Discussion
  • Indexing
  • Site Architecture
  • Technical SEO

When filter combinations have real search demand, where do you draw the faceted indexation line: two facets, a demand threshold, or something else?

Published 1 reply

Discussion starter: a technical SEO question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • admin
    Admin
    Original poster

    I would not draw the line at a facet count. “Maximum two facets” is a useful guardrail for crawl control, but it is a poor indexation rule. Some three-facet combinations match real queries with plenty of stock (“men’s wide-fit waterproof hiking boots”), while some single-facet pages have no demand and no distinct result set (“size 9”, “sort by price”). The line I would use is an allowlist. A combination earns an indexable URL when it passes a demand threshold set by commercial value, has enough stable inventory, returns a result set meaningfully different from its parent, and matches a query where Google ranks listing pages. Everything else stays out of the index and, on large sites, out of the crawl path.

    The main qualification is scale. On a catalogue of a few thousand products, crawling is rarely the constraint, and the bigger risk is thin pages cannibalising their parent categories. On a marketplace with hundreds of thousands of products and dozens of filters, uncontrolled facet URLs can absorb a large share of crawl activity, so crawl control has to be settled before indexation.

    Two decisions that usually get treated as one

    Crawl control asks whether Googlebot should spend requests on a URL pattern at all. Index selection asks which of the reachable URLs deserve to rank. Google’s documentation on faceted navigation (last updated December 2025) frames the problem mainly in crawl terms. It says crawlers “will typically access a very large number of faceted navigation URLs before the crawlers’ processes determine the URLs are in fact useless”, which slows discovery of new, useful URLs.

    For facets you do not need indexed, Google suggests disallowing them in robots.txt or using URL fragments for filters, and describes rel=”canonical” and rel=”nofollow” as less effective alternatives. For facets you do want indexed, it recommends the standard & parameter separator, a consistent filter order with no duplicate filters when filters live in the path, and an HTTP 404 when a filter combination returns no results.

    A two-facet cap mostly solves the crawl problem, because it limits how many URLs can exist. It says nothing about which of the surviving URLs deserve to be indexed.

    Why facet depth is a proxy, not the rule

    Depth caps are popular because combinations grow fast. Take a hypothetical category with 8 filterable attributes, each with 10 values. Single-facet URLs number 80. Two-facet combinations number 2,800 (28 attribute pairs × 100 value pairs). Three-facet combinations number 56,000 (56 attribute triples × 1,000 value triples). If the URL order is not normalised, every two-facet combination can exist in two orders and every three-facet combination in six.

    Demand does not follow that structure. It follows how people describe what they want. Some attributes work as query modifiers: brand, product type, material, use case, gender, fit, and location for services and property. Others are browsing conveniences with little standalone demand: price bands, in-stock toggles, ratings, sort orders, and size in most apparel categories. A two-facet rule treats “brand + type” and “colour + price band” as equals. They are not.

    Important distinction:

    Facet count predicts how many URLs you will create. Demand, inventory and distinctness predict whether a URL deserves to rank. Use the first to control crawl growth and the second to decide indexation.

    The four tests I would apply to each candidate combination

    Test Question Evidence to inspect Typical failure
    Demand, weighted by value Would the page earn enough qualified clicks to justify existing? Keyword tools, Search Console queries on the parent category that contain the modifier, internal site search logs, paid search terms reports Modifiers with volume but weak purchase intent
    Inventory Does it reliably return enough products to be a useful page? Product counts over the past 6 to 12 months, not today’s snapshot Seasonal or long-tail combinations that regularly empty out
    Distinctness Does its result set differ meaningfully from its parent and siblings? Overlap of product IDs between the combination and its parent “Black leather boots” when nearly all leather boots are black
    SERP fit Does Google rank category or listing pages for this query? Manual review of the results page for the head query Results dominated by product pages, editorial guides or marketplaces

    Two further checks are practical rather than strategic. Can you give the page its own title, H1, some introductory copy and deliberate internal links? And will merchandising keep the combination meaningful, or will product tagging drift until the page stops matching its name?

    Setting the demand threshold

    There is no universal number, and I would be wary of one borrowed from another site. A workable threshold compares estimated value with cost. Value is roughly monthly searches × expected CTR at a realistic position × conversion rate × average order value or margin. Cost includes crawl share, content and maintenance, and the risk of splitting signals with the parent category. A site with strong category rankings can justify a lower threshold because its new pages are more likely to rank. A newer site should set it higher.

    Look at demand across the cluster, not the single phrasing. “Wide fit hiking boots”, “hiking boots wide feet” and “wide hiking boots mens” may each look small in a keyword tool, but together they justify one page. Paid search terms reports are underused here, because they show real queries and which of them converted, not modelled volume.

    Hypothetical example:

    An outdoor retailer has 12,000 products and 14 filterable attributes. Keyword research and site search logs produce 900 candidate combinations. Applying the tests: 610 fail on demand, 120 fail on inventory (fewer products than the retailer’s chosen floor for most of the year), 70 fail on distinctness, and 30 fail on SERP fit because Google ranks buying guides for them. That leaves 70 allowlisted pages, a mix of one, two and three facets. A two-facet rule would have allowed thousands of URLs and still excluded some of the 70. These numbers are illustrative.

    For inventory, build in some tolerance so pages do not flip in and out of the index. For example, add a combination to the allowlist only after it has held your minimum product count for several weeks, and remove it only after it has stayed below that floor for longer. Constant flipping wastes crawl activity and makes the page’s ranking history useless. For combinations that permanently return nothing, follow Google’s 404 recommendation. For a seasonal allowlisted page you plan to restock, keeping the URL live with related products is a defensible judgement call, but a page that shows zero products for months risks being reported as a “Soft 404”.

    How I would implement the line

    1. Build candidates from demand data, not from the facet tree. Start with queries, then map them to combinations.
    2. Score each candidate on the four tests and record the result in an allowlist jointly owned by SEO and merchandising.
    3. Give allowlisted combinations clean, stable URLs with a fixed attribute order, self-referencing canonicals, their own titles and H1s, and a place in the XML sitemap. Putting them on a different URL pattern from general filter URLs (for example a path rather than parameters) makes crawl rules much easier to write.
    4. Link to them on purpose from the parent category (“Shop by material”), related categories and relevant content, so they do not depend on the filter interface for discovery.
    5. Handle everything else by scale. On large sites, keep general filter URLs out of the crawl path with robots.txt rules or non-crawlable filter interactions. On smaller sites, noindex or a canonical to the parent can be enough, accepting that a canonical is a hint Google may overrule.
    6. Return a 404 for empty combinations, as Google recommends.

    Watch for this:

    Do not combine a robots.txt disallow with a noindex and expect the noindex to work. Google cannot see a noindex on a URL it is not allowed to crawl, and blocked URLs can still be indexed without content if enough links point to them. Pick one control per URL pattern, and test your robots.txt rules against allowlisted URLs so you do not block the pages you want indexed.

    Let Google’s response recalibrate the threshold

    Signal after launch Likely interpretation Action
    Many allowlisted URLs sit in “Crawled - currently not indexed” Google does not see enough distinct value Raise the distinctness or inventory bar, or merge weak pages
    “Duplicate, Google chose different canonical than user” on allowlisted pages Result sets are too close to the parent or a sibling Check product overlap and consolidate
    “Discovered - currently not indexed” on a large site Crawl prioritisation or weak internal links Check logs for crawl spent on non-allowlisted filters, and strengthen links
    Parent and child pages trade rankings for the same queries Cannibalisation Tighten the child’s targeting or remove it from the allowlist
    Indexed, but few impressions after 3 to 6 months Demand was overestimated Prune and lower confidence in that demand source

    Experienced practitioners legitimately disagree on how wide to go. Property, jobs and classifieds sites often index millions of location × attribute combinations, because that structure is how people search (“2 bedroom apartments in [suburb]”). They use templated rules with inventory floors rather than hand-built allowlists. Retailers with shallow catalogues usually need to be much stricter. Some teams also prefer separately curated collection pages over indexing filter URLs at all, which gives more editorial control at the cost of maintaining two versions of similar result sets.

    How to test it

    Roll the allowlist out to a subset of categories and keep comparable categories as a control. Measure non-brand clicks and revenue at the category-family level (parent plus allowlisted children together, so cannibalisation shows up), plus indexation rate and the share of Googlebot requests hitting non-allowlisted filter URLs. Allow 8 to 12 weeks, and watch for seasonality and inventory changes that hit test and control differently. If the family’s combined clicks and revenue do not grow, or the parent loses roughly what the children gain, the threshold is too low or the pages are too similar.

    My decision rule: use a depth cap only as a crawl safety net, and decide indexation per combination using value-weighted demand, stable inventory, distinct results and SERP fit. Then let Search Console’s indexing states and each page’s revenue tell you whether the line is in the right place, and move it on evidence rather than on facet count.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.