What your internal links tell a machine about your site.

Whether other sites link to you, how a ranking system weighs a page, whether an answer engine selects a source — none of that is yours to set. Your own internal links are the exception. You choose which pages exist, which ones point at which, and what the pointing says.

Google describes links as a signal used both to find new pages and to judge what a page is relevant to, and it is explicit that only an anchor element with an href is reliably followed. Elements styled to look like links, or wired with a click handler and no href, may not be followed at all.

That makes internal linking unusually cheap and unusually neglected. It costs nothing but attention, it is entirely under the business’s control, and on most sites it is left to whatever the navigation happens to do.

A quick audit: pick any page and list the internal links inside the body text, excluding the header, footer and any component that appears on every page. On a lot of sites the list is empty, which means the page asserts no relationship to anything.

Site-wide navigation links every page to the same set of destinations. That is useful for people and almost uninformative as a statement about any particular page, because the same claim is being made from everywhere, including pages where it is irrelevant.

A link inside body text is different. It appears because this specific page had a reason to point there, at a specific moment in the argument, with anchor text chosen for that context. That is a claim about a relationship rather than a claim about the site.

The practical rule follows: if the only path from page A to page B runs through the navigation, the site has not said that A and B are related. It has said that B is reachable.

Anchor text is the short description you attach to the destination, and it is read by people deciding whether to click and by systems deciding what the destination is about. Google’s guidance asks for anchor text that is descriptive, reasonably concise, relevant to both pages and understandable out of context — and specifically advises against empty anchors and generic phrases such as “click here” or “read more”.

The same guidance warns against the opposite failure: stuffing the anchor with terms, chaining several links together with no text between them, and leaving a link without enough surrounding context to be interpreted.

The workable standard is a phrase that would still make sense read aloud on its own. “How to decide what your site should be found for” is a claim about a destination. “This article” is a pronoun.

Link structure is architecture. The arrangement a site falls into by default is rarely the one it would choose, and each of these communicates something different about where authority sits.

• FLAT

  Everything links to everything. Cheap to maintain, and it asserts nothing: if all pages are equally related to all others, no page is identifiably the authority on anything.

• HUB AND SPOKE

  One page owns a subject and the supporting pages link up to it and to each other where their contexts meet. The most legible arrangement, and the one that matches how people actually read a topic.

• SEQUENTIAL

  Pages link forward and back along a path. Genuinely useful where the order is load-bearing, and misleading where it is not — it implies a progression the content may not have.

• ORPHANED

  A page nothing links to. It may be in the sitemap and still be unreachable by anything that follows links, and it is invisible to the argument the rest of the site is making.

There is no published number of internal links a page should carry, and anybody quoting one is inventing it. What is observable is the effect of too many: a paragraph carrying four links is harder to read, each link is less likely to be followed, and the emphasis the links were supposed to create is gone.

A more useful constraint than a count is a reason. Each link should exist because the sentence around it raised a question the destination answers. Links that survive that test tend to be few and to be followed; links that fail it are decoration that costs reading.

The same applies to reciprocity. Two pages should link to each other when each genuinely raises the other’s question, which is often but not always. Adding the reverse link mechanically produces pairs where one direction makes no sense to the reader who is there.

A full link audit produces a large spreadsheet and rarely produces a decision. Working outward from the page the business most wants found produces both, and it finishes.

1. 01

   Name the page that should be the authority

   Decides

   Which page do we want to be the answer for this subject?

   Output

   One hub per subject, explicitly chosen rather than inherited.

2. 02

   List what it should be reachable from

   Decides

   Which existing pages raise the question this hub answers?

   Output

   A short list of genuine referrers, not every page on the site.

3. 03

   Place the link in the sentence, not at the end

   Decides

   Where in the argument does the question actually arise?

   Output

   Body-text links with context on both sides.

4. 04

   Rewrite the anchors as claims

   Decides

   Would this anchor make sense read aloud on its own?

   Output

   Descriptive anchors, no generic phrases, no chaining.

5. 05

   Find the orphans

   Decides

   Which pages does nothing link to?

   Output

   Each orphan linked from somewhere relevant, or retired.

6. 06

   Check the link works without JavaScript

   Decides

   Is it an anchor element with an href?

   Output

   Links that can actually be followed, rather than click handlers that look like links.

Each of these is visible on sites that have clearly put effort in, which is what makes them worth naming — the effort was real and it went somewhere that does not pay.

• THE RELATED-POSTS BLOCK

  An automated row of links at the foot of every article, chosen by tag rather than by relevance. It is a component, not a claim, and readers learn to skip it.

• COUNTING INSTEAD OF DECIDING

  A target of so many internal links per page. Hitting a number produces links nobody needed and does not produce the one link that mattered.

• LINKS THAT ARE NOT LINKS

  A span with a click handler, a styled div, a URL that only resolves in the client. It looks identical to a person and is not reliably followed.

• THE SITEMAP AS A SUBSTITUTE

  Assuming a page listed in the sitemap is covered. A sitemap is a list of addresses; it makes no statement about how the pages relate.

It makes a subject legible. A hub with genuine supporting pages linking into it gives a reader a path and gives a system repeated, consistent evidence that one page is the centre of a subject rather than one of twelve equivalent pages.

It also exposes architecture problems that are otherwise invisible. If two pages keep wanting to link to each other with the same anchor text, they are answering the same question and one of them should absorb the other — which is a coverage decision that link work surfaces for free.

None of this is a ranking technique and it should not be sold as one. It is the site explaining its own structure, which is worth doing whether or not anything downstream rewards it.

1. Internal links are the one discovery signal a business fully controls — and on most sites they are left to whatever the navigation happens to do.
2. A site-wide navigation link says a page is reachable; a body-text link says this page is related to that one, and why.
3. Google asks for anchor text that is descriptive, concise and understandable out of context, and advises against empty anchors and phrases like “click here”.
4. Only an anchor element with an href is reliably followed; styled elements and click handlers may not be.
5. Hub and spoke is the most legible arrangement; flat linking asserts nothing and orphaned pages are invisible to anything that follows links.
6. There is no published limit on internal links per page. The real constraint is that each one should answer a question the surrounding sentence raised.
7. Two pages that keep wanting to link to each other with the same anchor are answering the same question, and one should absorb the other.