The problem
Google's faceted navigationFilters on a listing page that narrow results by attributes such as brand, size or price, often creating a URL for each combination. documentation describes the issue plainly: filtering and sorting generate many URL variations that show overlapping content, which can lead to the crawler spending its effort on duplicates, slow discovery of new content, and dilution across near-identical pages. For a category with a dozen facets, each with a few values, the combinations run into the millions, and any crawler that follows links will find them.
The symptoms in Search Console (GSCGoogle's search performance and indexing tool.): the Page Indexing report dominated by "duplicate without user-selected canonical", "crawled, currently not indexed" and "discovered, currently not indexed"; crawl stats showing most requests going to parameterised URLs; new products taking weeks to appear. The crawling and indexing guide explains the underlying behaviour and Crawl BudgetHow much a search engine crawls a site..
The decision framework
For each facet, and then for the combinations, ask four questions:
- Is there search demand for this view? "Women's waterproof hiking boots" is a query; "hiking boots sorted by price, forty per page" is not. The keyword research section provides the data; the keyword map records which URL owns each cluster.
- Does the view have enough products to be a useful page? A filter combination that returns two products is a thin page; one that returns a hundred is a category.
- Is the view distinct from its parent? A filter that removes a tenth of the products is a near duplicate of the category; one that selects a coherent sub-range is a different page.
- Can the view carry its own content? A title, heading, introduction and markup that describe this sub-range specifically. If the template can only swap a word into the parent's title, the view is a filter, not a page.
The outcome is three tiers:
| Tier | Example | Treatment |
|---|---|---|
| Indexable landing page | Category + one high-demand facet (brand, type, a key attribute); occasionally two | Clean URL (path or canonical parameter form), self-canonical, own content, in the sitemap, linked from the category and the navigation |
| Crawlable but not indexable | Combinations the crawler will encounter through filter links but that do not deserve ranking | Canonical to the nearest indexable page, or noindex where the content differs too much to canonicalise honestly |
| Neither crawled nor indexed | Deep combinations, sorts, page sizes, view modes, session or tracking parameters | Not linked with crawlable anchors (script-driven or nofollow), robots rules for the parameter patterns, and canonical as a safety net |
The first tier is typically a small number of combinations per category, chosen by demand and reviewed with the keyword map; it is never "every value of every facet".
Designing filter URLs
Google's guidance asks that the engine be able to tell distinct views apart and that URL structure be consistent. Rules that make that true:
- Indexable combinations get a page-like URL:
/hiking-boots/womens/or/hiking-boots/brand/acme/, with the facet order fixed. - Non-indexable filters use parameters in a fixed order with fixed value encoding, so that the same
selection always produces the same URL:
?colour=blue&size=9, never?size=9&colour=bluefor the same view. Alphabetical parameter ordering enforced by the platform removes a whole class of duplicates. - Sorts, page sizes and view modes either do not change the URL (handled client-side) or use parameters that are canonicalised away and excluded in robots.
- Session, tracking and internal search parameters never appear in links the crawler can follow.
- Empty results return a page with a clear message and a noindexA robots directive telling search engines not to show a page in their results., or redirect to the parent; never a soft 404 that looks like a page.
- Pagination keeps its own parameter and is handled per the category and product guide.
The mechanics
| Tool | What it does | Where it fits |
|---|---|---|
| Canonical tag (Canonical TagTells search engines the preferred URL.) | Suggests the preferred URL among duplicates; passes signals | Filter views that are subsets of an indexable page; Google treats it as a hint and may ignore it where the content differs a lot |
noindex meta | Keeps the page out of the index while allowing crawl | Filter views that are too different from any page to canonicalise honestly, such as combinations with distinct products but no demand |
| robots.txt rules | Prevents crawling of matching URL patterns | Sorts, deep combinations, parameters that explode; note blocked URLs can still be indexed if linked, and cannot pass signals |
Link attributes (rel="nofollow") on filter links | Asks the crawler not to follow | Deep filter links; a hint, not a block |
| Script-driven filtering without URL change | Keeps selections off the URL entirely | Filters that never need a URL; must still expose the indexable tier as real links |
| Sitemaps | Declares the indexable tier | Only tier-one URLs; never filter noise |
| Internal linking | Decides what the crawler finds first | Link tier-one pages from the category and navigation; do not link deep combinations in crawlable form |
Google's documentation retired the parameter-handling tool that once let site owners declare parameters in Search Console, so these on-site mechanics are the whole toolkit. Of them, nofollowA link attribute asking search engines not to treat the link as an editorial endorsement. on filter links is the weakest on its own, because search engines treat it as a hint rather than a block.
Implementing on a platform
Hosted platforms expose some of these controls and not others, and the platform audit decides what is possible: whether filter URLs can be given canonical tags to the collection, whether chosen combinations can be turned into real collection pages with their own content, whether sort parameters can be excluded in robots, whether filter links can be rendered as buttons rather than anchors. Where a platform cannot canonicalise or noindex filter views, the fallback is to keep filters off the URL and build the indexableWhether a URL is eligible to be added to a search engine's index: reachable, returning success, not excluded by noindex and not canonicalised elsewhere. tier as separate collections.
An outdoor retailer's boots category
The category has facets for gender, activity, waterproofing, brand, colour, size, price and sort. Query data shows demand for gender, activity and brand, and for the gender-plus-activity pair; nobody searches by colour or price band. The decision: build indexable pages for each gender, each activity, each major brand, and the gender-activity pairs with enough products, each with its own introduction and title, linked from the category and in the sitemap. Waterproofing, colour, size and price remain parameters in fixed order, canonicalised to the nearest indexable page. Sort and page size are handled without changing the URL. Robots rules exclude the parameters that remain. Within a few months the Page Indexing report's duplicate counts fall, crawl requests shift from parameters to products, and the new gender-activity pages begin to appear for the queries they were built for.
Internal search results pages
A store's own search generates URLs (/search?q=...) that are, to the engine, another faceted view: an
unbounded set of listing pages with overlapping content. Google's guidance has long advised keeping
internal search results pages out of the index. The treatment is the third tier: noindex on the search
template, the search parameter excluded in robots.txtA file at a site's root telling crawlers which URLs they may and may not fetch., and search links rendered as a form submission
rather than crawlable anchors. The one exception is a query so common that it deserves a real landing
page — in which case it becomes a tier-one page built as a category, not an indexed search result.
Testing before rollout
Index-control changes on a large store affect thousands of URLs at once, and a mistake — a canonical rule that catches the tier-one pages, a robots rule that blocks the category paths — removes revenue pages from the index for weeks. Before rollout:
- List the tier-one URLs and confirm, against the new rules, that each remains self-canonical, indexable and allowed in robots.
- Sample the tier-two and tier-three URLs and confirm each resolves to the intended treatment.
- Crawl the staging site with the rules applied and compare the indexable URL set to the intended one; the diff is the bug list.
- Roll out to one category first on a large site, watch the indexing report and crawl stats for that section, then extend.
- Keep the previous rules ready to restore, and annotate the change for the measurement team.
The site migrations guide treats a large index-control change as the migration it is.
Monitoring
After implementation: the Page Indexing report by reason, over time; crawl stats by URL pattern; the share of crawl going to products and tier-one pages versus parameters; the time from product publication to first crawl; and the performance of the tier-one pages for their mapped queries. Server logs give the exact split of crawler requests between products, landing pages and parameters, using the method in the log file analysis tutorial. The technical audit tutorial includes the checks; the SEO page checker reports the canonical and robots directives on any individual URL for spot checks.
Filter and parameter copies frequently show up as duplicates Google has folded into another URL; the diagnosis for duplicates without a user-selected canonical covers how to align the signals.
Why filter views that users rely on take a canonical rather than a redirect is explained in the comparison of canonical tags and 301 redirects.
Filters are only one class of parameter a store emits; the tutorial on auditing URL parameters records sort, tracking, session and search parameters in the same register and gives each class its own rule.
Common mistakes
- Indexing every facet value. Thousands of thin, duplicative pages.
- Indexing nothing. The refinements people search for never get a page.
- Inconsistent parameter order. Every selection becomes several URLs.
- Canonicals that lie. A filter page with different products canonicalised to the category; the engine ignores the hint and indexes both.
- Blocking what is linked everywhere. Indexed without content, and no signal flow.
- Sort and view parameters in the sitemap. The sitemap becomes the problem.
- Forgetting internal search.
?q=URLs are faceted navigation by another name.
The site architecture guide covers where tier-one filter pages sit in the hierarchy; the e-commerce hub places this decision among the others a store makes.
