Article 9 min read
Internal Links: The Question That Decides Where Each Link Points
by Daniel Sócrates published
Every page on your site can point to another page on your site. Each of those pointers is an arrow the reader follows with one click. This is called an internal link.
Almost every article on the subject ends with the same advice. Put three to five links in each post. Point to related content. Use descriptive anchor text. Don’t overdo it.
None of that is wrong. It just never answers the question that matters: a link to which page, and why. Almost everything on your site relates to almost everything else, if you stretch far enough. Without a better yardstick, linking turns into a matter of taste. You end up choosing links at the end of the draft, when you are already tired.
The accounting firm that published for two years
Imagine an accounting firm that has posted weekly for two years, never missing one. There is an article on income taxes. There is a guide to starting a business. There is the post about the staff holiday party, with a photo of the whole team. There is the piece on artificial intelligence that everyone wrote that month. There is the Mother’s Day tribute. 1
None of these posts is a mistake. Someone signed off on each one, for a sensible reason. Now look at the end of the guide to starting a business. It carries three links: the holiday party, the AI piece, and the Mother’s Day tribute. They are all “from the firm,” so every one of them passed the relatedness test.
The corner butcher shop started selling cell phones. First came a display case of chargers by the register, because the margin was good. Then two shelves went up. Today a new customer needs fifteen seconds to figure out what the place sells. 1
None of those decisions was dumb. Added together, they erased what the shop stood for. The three links at the end of the guide do the same thing, one arrow at a time.
The center and the radius, in a few lines
Picture two pins stuck in a map of meaning. One stands for your whole site. The other stands for the page you published this morning. There is a distance between them. Call the site’s pin the center. Call how far your pages drift from it the radius.
The picture comes with field names. The Content Warehouse API documentation, leaked in 2024,
holds a field called siteFocusScore. Its description reads: “Number denoting how much a site
is focused on one topic”. 2
The same module holds siteRadius: “The measure of how far page_embeddings deviate from the
site_embedding”. 3
Two caveats travel with those names. The leak shows what is stored, and it never shows the weight. Whether Google uses any of this today, in the search you ran this morning, nobody on the outside knows. 4 And no Google dashboard displays these values. Any “focus score” for sale is an estimate from the tool that calculates it. 5
The full explanation of these fields is in the piece on semantic silos. That is where you will find the module, the limits, and the other proofs of site-level judgment. From here on, I use only the picture: a center, and a distance to it.
What nobody documented
Here I have to separate what I know from what I believe. There is no documentation tying internal links to these fields. The module that stores focus has five fields, and none of them mentions links. 6 The descriptions talk about embeddings, the points of meaning for the site and for its pages.
If someone tells you internal links shrink your siteRadius, be suspicious. That person is
making up precisely the part that matters.
What I claim is more modest, and it is editorial practice. My book hangs one question over every new piece. Does it move closer to the center, or does it widen the radius? I carry that same question to the moment you place a link. 7
The question behind an internal link stops being “is this related?” and becomes “does this move toward the center or away from it?”
The center, here, is the subject your site answers without needing an introduction. A link moves toward it when it takes the reader closer. It moves away when it takes the reader to a corner your business does not serve. This reading is mine. It organizes your decision and promises no ranking at all.
Three decisions the question changes
What receives links
In a structure with no declared center, every page links to every other page. What you get is a web with no hierarchy, where no page speaks for the subject. With a declared center, one page stands for the whole territory, and the others point to it. It becomes the front door, the place where the reader learns what that part of the site covers.
This page follows the same rule. It points to the front door of the semantic SEO silo, where the subject begins. It also points to the piece on silos, where the fields are explained. A link to a sales page would pull the reader away from what they came for.
What to do with the piece that sits far away
The holiday party post can stay live. What changes is its place in the web of links. The pages at the center stop pointing to it, and it leaves the income tax neighborhood. It keeps doing its job, for the staff and for existing clients.
The same goes for the piece on that month’s hot topic. It brought in visits that month, and nobody needs to pretend otherwise. But the reader who arrived through the guide to starting a business wanted to start a business.
How to read the count
Five links in one piece, pointing to five different subjects, scatter the reader. Two links, one to the center and one to a sibling page, keep the reader on track. The count alone says little. The direction of the arrows decides.
Two kinds of spread the industry mixes up
The same documentation has another field that measures spread, siteQualityStddev. It lives in
a different module, QualityNsrNsrData, and its description is about how quality scores deviate
across a site’s pages. 8
These are two different kinds of spread. The first is about subject: your pages cover topics that drift apart from one another. The second is about quality: half the site is very good, and the other half was rushed out in 2019.
One law firm published eighty articles, all on family law. Forty are flawless, and forty were written just to hit a quota. Small radius, scattered quality.
A second firm also published eighty flawless articles. Forty are on family law, and forty are on legal marketing. Even quality, wide radius. 1
When you place links, mixing the two up is expensive. An excellent article on legal marketing still sits far from the center of a family law firm. Being good does not earn it the arrows from the central pages. A weak article on joint custody still sits at the center, and it needs a rewrite. Removing the links to it only hides the page your subject needs most.
That is why the rule “delete everything that is off topic” appears in no source at all. Whoever repeats it has merged two fields that live in different modules. 8
Ten titles, one sentence, and the arrows
You can do this before your coffee gets cold. It takes ten minutes, and no tools at all. 9
Pull up the list of the last ten pieces you published. Just the titles. Don’t open the texts, don’t look at metrics, don’t judge quality. Write the ten titles in a column.
Underneath, write a single sentence that describes what the ten are about together. No semicolons, no list dressed up as a sentence, no “and so on.” If the sentence needs an “and also,” you just measured your radius by hand. Every “and also” is a piece pulling the average somewhere else.
Next to each title, note who pays your bills: client, prospect, or nobody. You will often find that the pieces farthest from the center got the most praise inside the company.
Now you can open the ten texts, one at a time. Draw the arrows. For every internal link you find, draw an arrow from the title to the destination page. If the arrows form a web with no direction, you have ten pages and no center. If they converge on one or two pages, you have a territory, and those pages are its front door.
Last, look at the arrows that leave a piece marked “client” and land on a piece marked “nobody.” Those are the first candidates to go. 9
This sheet is a long way from the value Google stores. You built, by hand, a rough version of the question that field asks. A broad sentence is no diagnosis of illness either. A general-interest publisher has a broad sentence by design, and that is fine.
What changes on Monday
Nobody can optimize a number that stays hidden, with no scale and no disclosed weight. You leave here with a criterion for the moment your cursor pauses over a word. The question fits on one line: does this link take the reader closer to the center?
Before you redraw the arrows, check that your pages make it into the index. A well-drawn territory with pages outside the index exists only inside your site. The ten-minute diagnosis is in how to get your site on Google. The crawler’s path to each page is in crawling and crawl budget.
Frequently asked questions
How many internal links should I put in each piece?
I have no source for that number, and I will not invent one. Look at the direction of each arrow. Every link should take the reader closer to the center of the site, or to a sibling page on the same subject. That criterion is my own reading, stated as such. 7
Do internal links improve my site’s siteRadius?
No documentation ties internal links to that field or to siteFocusScore. The descriptions talk
about page and site embeddings, and no field in the module mentions links. 3 6
Should I delete the pieces that sit far from my main subject?
That rule appears in no source. Distance from the subject and quality score are different measures, stored in different fields. A piece far from the center can stay live, outside the territory’s link neighborhood. 8
Notes and sources
- Scenes of the accounting firm, of the butcher shop that started selling cell phones, and of the two law firms: Daniel Sócrates, “AI Search Optimization” (“SEO de Entidade”, Portuguese edition), 2026, chapter 8, sections “A média de tudo o que você publicou” (The Average of Everything You Published), “Duas placas na mesma rua” (Two Signs on the Same Street), and “Duas coisas espalham, e elas são diferentes” (Two Things Spread, and They Are Different). They are illustrative scenes from the book, with no real client behind them. The three links at the end of the guide to starting a business are the author’s extension of the scene in this piece.
siteFocusScore, moduleGoogleApi.ContentWarehouse.V1.Model.QualityAuthorityTopicEmbeddingsVersionedItemin the documentation of the Content Warehouse API leaked in 2024. Literal description quoted, the same as in the piece on semantic silos. Live verification on the public mirror of the documentation on August 1, 2026, recorded in note 1 of chapter 8 of the book.siteRadius, same module, with the literal description quoted, the same as in the piece on semantic silos. Live verification on August 1, 2026, chapter 8, note 2.- The leak shows what is stored and never the weight. No field carries a percentage, a public scale, or a disclosed target, and the material is a snapshot of 2024 architecture. The existence of the fields is a checkable fact; current use by Google is unknown to anyone on the outside. Chapter 8, note 7, and section “O pino do site e o pino da página” (The Site Pin and the Page Pin).
- No Google API, dashboard, or report exposes
siteFocusScoreorsiteRadius. Any number sold under that name is an estimate from the tool that calculates it. Chapter 8, note 18. - The module has five fields:
siteFocusScore,siteRadius,siteEmbedding,pageEmbedding, andversionId(chapter 8, notes 1 to 3). None of them concerns links. No source in chapter 8 ties internal links to these fields; a literal search for “link” in the Portuguese text of the chapter on September 11, 2026 found no relevant occurrence. The documentation was not reopened for this piece, and the absence holds for what the book verified. - The question “Isso aproxima do centro ou aumenta o raio?” (Does this move closer to the center, or does it widen the radius?) is in chapter 8, section “Qual é o seu assunto, em uma frase?” (What Is Your Subject, in One Sentence?), framed for the decision of what to publish. Carrying that question over to the internal linking decision, and the three decisions that follow from it, is the author’s reading in this piece.
siteQualityStddev, inGoogleApi.ContentWarehouse.V1.Model.QualityNsrNsrData, described in chapter 8, note 6, as an estimate of the spread of quality scores across a site’s pages. It appears here in indirect speech. The distinction between topical spread and quality spread, and the finding that the rule “delete everything that is off topic” exists in no source, come from chapter 8, section “Duas coisas espalham, e elas são diferentes” (Two Things Spread, and They Are Different). The application to the linking decision is the author’s reading.- Exercise of the ten titles, the single sentence, and the column “quem paga a conta” (who pays the bills): chapter 8, section “Dez títulos e uma frase” (Ten Titles and One Sentence). The arrows step, and the reading of arrows that run from a piece marked “client” to a piece marked “nobody,” are the author’s extension for this piece and are not in the book.
Want to see SEO Genome more often on Google? Add it as a preferred source.
Let us look at what the machine understands about your company
A diagnostic conversation, with no slide deck. You bring your website address. We bring what Google and the AI assistants already know about it today.