Manufacturer H — Four-Language Rebuild and a Full Catalogue Audit

Car-care chemicals is an industry outsiders rarely think about and insiders find brutally competitive. Shampoos, waxes, coatings, interior cleaners, polishing tools — each line carries dozens of variants, and all of it has to be sold to three quite different buyers at once: brands wanting to put their own label on it, distributors stocking a channel, and detailing shops buying to use. This client ran an own brand alongside OEM and ODM work, and those two sides do not make the same argument.

The challenge: audit first, design later

A product structure like that almost always ends up reflected in the website. There is simply too much of it, the site accretes over years, and nobody ever has a quiet month to go back and tidy up. So our first deliverable was not a design comp. It was an audit.

What the crawl found was worse than expected. There was no Open Graph markup anywhere, which means any time someone pasted a link into a chat or a social post it arrived as a bare URL with no preview image and no title. No canonical tags. No language declared on the HTML element. Not a single h1 on the homepage. Image alt coverage sat below forty percent.

The genuinely damaging finding was that the English site barely existed. The English sitemap returned a 404, and not one of the URLs the sitemap did list was an English page. More than a dozen Chinese pages had no English counterpart at all. For a manufacturer whose business is export, that means being effectively unfindable in the language its buyers actually use.

Other problems sat deeper. Structured data covered only about a quarter of product pages, its image URLs pointed at a staging domain that had long since died, and it carried a few empty placeholder reviews — the kind of thing that, once a search engine reads it, damages trust across the whole domain. The brand colour failed contrast requirements against dark backgrounds. Login and registration pages returned healthy status codes while rendering the product list: two phantom pages.

What we did: rebuilding the owned site

1. Take the whole legacy site down first

We captured 360 URLs with a 100% success rate, re-crawled the product pages to recover complete specifications, converted content pages into structured documents, and preserved every PDF, embedded video and form specification. There is no design in this step. The point is simply that nothing existing should quietly vanish in the move —

The common migration disaster is not building something badly. It is losing something and nobody noticing.

2. Four languages, including right-to-left

The new site is statically generated: 118 products across four languages gives 472 product pages, and with category pages the build lands around 660. The four locales are English, Traditional Chinese, Spanish and Arabic, and the last of those needs genuine right-to-left layout support. That is not something a translation plugin solves; the layout has to accommodate it from the start.

3. Verify every redirect, one at a time

The biggest risk in any move is losing rankings. We generated a redirect map from the old URL structure and then verified it entry by entry, confirming each one landed on a real page rather than a 404 or a chain of hops. It is tedious work and it cannot be skipped: every redirect you get wrong is a page whose accumulated authority you simply throw away.

4. Handover: make the client self-sufficient

We connected a CMS and, before importing anything for real, set an acceptance gate: export the data, import it back, compare the round trip at byte level, and only proceed once everything passed. In testing, publishing from the CMS rebuilt and deployed the site within two to three minutes, and retiring a few SKUs updated the sitemap automatically. Day-to-day updates no longer require an engineer.

We also wrote a batch image pipeline to trim whitespace and square off product photography, cutting the total payload from 58.10MB to 15.98MB. The script decides whether an image has already been processed by its dimensions rather than its area, so repeated runs never compound.

What we did: auditing the marketplace storefront

For the same client we also ran a full-catalogue diagnostic on their Alibaba storefront. Across more than two thousand listings, the keyword fields were almost universally stuffed — long strings of loosely related terms crammed into a single field. That might have worked a decade ago. Today platforms treat it as abuse, and frankly it was never going to attract the right buyer anyway.

The more interesting finding was duplicate listing. A single model family had been published as more than three hundred separate listings, of which fewer than twenty drew any traffic, and the family as a whole had produced one inquiry. It comes from a reasonable-sounding assumption: more listings, more chances to be seen. What actually happens is that the same products dilute one another in search results and the platform cannot tell which one to favour. So we set a rule — no more than ten live listings per model family — relisted the selected products across two rounds, and rewrote titles that had been left far too short.

Two further problems were quietly eroding the store. One group of listings carried contact details in the product description, which the platform treats as a violation. And a store-wide keyword contamination had car-shampoo terms attached to brushes and clay bars, which distorts how the platform classifies the entire catalogue.

Keyword rewriting followed an add-only rule, leaving untouched any term already earning traffic. On a platform whose full ranking logic you cannot see, not breaking what already works matters more than chasing the optimal answer. Every write was read back in batches to confirm the platform had actually accepted it.

Results

The owned site went from a legacy build with no canonical tags, no Open Graph and effectively no English presence, to a four-language catalogue of 472 product pages, with around three hundred redirects individually verified and the product image payload cut to roughly a quarter of its original size — all maintained by the client themselves. On the marketplace side, the full catalogue has been audited and a relisting rule established.