I Audited Internal Linking on 300 Sites. Three Patterns Move Rankings. The Rest Waste Crawl Budget.

Most internal linking looks tidy in a crawler and does almost nothing in search. We've seen it hundreds of times now. A site looks beautifully organised in Screaming Frog, the boxes all connect, and yet the pages that pay the bills barely rank.

So our team did the unglamorous thing. Between mid 2024 and early 2026 we pulled the internal link graph on 300 sites and matched it against what those sites actually achieved. Rankings, clicks, conversions, and for the most recent batch, citations inside AI answers. This is what we found, and what we'd do with your site this week if you handed it to us.

We're going to skip the dictionary definition of an internal linking. If you need that, the current page-one results have it covered. This is for the SEO, content lead, or founder who already knows what an internal link is and has been told to "fix internal linking" with no further instruction. Welcome. It's less mysterious than people make it.

The TL;DR, because you're busy

• Four shapes show up again and again: hub-and-spoke, hierarchical, mesh, and contextual. Only three of them reliably move outcomes.
• Hub-and-spoke wins for ranking a specific commercial page. Contextual wins for AI citations. Hierarchical wins for sites past a few thousand URLs. Mesh wins almost nothing, despite looking the most impressive on a graph.
• The biggest lever nobody audits is internal anchor diversity. Money pages with one repeated exact-match anchor underperform pages with four to eight natural variations.
• Google has said the location of an internal link (footer versus body) barely matters for understanding a page. So stop arguing about placement and fix what the link actually says.
• If you do one thing this week, count how many unique anchor variations point at your top five revenue pages. Most sites have one. They should have several.

How we built the sample

A quick word on method, so you can decide how much to trust us.

Across roughly eighteen months we ran technical audits on 300 sites. The split was 180 commercial sites (services, ecommerce, SaaS), 70 publisher and content sites, and 50 local service businesses. For each one we pulled a fresh crawl of the internal link graph, layered Google Search Console clicks and impressions by URL on top, and where the client gave us GA4 access, conversions per URL.

For the last 90 audits we also tracked AI citations. We ran brand-name and category-level queries through ChatGPT, Perplexity and Google's AI Overviews and recorded whether the site got pulled in. That AI layer changed a few of our conclusions, which we'll get to.

None of this is a peer-reviewed study. It's practitioner data from real campaigns. We're telling you that up front because we'd want someone to tell us.

The four patterns we keep seeing

Almost every site falls into one of four internal linking shapes. People rarely build them on purpose. They emerge from how the site grew, which is usually "we added pages and hoped."

1. Hub-and-spoke

One central page (the hub) links out to a cluster of supporting pages (the spokes). Each spoke links back to the hub. Spokes link to each other sparingly. This is what most people mean by "topic cluster," and it's the same logic behind the knowledge graph and entity model we use for AI search.

Where we see it win: ranking a specific commercial or money page. The hub soaks up internal equity from many spokes and gets visible lift on its target term.

Where it fails: when the spokes have nothing to say. A cluster of thin pages pointing at one money page does not rescue the money page. Google notices, and frankly so do readers.

2. Hierarchical

The classic tree. Homepage to category pages, categories to subcategories, subcategories to products or articles. Each level mostly links downward, with breadcrumb links back up. This is the default for ecommerce and big publishers, and Schema.org formalises it with the breadcrumb property on WebPage, defined as "a set of links that can help a user understand and navigate a website hierarchy." Google uses it to render breadcrumb trails in results, so it's not just decoration.

Where we see it win: large sites. Google's own crawl budget guidance says you only really need to worry about crawl budget on sites with a million or more pages, or 10,000 or more pages with rapidly changing content. At that scale a clear hierarchy is the spine that keeps crawling predictable. Googlebot needs the map.

Where it fails: small sites that copy the pattern. If you have 80 pages and a six-level taxonomy, you've made things worse. Click depth balloons and pages get orphaned. We once found a 60-page accounting firm with its main service buried three clicks deep behind a category nobody clicked. Their best money page might as well have been invisible.

3. Mesh

Every page links to loads of other pages. The graph looks like a ball of yarn a cat got to. "Maximum PageRank flow," the theory goes. In practice it's what you get when an auto-linking plugin runs wild, or when "related posts" sit on top of "you may also like" sit on top of a tag cloud.

Where we see it win: almost nowhere. Across all 300 audits, mesh structures consistently underperformed hub-and-spoke for ranking and contextual for AI citations. Every time.

Why it fails comes down to maths. The PageRank model, the one Larry Page and Sergey Brin published back in 1998, is blunt about this: "the PageRank transferred from a given page to the targets of its outbound links upon the next iteration is divided equally among all outbound links." Put 80 links on a page and each one passes roughly an eightieth of the equity a single focused link would. Spread thin enough, every link is worth almost nothing. Mesh is dilution dressed up as generosity.

4. Contextual

Links sit inside body content based on what's actually being discussed, not where the page lives in a taxonomy. The anchor text describes the destination. Links are scattered across paragraphs, not bunched in a sidebar. This is the pattern Google's SEO link best practices doc keeps nudging you toward when it says good anchor text is "descriptive, reasonably concise, and relevant to the page that it's on and to the page it links to."

Where we see it win: AI citations. This is the pattern that shows up most when a site gets pulled into ChatGPT and Perplexity answers. Models read sentences and grab links the way a person would, so a link inside a paragraph is far more visible to them than a navigation or footer link. We dig into the why in our piece on getting your brand into AI answers.

Where it fails: when there's no editorial discipline. Contextual works because a human, or a careful automation, decided this link belongs in this sentence. Spray-and-pray "contextual" is just mesh wearing a nicer jumper.

Which pattern wins what

Here's the bit nobody prints clearly, so we will.

• Ranking a specific commercial page: hub-and-spoke wins most often.
• Getting cited in AI Overviews, ChatGPT or Perplexity: contextual wins most often.
• Managing a site with thousands of URLs: hierarchical wins.
• Looking impressive in a slide deck: mesh, and only there.

The honest answer is that a real site layers two or three of these. Hierarchical for the structural spine, hub-and-spoke for each service or topic cluster, contextual links woven through the body of every page. Mesh is the one we keep telling clients to stop doing, usually right after we find the plugin that caused it.

The lever nobody audits: internal anchor diversity

This is the part we get most animated about, because it's both the most ignored and the easiest to fix.

When we pulled the anchor text distribution for every internal link pointing at the top five revenue pages on those 300 sites, one pattern was loud. Sites where 70% or more of internal links to a money page used the same exact-match anchor tended to rank worse than sites where that distribution was spread across descriptive variants, partial matches and branded anchors. The top performers usually had between four and eight distinct anchor variations pointing at each important page. The laggards had one or two, copied everywhere.

Google's guidance backs this up without spelling it out. The link best practices doc says anchor text should be "descriptive, reasonably concise, and relevant." It does not say "identical every time." Repeat one anchor 80 times and you've handed Google a single signal shouted 80 times. Vary it and you've handed over 80 facets of context about the same page. We see this play out as the single most common own-goal in our work on anchor text for Google versus AI engines: people optimise external backlink anchors carefully, then let their CMS stamp the same internal anchor on every page.

One more myth to bury. Google has been clear that the location of an internal link doesn't change much. As John Mueller explained on the question of header and footer links, Google doesn't really differentiate, and won't treat a footer link as having less weight or quietly ignore it. So please, stop debating body versus sidebar. Audit your anchor variety instead. That's where the wins are.

What anchor diversity looks like in practice

Say you have a service page for SaaS link building. Internal links pointing at it might use:

• "SaaS link building" (exact match)
• "link building for SaaS companies" (partial)
• "how we build links for B2B SaaS" (descriptive)
• "our approach to SaaS link campaigns" (branded descriptive)
• the brand name itself (branded)

The goal isn't a perfect ratio. The goal is variety that reads naturally to a human. If your plugin keeps forcing the same exact-match anchor everywhere, that's the thing to fix.

What 300 audits taught us about crawl budget

Most small sites do not have a crawl budget problem. If you've got 200 pages and Googlebot is crawling hundreds of URLs a day, you're fine, and Google's own docs agree the threshold to even care sits well into the tens of thousands of pages. At that size your real issue is usually the opposite: your important pages don't get enough internal links, not too many.

The interesting cases were the mid-to-large sites. Two crawl killers showed up again and again.

1. Faceted navigation linked to every filter combination. Colour plus size plus price plus sort end up as crawlable URLs linked from listings, and Googlebot dutifully follows them into the void. We've seen 10,000 real pages turn into hundreds of thousands of crawlable URLs this way. Google's faceted navigation guidance is blunt: "Oftentimes there's no good reason to allow crawling of filtered items, as it consumes server resources for no or negligible benefit." We walk through the fixes in our guide to faceted navigation, pagination and crawl budget.
2. "Related posts" widgets with no editorial logic. A post on "best CRMs" pointing to 24 random articles in a sidebar passes meaningful context to none of them. It just adds noise.

The single biggest crawl efficiency improvement we made across the audits was the same move every time: shift from mesh toward hierarchical-plus-contextual. Less yarn, more spine.

How AI search changes the calculus

AI search reads internal links differently from Googlebot, and once we started tracking citations across the last 90 audits, three things shifted in how we think about this.

1. Link position matters for AI, even though it doesn't for Google

Google treats a footer link and a body link much the same. AI doesn't. Models consume content in chunks and tend to pull from earlier, denser content, so a link inside the first stretch of body content gets into the model's working context far more often than a footer or sidebar link. Same link, very different odds depending on where it sits.

2. Anchor text and the sentence around it become one unit

An LLM doesn't read your anchor in isolation. It reads "as we explained in our piece on X," sentence and all. So a link's value to an AI is the anchor plus the sentence or two wrapped around it. That makes lazy "click here" links practically invisible to models, and makes well-written contextual links more powerful than they were eighteen months ago.

3. Mesh confuses models more than it confuses Google

Google has had 25 years to cope with messy link graphs. LLMs haven't. When a model is summarising a topic, a clean hub-and-spoke or contextual structure gives it strong signals to pull from. Mesh gives it 30 weak ones. That's the core reason contextual wins for AI citations while hub-and-spoke wins for traditional rankings. You're optimising for two different readers with different patience levels.

Three things we never do, because we've watched them backfire

• We don't install "AI internal linking" plugins on small sites. On a 50-page site you don't need a plugin, you need an hour and a spreadsheet. Most of these tools build mesh by default, and you'll spend longer pruning than you'd have spent linking by hand.
• We don't force every page into a cluster. Some pages are genuinely useful standalone resources. Bending them into a hub with awkward links makes the cluster weaker, not stronger.
• We don't do programmatic internal linking when the surrounding content is thin. If you can't write a sensible anchor and a sentence around it, the link probably shouldn't exist. We've written about why programmatic SEO backfires when the content underneath is thin, and programmatic internal links sit in exactly the same bucket.

What Google has actually said, and what people misquote

The internal linking conversation is full of confident claims that turn out to be loose paraphrases. So, straight from Google's published guidance:

• The link best practices doc says "every page you care about should have a link from at least one other page on your site." A minimum bar, and plenty of sites fail it.
• The same doc warns against generic anchors like "click here" and "read more," because they tell neither the reader nor the crawler anything about the destination.
• Pages with no internal links pointing at them are orphans, and as Ahrefs describes them, they're "pages that search engines may have difficulty discovering because they have no internal links from elsewhere on your website." If a page can't earn a link from somewhere on your own site, that's a flag.
• The breadcrumb property in Schema.org is recognised by Google for site hierarchy. Use breadcrumbs. They reinforce the hierarchical spine for free.

A practical audit you can run in an afternoon

If you want to do something with all of this, here's the order we run it in. On a small site this is a single afternoon with a crawler and a spreadsheet. On a big site we lean on a custom crawler plus the open Common Crawl dataset for external context, but the logic is identical.

1. Pull a fresh crawl. Screaming Frog, Sitebulb, whatever you've got. Export internal links by source and destination URL, with anchor text.
2. Pick your top five revenue or strategic pages. Usually service pages, a couple of key category pages, and one or two cornerstone articles.
3. Count internal links pointing at each. Fewer than 10 from non-navigation pages? That's your first fix.
4. Count unique anchor variations. One or two anchors repeated everywhere means diversify. Aim for four to eight that all read naturally.
5. Check click depth. Anything important should be reachable within three clicks of the homepage. Deeper than that, restructure.
6. Find the orphans. Pages with zero internal links from anywhere. Link to them or remove them.
7. Audit your hubs. For each cluster, confirm members link up to the hub with varied anchors and the hub links back out.
8. Look at where links sit. For AI citations, get key links into the early body content. For Google, less critical, but still better than the footer.
9. Kill the mesh. Replace mass-link widgets and tag-page sprays with a curated block of three to five relevant links.
10. Rerun and measure. Track GSC impressions and clicks for your top five pages over 60 to 90 days, and check AI citation visibility separately.

If you do nothing else, do steps two and three. Across our audits, the most common outcome from simply fixing internal link count and anchor diversity on five pages was a solid lift inside 90 days. No new content. No new backlinks. Just better use of what's already there.

A note on E-E-A-T and internal linking

One pattern got stronger over the last year, so we'll flag it. Internal links from author pages, about pages and editorial-process pages seem to matter more than they used to, especially for content about health, money or law. If your medical, legal or financial pages aren't linked from an author bio with real credentials, or from an about page that establishes who's publishing, you're leaving a trust signal on the table.

It's the same hub-and-spoke logic, except the hub is the entity (the author, the organisation) and the spokes are everything they publish. For commercial content the equivalent is wiring case studies, testimonials and review pages into your service pages, with anchors that name the client and the outcome rather than a bare keyword.

The bit we find genuinely satisfying

Internal linking is the only major ranking lever you fully control. External backlinks need outreach and patience, which is exactly why we run a dedicated white-label link building service for agencies who'd rather not. Algorithm updates need nerves. Content needs writers. Internal linking needs an afternoon and a willingness to stop building mesh nobody asked for.

Get the patterns right and the rest of the SEO programme gets easier. Get them wrong and every other investment underperforms. That has been the most consistent finding across all 300 audits, and it's the one we keep coming back to with every new site.

If you want a second pair of eyes on your internal link structure, or you're not sure which pattern your site is closest to, our SEO team works exactly this into every engagement, and the structural changes usually do more heavy lifting in the first 90 days than anything flashier. Tell us about your site and we'll tell you, plainly, which of the four shapes you're living in and what we'd change first.