Rankelle

SEO for marketplaces

The site is mostly pages you did not write. Deciding which of them Google should see is the whole job, and it is a decision you can measure.

A marketplace's pages are generated by its users and its filters. Listings expire, sellers write their own titles, and every combination of category, location and attribute is a URL. The site is effectively infinite, the crawl budget is not, and the central question is which finite subset of pages you want the crawler to spend its visits on.

The pages that earn impressions are usually the category-plus-location landing pages — 'used bikes in Bristol' — not individual listings, which are short-lived and thin. So the strategy is to make those landing pages excellent and indexable, make the rest reachable but not indexed, and make expired listings leave the index cleanly instead of lingering as thin pages.

Measurement at this scale is easier than it looks, because the landing pages are hundreds of comparable units. Change the template for some, hold the rest, and the comparison is built in. What is hard is knowing the fix actually reached every page, which is why checking it landed is not optional here.

The findings that recur on this kind of site

  • Every filter combination indexableCategory × location × price × condition, each a 200 with a canonical to itself. The largest duplicate-content cause on any marketplace, and the one eating the crawl budget.
  • Expired listings returning 200'This item has sold' on a page that still asks to be indexed. Thin content in the tens of thousands, and the sitemap still lists them.
  • Seller-written titles with no structure'BIKE!!!' as the title tag. Titles too short, or missing where the seller left the field blank, on every listing rather than any one.
  • Landing pages reachable only through searchThe category-location pages that should earn everything, orphaned because the only way to reach them is to type in the site's own search box.
  • Redirect chains from old URL schemesThree generations of listing URL, each redirecting through the last, on links that are still in the wild.

What to change first, and what to compare it against

  1. Decide the indexable set, then enforce it in the templateCategory and location pages indexable; deeper facets noindexed or canonicalised; expired listings 410 after a grace period. One decision per page type, ranked by the impressions it protects.
  2. Treat landing pages as the experimental unitHundreds of comparable pages, easily matched by traffic band. Control-group testing was practically designed for a marketplace.
  3. Verify the fix reached every page before starting the clockA template change that deployed to one region's pages and not another's is a common failure. The sentinel re-checks the affected set and only marks the fix landed when the defect is gone from all of it.
  4. Expect the first verdict to be about crawl, not rankReducing the indexable set often shows first as more frequent crawling of the pages that matter and only later as impressions. The window is four weeks; the crawl effect may be visible sooner and is worth noting as 'so far'.

Questions from people running one

Should listings be indexed at all?

Usually some. A listing with substantial seller-written content and a long life — property, vehicles, jobs — can earn its own impressions. A listing that lasts three days and says 'bike, good condition' cannot, and indexing it costs crawl budget the landing pages need. Decide per category by how long listings live and how much is on them.

Can a crawler even cover a site this size?

Not all of it, and it does not need to. Rankelle crawls up to the page limit of your tier from the home page outward, which reaches the landing pages and a sample of listings — enough to identify every template-level cause. A cause found on a sample of listings applies to all of them; that is what a template means.

Everything above is the method, usable with any crawler and a Search Console export. Rankelle runs it on a schedule: the crawl, the grouping into causes, the watch for the fix landing, the frozen control set and the verdict four weeks on — including the verdicts that say the work did nothing. One site is free, permanently, because the first verdict takes six weeks to arrive.

Try it on your own site, free

The whole audit of every website you add, no card and no expiry. Pay only when you want the proof.