thousand SKUs before that.

Seven problems. Every time.

If you'd rather know which ones your store is running before they surface somewhere expensive, OKART's Infrastructure Stress-Test reads your catalog end to end, read-only, in 48-72 hours.

Sources

1. Shopify's Standard Product Taxonomy: what the category field controls: tax, sales channels, category metafields, and Shopify Magic suggestions.
2. Shopify/product-taxonomy on GitHub: open source taxonomy under MIT license, 25+ verticals, quarterly releases, current version 2026-05.
3. The new product model: products, options, variants hierarchy; productVariantsBulkCreate accepts up to 2,048 variants per operation.
4. Metafields and metaobjects: typed key-value custom data on products and other Shopify resources.
5. Google Merchant Center product data specification: required attributes, 150-character title limit, disapprovals for missing or conflicting product data.
6. Adding alt text to product media: 512-character maximum, 125 recommended, accessibility and SEO roles of alt text.