🧭 Website architecture for SEO: a structure that crawlers and people understand

Most sites lose in search not because of weak copy, but because the crawler simply never reaches half of the important pages. In this piece, we will look at site architecture as an engineering problem: where crawler attention goes, why the third level of depth often becomes the last, and how links between documents work the same way as links between sites, if you have already tried guest posting to grow visibility.

How to design site structure in six steps

Each step is covered in detail below, but the overall route is compact and convenient to keep in front of you during an audit.

💡 Quick overview:

• Step 1: Pull a full list of URLs with a crawler and compare it with what the sitemap actually returns.
• Step 2: Calculate click depth for every page from the homepage and find everything that sits deeper than the third level.
• Step 3: Group pages into topic clusters by search intent, not by internal company departments.
• Step 4: Bring URLs into predictable directories, where the path reads like a section address.
• Step 5: Set up internal links from hub pages to end pages and back, removing documents with no inbound links.
• Step 6: Lock in the result with navigation, breadcrumbs, and an up-to-date sitemap, then repeat the measurement in a month.

What the crawler sees instead of your menu

The bot does not perceive a site as a picture with a menu and a footer. It receives a set of addresses, follows the links it can recognize, and spends a limited resource on that. In Google's guide to crawling large sites, updated in July 2026, the threshold for concern is stated directly: owners of projects with 1,000,000 or more unique pages with weekly changing content should think about crawl budget. The same guide names a second threshold, from 10,000 unique pages, if content is updated daily.

Below these values, crawl budget is almost never the real cause of problems. Far more often the structure is to blame: the bot honestly spends visits on service addresses, filter duplicates, and endless pagination pages, while product cards and articles get whatever is left over.

A separate trap concerns the links themselves. Google, in its documentation on link recognition, describes a strict condition: a transition is counted when it is a regular element with an address attribute. A menu built on click handlers, buttons, or infinite loading looks like a dead end to the bot, even if users navigate it without any trouble.

There is a simple way to tell that the structure is working against you. Open the indexing report and look at the "Discovered, not indexed" status. When hundreds of addresses pile up there, it is almost never a text problem. It is a signal that pages sit too far from entry points and look optional to search.

Click depth: why the third level becomes the last

Click depth is measured not by the number of slashes in the address, but by the number of transitions from the homepage to a document along the shortest path. These are different things, and confusing them is costly: a page can have a short URL but no links at all except from the sitemap.

The industry benchmark has held for years: in Ahrefs' guide to internal links, the architecture is proposed as a pyramid, so that any important page is no more than three transitions away from the homepage. The point is not the magic of the number, but the distribution of weight: every extra level noticeably dilutes the signal that reaches from the most authoritative pages.

In practice, depth grows unnoticed. A section expands, pagination appears, then a subcategory inside a subcategory, then an archive by year. A year and a half later, half the catalog lives at the fifth level, and nobody planned it. That is why measuring depth should be a regular procedure, not a one-time cleanup.

The fix is not moving files, but adding short paths. Collections, "important in this section" blocks, links from popular content, and thoughtful navigation shorten the route without breaking existing addresses. A URL migration is worth starting only when the very logic of the sections has already been recognized as wrong.

URL as an address, not as a random identifier

A page address does double work. For a person, it is a hint about where they are; for a bot, it is a sign of a group. In Google's recommendations on URL structure, it is stated directly that grouping similar topics into directories helps search understand how often each section is updated and crawl different directories at different frequencies.

The flip side of the same logic: redundant parameters spawn nearly identical addresses, and crawling goes nowhere. Typical cases and clear replacements are summarized below.

A separate conversation about stores. Google, in its material on address structure for ecommerce, advises keeping faceted navigation under control and not allowing filter combinations to turn into independent indexable addresses. This is the case where one catalog setting affects visibility more than a dozen rewritten descriptions.

Clusters instead of rigid silos

The classic silo assumed that sections barely linked to each other so that weight would not spread out. Today, editorial teams work differently. A topic cluster is built around a pillar page that answers a broad query, and around it live narrow articles covering specific cases. The pillar page links to the satellites, and the satellites link to the pillar and to each other wherever that helps the reader.

The difference is fundamental. A silo forbids connections, a cluster encourages them, but requires meaning: a link is placed where the reader genuinely needs the next step. This network gives search engines a clear set of signals about which page in the group is the main one, while also reducing the risk that two of your articles compete for the same query.

A practical rule when designing: first map out the clusters on paper, then decide how they will fit into directories. If you do it the other way around, the structure will mirror the company's organizational chart rather than the logic of demand. The user is looking for a solution to a problem and does not care which department is responsible for it.

Internal links distribute authority, they don't just connect pages

Internal linking is often treated as a decorative touch for content. In reality, it is the only tool you have to control how authority is distributed within a site, and the only way to tell search engines what matters more.

A density guideline appears in the 2025 internal linking guide: 2 to 5 meaningful links per 1,000 words of text. This is not a hard rule, but a reminder that both an empty article with no links and a wall of thirty links in a row are equally useless. The same source emphasizes the value of anchor diversity: the same page receiving links with different anchor texts gets a more complete description of its topic.

The second half of the job is finding pages that have no internal links pointing to them. In Semrush's breakdown of orphan pages, such documents are described as a typical byproduct of redesigns and migrations: the content exists, sometimes even collects external links, but within the site there is no click path to reach it. You can find them by comparing the crawler's list with the sitemap list and analytics data.

Another layer that is usually missed is link direction. In Sitebulb's practical guide to internal link optimization, it is shown that one-way top-down internal linking leaves bottom-level pages with no way to pass signal back up. Reciprocal links between related content strengthen the entire cluster, not just its top.

Sitemap, navigation, and breadcrumbs solve different problems

These three elements are often lumped together, even though they work differently. Navigation defines the main entry points and holds the structure together for humans. Breadcrumbs show a document's position in the hierarchy and provide an additional path upward. The sitemap does not affect hierarchy at all; it simply lists the URLs you consider worth crawling.

The technical limits of a sitemap are strict and have not changed for years: according to Google's sitemap file specification, a single file holds up to 50,000 URLs and weighs no more than 50 MB uncompressed, and when limits are exceeded, an index file is created with links to multiple sitemaps. This is a convenient control point: if the number of rows in the sitemap diverges sharply from the number of actual pages, you either have crawl junk or missing sections.

An important detail about menus: cramming everything into them is harmful. When top navigation contains several dozen links, authority gets spread in an even layer, and search engines stop understanding priorities. The menu should reflect the main sections, with depth revealed inside them.

Case study: a store that drowned in its own filters

A typical story familiar to anyone who has audited a catalog. A mid-sized store adds faceted navigation: color, size, material, brand, sorting. Every combination returns a separate URL accessible via a link. A few months later, search reports show an order of magnitude more URLs than products, a significant portion sits in "Discovered, not indexed" status, and new product pages take weeks to get indexed.

The analysis always starts with one question: which of these combinations does anyone actually search for. It usually turns out that demand exists for one, at most two facets, such as brand plus category. Those are kept accessible and turned into full landing pages with descriptions. Everything else is blocked from crawling via parameters, and the canonical URL of the category gets an unambiguous directive, as recommended in the Google merchant guide mentioned above.

Then the second part kicks in: product pages stop being orphans. Links to them appear from curated collections, from related product blocks, from blog reviews. Crawling returns to pages that generate revenue, and indexing speed for new items drops from weeks to days. There is nothing exotic about this procedure, but it delivers the effect that rewriting descriptions never does.

Structure audit checklist

It is convenient to run the check against a fixed list so you don't forget the boring items. Collect data with a crawler and go through the table top to bottom.

⁉️🤔 Frequently asked questions

Should I move pages if they sit too deep?

Usually not. Moving URLs means a chain of redirects and a temporary dip. First shorten the path with links: collections, blocks within the section, mentions from popular content. Change URLs only if the section logic itself is recognized as flawed and would have to be redone anyway.

Does the number of slashes in a URL directly affect ranking?

There is no direct penalty for nesting in the URL. What matters is click depth, meaning the number of clicks from the homepage. The URL matters as a group signal: directories help search engines determine which pages belong to the same section and how often it is updated.

How many internal links should I put in an article?

Focus on meaning, not a counter. A reasonable range, two to five meaningful links per thousand words, works as a reminder of balance. Each link should lead where the reader would logically go next, and the anchor should describe the target page's content.

Does a sitemap replace good internal linking?

No. A sitemap reports URLs, but it does not pass authority or define hierarchy. A page that appears only in the sitemap and has no inbound links is treated as secondary. The sitemap complements the structure, but does not substitute for it.

What should I do with filters in an online store?

Keep open only the facets with confirmed demand and turn them into full landing pages. Block the remaining combinations from crawling and set the canonical URL to the category, otherwise filters will consume crawl budget and create a mass of nearly identical documents.

Summary: structure as an agreement with search engines

Site architecture rarely looks like an urgent task. It has no visible deadline, it does not show up in a rankings report, and restructuring almost never looks impressive. But it determines which pages get a chance to rank at all and which remain in discovered-and-forgotten status.

Let's boil it down to three pillars. First: the path to any important page should be short and follow regular links that the crawler can recognize. Second: URLs should read as clear directories, not as a warehouse of parameters. Third: links within the site should reflect semantic clusters, not the company's organizational chart.

Start with the cheapest step. Export the URL list with a crawler, calculate click depth, and find pages with no inbound links. This measurement takes an evening and almost always reveals a couple of sections you forgot about. Then work through the checklist above, and a month later repeat the measurement and compare: the trends will show whether the new scheme is working. If the topic of links resonates with you, look at the external landscape too, where the same logic of useful transitions works between sites.