Faceted navigation: let shoppers filter everything, let Google index almost none of it
Count your URLs before you argue about them
Pull the parameter URLs from your server logs or a crawl and put a number on the problem. Stores are routinely surprised: a catalogue of a few thousand products wearing a filter space in the hundreds of thousands, with the crawler spending more time inside filter combinations than on product pages. Crawl budget stops being an abstraction at roughly a thousand SKUs, when the permutations start eating the crawling your revenue pages need.
The tell in Search Console is coverage bloat: huge counts of alternate or duplicate URLs, and important new products taking days to get crawled while filter URLs refresh constantly. If Google is spending its visits rechecking black plus size nine plus under fifty dollars sorted by rating, it is not spending them finding the product you launched this morning.
The two list decision
List one is short: filter combinations with real search demand, stable inventory, and a reason to exist as a landing page. Leather sofas, running shoes for wide feet, dresses in a fabric people search by name. Those earn a permanent URL, unique copy, and internal links, and they are built deliberately, not generated by the filter machinery and left to fate.
Everything else is list two: usable, noindexed, and canonicalised to the parent category or the nearest list one page. Noindex beats a robots block for pages users should still reach, because a blocked page can still be indexed from links while its content stays unexamined, whereas noindex lets the crawler read the instruction and move on. Empty combinations, filters that return no products, should return a 404 rather than a smiling empty page with a 200 status.
Sort order is a facet too
Shoppers sort; crawlers follow. Each sort state minting its own indexed URL splits the same product set across half a dozen addresses that differ only in order. Sort parameters belong firmly in list two, always. The same goes for session and tracking parameters that leak into internal links: every parameter that reaches an href is a page you are asking a crawler to consider, whether you meant to ask or not.
Watch what your own site links to. Internal links are instructions. If collection pages link their filtered states in menus, promotional blocks, or recently viewed modules, you are publishing the filter space yourself and no robots file will talk Google out of reading it. The fix is usually in the templates: link the clean category and the handful of earned landing pages, and let filters live in the interface where they belong.
A combination with three products is a thin page; the same combination after a range expansion might be your best landing page next season. Review list one quarterly against inventory depth and search demand, promote combinations that earned it, and demote pages whose inventory collapsed, because a thin indexed facet is worse than none: it ranks for nothing, converts nobody, and testifies that the store does not curate its own search presence.
Review the lists when the catalogue changes
Measure the win in crawl stats before rankings: crawl share going to product and category pages should rise, and new product discovery time should fall. Those are the mechanics you actually fixed. Ranking movement follows mechanics with a lag, and teams that demand ranking proof in week one abandon a correct fix before it has finished working.
Keep reading: The zero results page: search demand your store collected and then dropped, New arrival badges and the 30 day rule: a label that means something or nothing.
Frequently asked questions
Why is faceted navigation a search problem if shoppers love it?
Because each filter combination can become a separate crawlable URL, so a small catalogue can present hundreds of thousands of near duplicate pages. They split link equity and consume crawl budget that product pages need, a pattern Google itself flags as a cause of overcrawling. The answer is not removing filters. It is controlling which combinations get indexed.
Should filter pages be blocked in robots.txt or noindexed?
Usually noindexed. A robots block stops crawling, but a blocked URL can still be indexed if linked elsewhere, and the crawler never sees any instruction on the page. Noindex lets the page be crawled, read, and dropped from the index while shoppers keep using the filter normally. Reserve robots blocks for URL spaces you never want crawled at all.
Which filter combinations deserve their own landing page?
The few with genuine search demand, stable and sufficient inventory, and a permanent purpose: the combinations people type into Google, not just click inside your store. Those get a clean URL, unique copy, and internal links. Everything else stays a browsing state. Review the list quarterly, because inventory changes move combinations between the two lists.
What should an empty filter combination return?
A 404. A combination that cannot exist or has no products is not a page, and serving it as a normal 200 creates soft empty listings that waste crawl and confuse the index. That differs from a real category temporarily out of stock, which is a valid page with an inventory problem, not an invalid URL.