Internal Linking for SEO: How to Build a Strategy and Silo Structure

Internal linking is the practice of linking one page on your site to another, so people and search engines can follow a clear path between related topics. Google finds most new pages through links, reads anchor text to understand them, and uses link analysis, including PageRank, to judge which pages matter. A silo structure is one way to organise those links by topic.

This guide covers link types, crawl budget facts, a seven-step strategy, anchor text rules, silos versus topic clusters, and how to audit links. Deciding which pages should exist at all is covered in our guide to planning a website structure.

An internal link points from one page on a domain to another page on the same domain. According to Google's SEO Starter Guide, Google primarily finds pages through links from pages it has already crawled. A page with no incoming internal links is therefore close to invisible, unless a sitemap or another site happens to point at it.

Internal links do three separate jobs, and mixing them up is where most link plans go wrong. They let a crawler discover a URL, which an XML sitemap helps with but does not replace. They describe the destination, because the anchor text and the sentence around it tell search engines what the page covers in words the page did not write about itself. And they signal importance: the pages you link to most often, from your strongest pages, are the pages you are nominating.

External links point at other domains. Only internal links are fully under your control, which makes them the cheapest structural fix on almost any site.

Internal links do not all behave the same way. Navigation and footer links repeat on every page and mostly help discovery. Contextual links sit inside body copy, next to the words the destination should rank for, and do most of the descriptive work. Breadcrumbs, hub pages and related-content blocks sit in between. The table shows what each type is for and the rule we apply to it.

For breadcrumbs, Google's breadcrumb structured data documentation supports BreadcrumbList markup in JSON-LD, RDFa or Microdata, needs at least two list items, and asks that the trail reflect a typical path a user would take to the page rather than simply mirror the URL.

Crawling starts from URLs a search engine already knows and spreads outward along links. As Google's overview of how Search works describes, a crawler fetches a page, extracts its links and queues the new addresses. Pages a crawler reaches cheaply get refreshed often. Pages several hops away get visited less, and sometimes not at all.

Three measurements explain most crawl problems:

• Click depth is the number of clicks from the homepage to a page. Google publishes no depth limit. Three clicks is the working ceiling we apply to commercial pages, and we flag a site when more than 20% of its indexable URLs sit deeper than that.
• Orphan pages are URLs with zero incoming internal links. Google's guidance on links says every page you care about should have a link from at least one other page on your site. For pages that carry revenue, the only acceptable share is 0%.
• Crawl budget is how much Google is able and willing to crawl on your site. Google's crawl budget guide is aimed at sites with roughly 1 million or more unique pages changing about once a week, sites with 10,000 or more pages changing daily, and sites where Search Console reports a large share of URLs as discovered but not indexed. Google calls these rough estimates, not exact thresholds.

Below those sizes, what looks like a crawl budget problem on a 400-page site is almost always a linking problem: duplicate parameter URLs, uncontrolled faceted navigation, or a section reachable only from the footer. The guide does name two wastes that apply anywhere: soft 404 pages keep getting crawled, and long redirect chains, in Google's words, have a negative effect on crawling.

A link passes three things at once: a crawl path, a description of the destination, and a share of the linking page's importance. Google's guide to its ranking systems confirms that PageRank, which looks at how pages link to each other, is still one of its core ranking systems, although Google says how it works has changed a lot since launch.

Google does not publish how much value an internal link carries, so any link equity formula you see quoted is a third-party model. What we see in practice tends to show up in a predictable order:

• New pages get indexed sooner when an already-crawled page links to them, because the link puts them in front of the crawler on its next visit
• Pages start ranking for phrases used in their anchors, since anchor text is one of the few descriptions of a page written by someone other than its author
• Deep pages stop dead-ending, because importance that used to pool on the homepage now reaches category and detail pages

Knowing which of the three effects you need decides where the link belongs. A link added for discovery works from navigation or a hub page. A link added for relevance has to sit inside a paragraph, next to the words the destination should rank for.

An internal linking strategy is a written decision about which pages should collect links, which pages should send them, and what each link should say. Without one, links pile up wherever an editor remembered a related post, and revenue pages end up with fewer links than last year's news. We work in this order.

1. List the pages that make money or bring in leads: service pages, product categories, and the SEO landing pages behind campaigns. These are your targets.
2. Group every other page under one of those targets by topic. A page that fits nowhere is a candidate for merging, not for more links.
3. Pick or build one hub page per group. On a service business that is usually the service page itself, for example a corporate website page that links to its guides, case studies and pricing.
4. Link every supporting page back to its hub with a descriptive anchor, inside the body copy rather than only in a sidebar.
5. Link the hub out to each supporting page, so the whole group can be crawled from one place.
6. Add cross-group links only where a reader needs the other topic. They are fine, as long as they do not outnumber the links inside the group.
7. Put linking on the publishing checklist. Every new page gets at least two contextual links from older pages on the day it goes live, plus one link out to its hub.

Most teams skip step 7, which is why orphan counts creep back up a few months after an audit.

Anchor text is the visible, clickable text of a link. Google's link guidelines ask for anchors that are descriptive, reasonably concise, and relevant both to the page they sit on and to the page they point to. The same document suggests a quick test: read the anchor on its own and check whether it still makes sense.

Some examples from real edits:

• Weak: "Click here to see our pricing." Better: "See what a corporate website costs."
• Weak: "Read more." Better: "Our SEO audit checklist covers the technical checks."
• Too long: an anchor that wraps a whole sentence. Better: link only the three to six words that name the destination.
• Stuffed: "cheap SEO web design agency services company". Better: "our SEO services".

Google also warns against stuffing keywords into anchors and against placing links right next to each other, where readers cannot tell them apart. For image links, Google uses the image's alt text as the anchor, so an icon link with an empty alt describes nothing.

Google has not said that repeated exact-match anchors on internal links are penalised, but twenty identical anchors pointing at one page read like automation. Vary the wording as a person would. The internal Google Content Warehouse API documentation that leaked in May 2024 listed attributes for anchor text and the text around links, which fits this advice. Google cautioned against assumptions based on out-of-context, outdated or incomplete information, so treat the leak as a hint about what Google stores, not as a list of ranking rules.

A silo structure groups every page under exactly one parent topic and keeps most contextual links inside that group. A topic cluster is the softer version: one pillar page covers a subject broadly, supporting pages cover its parts, and each supporting page links back to the pillar. Both make the boundaries of a subject easier for a search engine to read.

Silos come in two forms. A physical silo puts the grouping into the URL, so every page about one service lives under the same directory. A virtual silo leaves URLs alone and builds the grouping through links only. Virtual silos are easier to retrofit onto an existing site, because they change no addresses and need no redirects.

The difference between a silo and a cluster is how strict the walls are. A silo limits cross-links between groups, while a cluster allows them wherever a reader benefits. Google has never described silos as a ranking concept, so both models are conventions for organising links. Our working rule: a hub page should send at least 70% of its contextual links to pages inside its own group and spend the rest wherever the reader gains something.

Silos fail in one specific way. When a business sells three services and a customer question spans two of them, a strict silo forces the writer either to duplicate the answer or to leave the reader at a wall. If the choice is open, start with clusters and tighten toward silos only where topics do not overlap.

The URL structure that holds up is short, lowercase, hyphenated, and organised by topic rather than by whatever software generates the page. Google's URL structure guidance asks for readable words instead of long ID numbers, hyphens rather than the _ character, and as few parameters as possible. Its Starter Guide adds that keywords in a URL path alone have hardly any effect on ranking.

Where URLs do make a difference is in how easy a site is to manage and crawl:

• URLs are case sensitive, so pick one case and keep it; lowercase avoids accidental duplicates
• Session IDs, sort orders and referral tags multiply URLs without changing content, and Google recommends cookies over session IDs in URLs
• Filter combinations on listing pages can produce huge numbers of near-duplicate URLs, a frequent source of crawl waste on large catalogues
• Grouping similar pages in directories helps Google learn how often each section changes, which the Starter Guide says matters mostly on sites with more than a few thousand URLs
• Stable slugs save work, because every change needs a redirect and every old internal link should then be updated at the source

Deep directory paths are not a problem on their own. Click depth is a linking measure, and you fix it with links, not by shortening the path.

Flat and deep describe the same site measured from the homepage. A flat structure keeps most pages within two or three clicks. A deep one nests content across four levels or more. Neither is correct in the abstract, because a 60-page consultancy site and a 200,000-item catalogue need different answers. Choosing between types of website structure is a planning decision; the table below covers only what each shape does to internal linking.

There is no fixed limit. Google's older webmaster guidelines once suggested keeping a page under about 100 links, but its current documentation on links says there is no magical ideal number of links a page should contain, adding that if a page looks like it has too many links, it probably does. Any hard ceiling you see quoted is a convention.

What replaced the number is a judgement about attention. Every extra link divides the reader's focus and, under any PageRank-style model, the share each link passes. The conventions we apply on client sites:

• A long-form article carries five to fifteen contextual links, weighted toward the pages it supports
• A category page links to its children and its parent, and little else
• A navigation menu exposes sections, not every page inside them
• Footers carry legal and utility links, not a second sitemap

Count the links a template already emits before adding more. A header that repeats 80 links sitewide spends most of every page's links on URLs that were already easy to reach.

Most linking damage is invisible in a browser. A page looks fine, ranks for nothing, and nobody investigates, because the failure sits in how the page connects rather than in what it says. These errors turn up in almost every audit, roughly in order of how much traffic they return once fixed.

Orphan and Near-Orphan Pages

An orphan page has no incoming internal links at all. A near-orphan has one, usually from a paginated archive that pushes it off the first page within a month. Both behave the same way in search: found slowly, recrawled rarely, and ranked as though nobody on the site considers them important.

Links That Only Exist in JavaScript

Google's link best practices say it can generally crawl only links that use an <a> element with an href attribute. Links that JavaScript inserts are fine if they use that same markup. Patterns like <span href>, <a routerLink> without an href, or <a onclick> handlers are not recommended, and javascript: URLs should be avoided. A menu built that way looks like navigation to a person and like nothing to a crawler.

Nofollow on Internal Links

Google's guidance on qualifying links reserves nofollow, sponsored and ugc for links you do not vouch for. For links within your own site, it points to the robots.txt disallow rule when you want to keep crawlers out, and to noindex when you want a page out of the index. Nofollow does not redirect value either: Matt Cutts, then head of Google's webspam team, explained in 2009 that a nofollowed link's PageRank share is not passed to the other links. The one place Google's URL guidance does suggest nofollow internally is on links to endlessly generated calendar pages.

Redirect Chains Left Behind by Renames

Each rename leaves old links pointing at an address that now forwards somewhere else. One hop is tolerable. Long chains waste crawling, which Google's crawl budget guide says directly. Update the links in the source content rather than relying on the redirect map to carry them forever.

Yes, in the same way it matters for regular search. Google's page on AI features and your website says there are no additional requirements or special optimizations for AI Overviews or AI Mode, and it lists making content easy to find through internal links among the standard practices that apply. A page also has to be indexed and eligible for a snippet before it can appear as a supporting link.

The same page describes query fan-out, where Google issues several related searches across subtopics to build one answer. Subtopic pages that are crawlable, indexed and linked to their hub give that process more pages to find. Google has confirmed nothing beyond that, so treat claims that a linking pattern earns AI citations with caution.

We still write each section to answer one question on its own, because it helps readers who land halfway down a page.

An internal link audit answers four questions: which pages have no incoming links, which important pages have too few, which links are broken or redirected, and which anchors describe the wrong thing. You need a full crawl of the site plus a list of the pages that matter commercially. Everything after that is comparison work, and it fits inside a wider SEO audit if you are running one.

The sequence that produces usable output:

1. Crawl the site and export every URL with its inlink count, click depth and status code
2. Export the same URL list from the XML sitemap and from analytics, then compare the three to expose orphans
3. Rank the commercial pages by inlink count, and mark anything receiving under 1% of the site's internal links as under-linked
4. Pull anchor text per destination and look for pages where every anchor is identical or generic
5. Filter for links returning 3xx or 4xx codes, and fix them at the source rather than in the redirect map
6. Write the fix list as page-level tasks, because a linking task nobody owns never ships

Repeat the crawl a month after the fixes land. Inlink counts move immediately, but ranking movement on the target pages usually needs a recrawl cycle to show up in post-launch performance data.

Any crawler that reports inlinks per URL handles the structural half of the job. The other half needs search performance data, because a page with plenty of internal links and no impressions has a different problem from a page with impressions and almost no links. Most teams settle on one crawler plus one performance source.

• Desktop crawlers such as Screaming Frog and Sitebulb map the whole link graph, report click depth per URL, and export inlink lists that a spreadsheet can compare.
• Cloud site audits in Ahrefs and Semrush run the crawl on a schedule and flag orphan pages against their own index, which catches URLs a fresh crawl would miss.
• The Links report in Google Search Console shows which internal links Google recorded, the only view that reflects what Google saw rather than what your crawler found.
• Server log analysis shows which URLs bots requested and how often, and settles crawl budget arguments that no crawler can.

Automation needs a lighter touch. Plugins that turn every mention of a phrase into a link produce the pattern Google's guidance warns about: repeated keyword anchors placed where no reader needs them. AI link suggestions are useful as a shortlist, as long as someone accepts them one at a time. The tools surface gaps; deciding what should link to what stays with whoever knows the business.

A redesign is where internal linking most often goes backwards. Templates change, sections merge, and contextual links that lived inside old body copy disappear. The pages still exist and still rank, right up until crawlers stop finding paths to them. Capturing the old link graph before launch is the only defence that reliably works.

The checklist we run on every corporate website project that replaces an existing site:

• Crawl the live site and archive the full inlink and outlink export before anything changes
• Map every old URL to a new one, including URLs that will be merged rather than moved
• Rebuild contextual links inside the migrated body copy, not just the navigation
• Verify that the new templates emit real anchor tags, not scripted click handlers
• Recrawl the staging environment and compare inlink counts page by page against the archive
• Recrawl again 48 hours after launch, because template edits made during launch week are where links usually vanish

Redirect mapping and tracking are covered in our website redesign SEO checklist and the website migration guide. Our pricing starts from €1,500 for a landing page, €4,500 for a corporate site and €9,000 for a web portal, and our development process and portfolio show how the work runs.

Internal linking is the part of technical SEO a content team can own without touching the codebase, and a paragraph link costs nothing to add.

A silo structure is worth the extra discipline when a business covers several distinct subjects and needs each one to read as deep rather than incidental. When the subjects overlap, looser topic clusters do the same work with fewer arguments about where a page belongs.

Whatever the model, the maintenance habit matters more than the diagram. Crawl the site quarterly, watch the orphan count, keep anchors descriptive, and repair links at the source instead of stacking redirects.