SEO for ecommerce: the URLs nobody decided to publish

Google's documentation on faceted navigation, updated on 18 December 2025, says the common implementation of filters "can generate infinite URL spaces which harms the website". Nobody writes those URLs. The filter software produces them from the catalog, and it keeps producing them while the SEO work happens somewhere else. In Ahrefs' US data, pulled for this blog on 5 September 2026, seo for ecommerce carries 2,800 searches a month against a keyword difficulty of 57, a 4.00 USD click and an AI Overview on the results. The four guides I opened that day teach roughly the same sequence, and three of them never name the free report that says whether a store has this problem at all.

How much of an ecommerce site should Google index?

Ecommerce SEO succeeds when Google indexes the pages a store meant to publish and skips the ones its filters generated. Below roughly 10,000 URLs the crawl side is rarely the constraint; a mid-size catalog with filters open to crawling usually is; past a million URLs, deciding which pages exist becomes the work itself.

Quick answer

• The index count is the first measurement, before the keyword list. Search Console's page indexing report says how many of a store's URLs Google kept, and the distance between that number and the catalog is the brief.
• Filters publish URLs that nobody authored. Google's own wording for the common implementation is that it can generate infinite URL spaces, and those pages compete for the same crawling as the categories that sell.
• Crawl budget is not everybody's problem, and Google names the sites it is for. Two of its three conditions are about size. The third is not, and it is the one that catches stores.
• Part of an ecommerce presence is not reachable by SEO at all. Four of the six Google surfaces documented for product data route through a Merchant Center feed; Search is not one of them.

The belief that the giants took the rankings

Every store owner arrives with the same diagnosis, and in the United States it has a name. The marketplaces and the national retailers own the shelf, the reasoning goes, so a catalog of nine hundred products is fighting for scraps.

Sometimes that is exactly what is happening. It is also the most expensive first hypothesis available, because testing it means competing, and a cheaper test comes before it: whether the store's own pages are in the index at all. A page Google has not indexed is not losing to Amazon. It is not in the contest.

That distinction decides what the budget buys. If the pages are indexed and outranked, the work is authority, offer and page quality, all slow and all expensive. If a large share of the URLs were never crawled, the work is deciding which URLs deserve to exist, which is unglamorous, fast and mostly free. The two briefs share almost nothing, and one free report tells them apart.

What Google's crawl budget guide actually says

The mechanics of ranking do not change for a store. Crawling, indexing and ranking treat a catalog like any other site, and the foundations in what SEO is and what it is for transfer without edits. What changes is the arithmetic of how many pages the site asks Google to consider in the first place.

Google publishes a guide for exactly that, and it opens by telling most readers to leave. Updated on 22 July 2026, it states: "If your site doesn't have a large number of pages that change rapidly, or if your pages seem to be crawled the same day that they are published, you don't need to read this guide." Then it names who it is for: large sites of a million or more unique pages whose content changes about once a week, medium or larger sites of 10,000 or more with content changing daily, and, third, "Sites with a large portion of their total URLs classified by Search Console as Discovered - currently not indexed".

The first two conditions are about size and most stores fail them. The third is not about size at all: it is a status in a free report, and a catalog of four hundred products can meet it if its filters are open to crawling. Search Console's help defines that status as follows: "The page was found by Google, but not crawled yet. Typically, Google wanted to crawl the URL but this was expected to overload the site; therefore Google rescheduled the crawl." Google found the URL, judged that crawling it would cost the site too much, and postponed.

The report lives inside the property, so the first hour of an ecommerce engagement is an export rather than an audit deck. Setting up Search Console is the prerequisite, and the number worth writing down is not sessions. It is indexed pages against the count of products and categories the store believes it publishes.

Where the extra URLs come from

Filters are the loud answer and pagination is the quiet one.

Google's faceted navigation page describes two distinct harms. One is overcrawling: crawlers spend requests on faceted URLs before working out that they are useless. The other is slower discovery, because the requests spent there are not spent on new pages. A category published on Tuesday waits behind a queue of color and price combinations nobody asked for.

Its two prevention methods are short. One is disallowing the faceted navigation URLs in robots.txt. The other is expressing filters as URL fragments, the #products=fish form, so a filter never becomes a separate URL. For the faceted URLs that do need to be crawled and indexed, the same page asks for a standard & separator, a consistent order of filters in the path, and an HTTP 404 for combinations that return nothing.

Which attributes earn that treatment is a demand question, answered the way any keyword decision is, with keyword research done on free tools applied attribute by attribute rather than to the catalog as a whole. A color has demand of its own. A price band crossed with a sort order has none, and nobody will ever type it.

Two pieces of advice that circulate for this exact problem stopped being true and are still repeated. The Search Console URL Parameters tool, which a generation of guides names as the place to fix parameter crawling, was deprecated in March 2022; Google's announcement gave as its reason that only a small share of the configurations were useful and that its crawlers work parameters out on their own. And rel=next and rel=prev, still recommended for paginated category listings, are settled in one sentence on Google's pagination page: "Google no longer uses these tags." The same page rules out the other habit it travels with: "Don't use the first page of a paginated sequence as the canonical page. Instead, give each page its own canonical URL."

A store following both retired recommendations ends up with a paginated catalog canonicalized into its own first page, which removes everything past page one from consideration and is one of the reasons a page can stay out of the index while every visible thing about it looks correct.

When more pages are not the problem

When the indexed count already matches the catalog. If Search Console reports every category and product as indexed and the store still has no traffic, the constraint is demand or the offer, and pruning URLs moves neither.

When the catalog is a few dozen products. Google's guide excludes that store by its own thresholds, and the crawl story becomes a way of billing hours against a target nobody measures.

When the navigation is not crawlable at all. Google's ecommerce structure page asks for <a href> tags and says not to use JavaScript events on other HTML elements for navigation. A store whose menu is a script has an indexing defect rather than a budget one.

When most of the revenue comes from a marketplace. The site is a showroom while that holds, and the return on a retainer arrives later than the invoices for it. What those bands buy is in what SEO costs a small business, where SE Ranking's survey of 260 agency responses, published on 13 December 2024, puts 64 % of them below 1,000 USD a month, with its own authors warning that the sample may not reach statistical significance.

The audit that found the mirror image

The opposite defect produces the same symptom and needs the opposite fix. In an audit of a 312-page client site, a recipe publisher rather than a store, 35 pages had no internal link pointing at them at all. Between them they carried around 180,000 impressions a quarter that the site's own structure never reached. Of the links that did exist, 79 % were piled into the footer list, which is a template linking rather than an editor.

That is the same disease as an open filter set, read from the other end. In both cases a template decides which URLs exist and which pages get reached, and nobody at the company chose either outcome. Google states the consequence plainly on its ecommerce structure page: "If category pages don't include direct links to all products in a category, Googlebot might not find all of your products by crawling alone", and, on internal weight, "The more links a page has to it within a site, the higher the relative importance of the page to other pages on your site."

The repair followed a rule worth carrying into a catalog: no sentence gets written in order to place a link. Mentions that already existed were wrapped instead. Links in prose went from 251 to 367, orphan pages from 35 to zero, and broken internal links from 9 to zero. It is not an ecommerce case, and what transfers is the mechanism rather than the figures: a catalog that links its categories from the menu and its products only from a paginated listing has orphans it has never counted, and both halves are template decisions with a long shelf life, closer to the checks a site passes before it launches than to anything in a monthly report.

The half of the shelf a feed reaches and SEO does not

Google documents where ecommerce content can appear, and the list runs longer than the blue links: Search, Images, Lens, the Shopping tab, the Business Profile and Maps. Search asks for structured data on the product pages and Images for image best practices. The other four route through Google Merchant Center, with Maps additionally requiring inventory location data inside the feed.

The sentence that settles the confusion in both directions sits on Google's page about sharing product data: "While uploading product data to Google Merchant Center isn't mandatory to appear in Google search results, it can enhance Google's understanding of your products." And, in the same document: "Participation in Google Merchant Center is mandatory for some Google surfaces, such as listings in the Google Shopping tab."

So a store does not need a feed to rank, and a store without one has conceded four documented surfaces. In most companies the feed belongs to whoever runs the shopping ads and the SEO work never opens it, so product titles and attributes are maintained twice, drift apart, and get audited once. Whether a query is better bought or better ranked is the trade examined in SEO against Google Ads; what this list adds is that for a catalog the two channels are fed from the same product data, so an error in it is inherited by both.

The markup itself deserves the modest framing Google gives it. Its ecommerce page recommends Product and ProductGroup, BreadcrumbList, Organization, Review, and LocalBusiness for stores with a physical location, and presents them as a way to improve the accuracy of Google's understanding of a page rather than as a ranking factor. A store that adds markup and waits for positions has bought the wrong expectation with the right implementation.

What the summaries changed for a store

The results page a shopper lands on is not the one these guides were written against. Ahrefs, in an update published on 4 February 2026 over 300,000 keywords, half of them with an AI Overview and half without, comparing December 2023 with December 2025, reports that the presence of an AI Overview now correlates with a 58 % lower average clickthrough rate for the top-ranking page. That is a correlation between summary presence and average click-through, not a fixed set of pages followed through time, and the difference matters when somebody quotes the number at you as a measured loss.

The part of a catalog carrying that exposure is the informational part: the buying guides, the how-to-choose articles, the material a store publishes to attract readers earlier. Commercial queries have held up better, which turns "categories before blog" from a preference into a measured argument. A summary is also a poor answer to a request to compare and buy, so the buying query still returns a shelf.

What moves is who gets named inside the summary, and that runs on the raw material extraction has always run on: one narrow question answered in one place, with the entity written out instead of implied. The mechanics are in what it takes to get cited by ChatGPT, and the product page that survives the treatment carries a specification somebody verified, not the manufacturer's paragraph that forty other stores also published.

Mistakes that repeat

• Reading the performance report and never the page indexing report. The first says what the store earned. The second says how much of the store is eligible to earn anything.
• Opening every filter combination to crawling. The combinations grow multiplicatively, and Google describes the result as an infinite URL space competing for the same requests as the categories.
• Sending someone to the URL Parameters tool. It was deprecated in March 2022, so the most repeated fix for this problem is a control that no longer exists.
• Canonicalizing a paginated sequence into its first page. Google's pagination guidance says the opposite, and the products on page four stop being candidates for anything.
• Treating the product feed as the advertising team's file. Four of the six documented surfaces run through it, and the titles it carries are the titles the site publishes.
• Judging the engagement by pages produced. Descriptions rewritten is a countable deliverable on a site whose problem is that it already has too many URLs, and it is a common reason a flat curve outlives an invoice; the order for reading a flat curve is in diagnosing an organic traffic drop.

Data and transparency

The 2,800 monthly US searches, the keyword difficulty of 57, the 4.00 USD cost per click and the AI Overview present on the results come from an Ahrefs pull for this blog's keyword research on 5 September 2026. The reading of the first page of results is my own, done that day over four guides: one published 22 May 2025, one updated 28 July 2025, one undated, and one published 4 February 2026 and updated 24 April 2026, the only one of the four that covers facets, pagination and out-of-stock URLs. The first of them attributes two click-through figures to Backlinko, 0.63 % of searchers clicking page two and 27.6 % of clicks going to the first result, and does not name the window either was measured over, so they are examined here and not used. Every quotation of Google's documentation was taken from the pages themselves on 5 September 2026: the faceted navigation guidance last updated 18 December 2025, the crawl budget guide last updated 22 July 2026, and the ecommerce structure, pagination and product data pages last updated 10 December 2025. The first two have moved to a new crawling section of the developer site; the old Search Central addresses still redirect, and the links below point at the current path. The click-through figure comes from Ahrefs, published 4 February 2026 over 300,000 keywords comparing December 2023 with December 2025; it is a correlation between AI Overview presence and average click-through, not a loss measured on a fixed cohort. Retainer context comes from SE Ranking's pricing survey published 13 December 2024 over 260 agency responses, whose authors warn the sample may not reach statistical significance. No site in the portfolio I operate and audit is an ecommerce, so no store case appears here; the 312-page audit is a client content site, cited for the linking mechanism and with rounded figures. The operating judgment comes from auditing sites that record more than 300 million impressions a year in Search Console. Verified as of September 2026.

Primary sources, opened on 5 September 2026: Google on managing faceted navigation; the crawl budget guide; the Search Console page indexing report; sharing product data with Google; ecommerce pagination and incremental loading; the URL Parameters tool deprecation; the Ahrefs update; the SE Ranking survey.

What this changes

The work that decides an ecommerce site's search results was done by whoever configured the filters, and it was never framed as SEO work. It shipped as a merchandising feature in a sprint, and it has been publishing URLs ever since without a single person reviewing one of them.

Which is why the first deliverable of this channel is subtraction, and why subtraction is close to unsellable. A proposal to write two hundred product descriptions describes a month of visible activity. A proposal to remove eighty thousand URLs from the index describes a week and looks like less work, because it is. The store that does the second one first spends the rest of the year competing with the pages it meant to publish. The one that does the first one first spends it competing with itself.

Frequently asked questions