Catalogue SEO: category architecture, facets and variants

Where the traffic actually is: Category and facet pages carry the non-brand demand; product pages win queries containing a brand and a model number and very little else. The last three catalogues we took through this ended with between 62 and 78 percent of previously indexable URLs removed from the index, and non-brand category traffic up in each case within two quarters. Nothing was deleted from the shop. What changed was which URLs were allowed to ask Google for attention. Facets: The facet decision is the whole engagement in miniature. A facet becomes an indexable landing page when the combination has demand and stock behind it — a category plus one attribute, where people actually search that way. Two or more attributes combined, price bands, sort orders and view toggles never do. We look at the server logs before deciding anything, because the argument between robots.txt disallow, noindex and canonical is usually settled by what Googlebot is already spending its requests on. Each of the three has a different cost: disallow keeps the crawler out but strands any equity pointed at those URLs, noindex needs the page crawled to work, and canonical is a hint that a large enough contradiction will overrule. We pick per pattern and write down why, so the next developer does not undo it. Variants: Collapse to a parent unless the variant has independent demand, and in beauty that line is unusually clear. Shade names have no search volume and belong on one product URL with a selector. Active ingredient concentrations do — someone searching a serum at fifteen percent means it, and that variant deserves its own page with its own copy. Apparel sizes collapse; colours occasionally do not. We decide it per attribute against real query data, not by platform default. Product data and the law it sits under: Structured data has to agree with the page and with your Merchant Center feed. Price including BTW, marked as tax-inclusive, matching what the customer sees. Availability that is true at render time. Where a promotion runs, the displayed prior price must be the lowest applied in the preceding thirty days under the Omnibus rules, and your markup must not tell a different story than your promotional badge. Feed and page disagreeing on price or stock is a common and expensive error, since it costs the free listing as well as the paid one. Ratings get marked up only from real reviews collected on your own product pages. Copy, and the claims we will not write: In cosmetics we write to what your safety dossier substantiates and no further. If keyword research says the market searches for a barrier-repair claim and your dossier supports helping to maintain, the dossier wins, under Regulation 1223/2009 and the common criteria that govern claims. We will not argue that boundary back with your regulatory lead, and we would rather lose the query than hand you copy the NVWA can take apart. We also do not write blocks of text under the fold to hold keywords. Category copy is buying guidance placed where somebody choosing between two products needs it, or it does not go on the page. Out of scope: Writing product descriptions at volume. Generating category pages for keyword permutations with no stock behind them. Paid shopping management, feed operations, and CRO testing. Platform development: we specify, your team or your agency builds it on Shopware, Magento, Shopify or Lightspeed, and we review the output. Not for: Catalogues below a few hundred SKUs, where the useful work is about a day of your own developer's time and we will tell you which day. Sellers operating only on marketplaces, with no storefront of their own. And any team that wants the index shrunk without a stakeholder willing to defend the deleted URLs in a meeting, because that meeting always happens.