---
title: "Semantic Distance Modeling for Keyword Clustering"
description: "Use semantic distance and vector embeddings for keyword clustering, cluster validation, intent review, and evidence-based content mapping."
canonical: "https://nikoalho.fi/writing/semantic-distance-modeling/"
language: "en"
---

> Canonical source: [https://nikoalho.fi/writing/semantic-distance-modeling/](https://nikoalho.fi/writing/semantic-distance-modeling/)

[← writing](https://nikoalho.fi/writing/)

Lab Published 2026 · 05 · 20 Updated 2026 · 07 · 18

# Semantic distance modeling: grouping keywords for authority

Group queries by semantic distance instead of surface overlap. A practical embedding workflow for clustering, validation, and content mapping.

![Niko Alho](https://nikoalho.fi/assets/niko-alho-avatar-96.webp)

**Niko Alho**Operator in Turku · firsthand systems

ON THIS PAGE

[01 What is Semantic Distance in SEO? (The Theory)](#what-is-semantic-distance-in-seo-the-theory) [02 Visualizing Topic Clusters in 3D Space](#visualizing-topic-clusters-in-3d-space) [03 See the galaxy model in motion](#see-the-galaxy-model-in-motion) [04 How to Reduce Semantic Distance to Boost Authority](#how-to-reduce-semantic-distance-to-boost-authority) [05 The Math Behind the Model](#the-math-behind-the-model) [06 Simulate cosine similarity between two vectors](#simulate-cosine-similarity-between-two-vectors) [07 The Revenue Impact of Semantic Precision](#the-revenue-impact-of-semantic-precision) [08 How semantic distance maps to relevance](#how-semantic-distance-maps-to-relevance) [09 Stop Guessing, Start Modeling](#stop-guessing-start-modeling)

PROGRESS

0%

ON THIS PAGE 9 sections

[01 What is Semantic Distance in SEO? (The Theory)](#what-is-semantic-distance-in-seo-the-theory) [02 Visualizing Topic Clusters in 3D Space](#visualizing-topic-clusters-in-3d-space) [03 See the galaxy model in motion](#see-the-galaxy-model-in-motion) [04 How to Reduce Semantic Distance to Boost Authority](#how-to-reduce-semantic-distance-to-boost-authority) [05 The Math Behind the Model](#the-math-behind-the-model) [06 Simulate cosine similarity between two vectors](#simulate-cosine-similarity-between-two-vectors) [07 The Revenue Impact of Semantic Precision](#the-revenue-impact-of-semantic-precision) [08 How semantic distance maps to relevance](#how-semantic-distance-maps-to-relevance) [09 Stop Guessing, Start Modeling](#stop-guessing-start-modeling)

**TL;DR** The useful bits

-   8-min read
-   4 takeaways

1.  01 LSI keywords don't exist — that concept comes from a 1988 patent designed for closed databases, not the open web.
2.  02 Modern relevance is vector cosine similarity, not synonym sprinkling — measure mathematical distance between concepts, not string overlap.
3.  03 Whiteboarding 'hub and spoke' diagrams is doodling; semantic distance modeling makes cluster decisions mathematically defensible.
4.  04 Group keywords by vector proximity to the topic centroid; pages that score above the threshold belong in the cluster, the rest don't.

A/01 Direct answer

What is semantic distance modeling?

Semantic distance modeling is the mathematical process of calculating proximity between concepts in a high-dimensional vector space. In SEO, it quantifies how closely a keyword or page relates to a core topic — letting you build content clusters by vector distance instead of intuition or whiteboard diagrams.

**Semantic distance modeling is the mathematical process of calculating the proximity between two concepts within a high-dimensional vector space. In SEO, it quantifies how closely a specific keyword or page relates to a core topic entity, allowing architects to [build content clusters based on mathematical relevance](https://nikoalho.fi/writing/programmatic-seo-architecture/) rather than intuition.**

* * *

### The Brutal Truth: Your Content Clusters Are Built on Guesswork

Most SEO strategies are built on a whiteboard. You draw a “Hub,” draw some lines to “Spokes,” and congratulate yourself on a job well done. That is not architecture. That is doodling. While Google still uses traditional keyword matching (lexical search) for retrieval, its ranking engine has evolved significantly since the introduction of BERT in 2019. The algorithm doesn’t just look for strings of text; it calculates **Vector Space**. It measures the mathematical distance between the user’s query and your content’s meaning. If you rely solely on manual brainstorming to group keywords, you are failing. To dominate a vertical in 2026, you must stop thinking exclusively in keywords and start thinking in vectors. You need **Semantic Distance Modeling**.

* * *

## What is Semantic Distance in SEO? (The Theory)

We need to demystify the “black box” of relevance. For years, the industry relied on *Latent Semantic Indexing (LSI)* to explain how search engines understood context. Let’s be clear: LSI is deprecated technology. It is a concept from the 1980s designed for small, static databases. It has no place in a modern SEO conversation. Today, relevance is defined by **Vector Embeddings** and **Neural Matching**.

### From Strings to Things

When Google deployed BERT and later MUM, it shifted toward understanding “things” (entities). The search engine maps these entities as points in a multi-dimensional geometric space. In this space, concepts that are semantically similar are positioned closer together.

-   **High Similarity:** “SaaS” and “Subscription Revenue” are close neighbors.
-   **Low Similarity:** “SaaS” and “Cat Food” are miles apart.

### The Math of Relevance

“Relevance” is not a feeling. It is a calculation of **Cosine Similarity** between two vectors. The underlying representation is explained in [vector embeddings for search](https://nikoalho.fi/writing/vector-embeddings/), including what the dimensions encode and where retrieval systems use them. When a user searches for a solution, Google converts that query into a vector. It then scans its index for content vectors that align most closely with that query vector. Your goal as an SEO Architect is to minimize the semantic distance between your content ecosystem and the core entities you want to own. If your content clusters are loose—filled with fluff or irrelevant diversions—the average distance increases, and your authority signal dilutes. This is architecting authority stripped of the magic and reduced to its raw mechanics: reducing the distance between points in a dataset. A solid [semantic SEO strategy](https://nikoalho.fi/writing/semantic-seo-for-ai/) operationalizes this math into a repeatable framework for dominating topic verticals.

* * *

## Visualizing Topic Clusters in 3D Space

Stop visualizing your site structure as a flat sitemap. To engineer revenue growth, you must visualize your content as a 3D cloud of data points.

### The Galaxy Model

Imagine your core entity—the primary revenue driver (e.g., “Enterprise ERP”)—is the sun at the center of a solar system. Every supporting article, case study, or technical documentation page is a planet orbiting that sun.

-   **High Authority:** The planets (supporting content) orbit tightly around the sun. The semantic distance is short. The gravitational pull (relevance) is strong.
-   **Low Authority:** The planets are scattered. You have blog posts drifting into irrelevant topics. The distance is vast. The system collapses.

### The Void (Identifying True Gaps)

Traditional gap analysis involves looking at a competitor’s blog and copying what they wrote. This is reactive. In **vector space SEO** , a “content gap” is a literal void in your data cloud. By plotting your existing content vectors against the query vectors of your market, you can see empty spaces where users are searching, but you have no matching entity. This is how you win. You don’t write content to “fill a calendar.” You deploy assets to fill a coordinate in vector space.

> **Technical Note:** This approach aligns with Google’s patents regarding the Knowledge Graph. The engine assesses the “confidence score” of a relationship between two entities. If your content makes the connection explicit and mathematically proximate, the confidence score rises.

* * *

## See the galaxy model in motion

01

Visual model

Topic Authority as Gravitational Pull

Vector embedding proximity limits index dilution.

Programmatic  
SEO Architecture

Agentic Workflows  
Cos: 0.92

Vector DBs  
Cos: 0.88

Next.js Hydration  
Cos: 0.76

XML Sitemaps  
Cos: 0.71

Facebook Ads ROI  
Cos: 0.12 (Prune)

General Marketing  
Cos: 0.25 (Prune)

## How to Reduce Semantic Distance to Boost Authority

You cannot achieve this with a spreadsheet and a “gut feeling.” You need **Automated Topical Mapping** and programmatic execution. Here is the architecture for tightening your semantic signal.

### 1\. Automated Topical Mapping

Stop guessing which keywords belong in a cluster. Use Python libraries (like Scikit-learn) or OpenAI’s embedding API to automate the process. **The Workflow:**

1.  **Scrape the SERPs** for your target high-value queries.
2.  **Generate Embeddings** for the top-ranking pages.
3.  **Map the Average Vector.** This gives you the mathematical “center” of the topic.
4.  **Audit Your Distance.** Compare your current content’s vector against that average.

If your content vector is at a 0.75 distance (relative to your specific embedding model) and the market leader is at 0.15, you don’t need “better writing.” You need to re-engineer the semantic focus of the page.

### 2\. Architecture & Internal Linking

Internal links are often treated as navigation tools. In this model, they are **bridges in vector space**. When you link Page A to Page B, you are telling the search engine, “These two concepts are related.”

-   **High-Value Link:** Linking “Cloud Security” to “Data Encryption” reduces semantic distance.
-   **Toxic Link:** Linking “Cloud Security” to “Company Picnic Photos” introduces noise.

**Semantic Content Clusters** must be structurally sound. Links should flow vertically through the hierarchy (Parent to Child) and horizontally between highly similar vectors (Sibling to Sibling). Do not cross-link distinct clusters unless there is a calculated, mathematical overlap. *(Once the map is built, scale the connections using **[linking via vector embeddings](https://nikoalho.fi/writing/automating-internal-linking/)**.)*

### 3\. Pruning the Noise

This is the hardest pill for marketing teams to swallow. To reduce the average semantic distance of your domain, you must cut the outliers. If you are a B2B FinTech company, that blog post from 2021 about “Top 10 Coffee Shops for Remote Work” is a liability. It is a data point located far away from your core entity. It stretches your vector cloud, lowering the overall density of your authority signal. **The Directive:** Delete it. Pruning irrelevant content tightens the cluster and spikes the relevance of the remaining assets.

* * *

## The Math Behind the Model

We are not dealing with abstractions. We are dealing with linear algebra. The most common method for determining **semantic distance** is Cosine Similarity. You don’t need to be a mathematician, but you must respect the math that governs your revenue. $$ text{similarity} = cos(theta) = frac{mathbf{A} cdot mathbf{B}}{|mathbf{A}| |mathbf{B}|} $$ Where:

-   $mathbf{A}$ is the vector of the user’s query.
-   $mathbf{B}$ is the vector of your content.
-   The closer the result is to 1, the higher the relevance.

Google’s algorithms run variations of this calculation billions of times a day. If you optimize for keywords (strings), you are hoping for a match. If you optimize for vectors (concepts), you are engineering a mathematical inevitability.

* * *

## Simulate cosine similarity between two vectors

03

Working tool

Semantic Distance Simulator

Pillar Topic Node (Anchor)

Target Page Topic (Satellite)

Cosine Similarity Score

0.87

High relevance. Algorithm will automatically inject internal link.

## The Revenue Impact of Semantic Precision

Why should the CFO care about vector space? Because semantic precision is a proxy for **Operational Intelligence**.

### 1\. Efficiency & Crawl Budget

Tighter clusters are easier for bots to crawl. When the semantic distance is low, Googlebot understands the site structure instantly. You waste less crawl budget on low-value pages and get your money pages indexed faster.

### 2\. High Intent Conversion

There is a direct correlation between semantic relevance and user intent. A user searching for specific, technical solutions has a high-intent vector. If your content matches that precision, you aren’t just getting traffic; you are getting qualified leads.

### 3\. The Revenue Correlation

We can model organic growth potential through this heuristic: $$ Revenue propto frac{Authority}{Distance} $$ While not a literal law of economics, the correlation is clear: as you decrease the semantic distance between your content and the user’s need, your authority relative to that need increases. High authority leads to dominance. Dominance leads to revenue.

* * *

## How semantic distance maps to relevance

02

Reference table

| Cosine Similarity Score | Semantic Relationship | Automated Architecture Action |
| --- | --- | --- |
| 0.85 - 1.00 | **Tight Orbit (Core Cluster)**  
Topics are intrinsically linked. | **Hard Linking.** System automatically generates contextual anchor text and two-way internal links. |
| 0.50 - 0.84 | **Loose Orbit (Support Topic)**  
Related conceptually, but distinct. | **Category Association.** Included in XML sitemaps naturally, but no hard-coded contextual links injected. |
| 0.00 - 0.49 | **Semantic Outlier (Dilution)**  
Topic harms overall domain focus. | **Quarantine.** Agentic auditor flags page for review, 410 Deletion, or 301 Redirect. |

## Stop Guessing, Start Modeling

The era of “content is king” is dead. The king is dead; long live the **Model**. If you want to scale revenue in 2026, you cannot afford to treat SEO as a creative writing exercise. It is a data science problem.

-   **Audit your vectors.**
-   **Calculate your distance.**
-   **Prune the noise.**

Google is a machine. It does not feel; it calculates. If you want to rank, stop trying to be human and start speaking its language. **Audit your system. Engineer the result.** Written by ![Niko Alho](https://nikoalho.fi/assets/niko-alho-avatar-192.webp) Niko Alho Technical SEO specialist and AI automation architect. Building systems that drive organic performance through data-driven strategies and agentic AI. [Connect on LinkedIn →](https://www.linkedin.com/in/nikoalho/) Related Articles

Questions people actually ask

FAQ · 4

Q01 How does cosine similarity decide cluster membership? +

Embed every candidate page and the topic centroid. Pages with cosine similarity above a threshold (typically 0.78-0.85) belong in the cluster. The rest stay out or form their own cluster.

Q02 What's a topic centroid? +

The average vector of pages that definitively belong to a cluster. You compute it once, then measure new candidate pages against it. As the cluster matures, recompute the centroid.

Q03 Why is this better than 'hub and spoke' brainstorming? +

Brainstorming relies on human guesses about relationships. Vector distance gives you a defensible number — you can audit cluster decisions and prove that page X belongs with page Y.

Q04 Does Google literally use cosine similarity? +

Google uses neural matching and vector-based retrieval (since RankBrain and BERT), which approximate this behavior. The exact algorithm is proprietary, but the principle aligns with modern information retrieval.

Sources & further reading

1.  \[01\]
    
    [Vector embeddings explained](https://platform.openai.com/docs/guides/embeddings)
    
    OpenAI
    
    DOC
2.  \[02\]
    
    [BERT for search](https://blog.google/products/search/search-language-understanding-bert/)
    
    Google Blog
    
    BLOG

INBOX · TWICE A MONTH

Notes from the lab, in your inbox.

The same pipelines I run for paying clients — written up first for subscribers.

Written for operators, not marketers

![Niko Alho](https://nikoalho.fi/assets/niko-alho-avatar-192.webp)

Niko Alho

I run agentic SEO and build custom AI for B2B companies. Based in Turku.

[About →](https://nikoalho.fi/about/)

KEEP READING

## More on lab.

-   [
    
    Lab 2026 · 05 · 20
    
    Solving Next.js hydration issues that tank SEO
    
    Hydration mismatches tank INP and ghost pages from Google's index. Diagnose with DevTools and Search…
    
    read →](https://nikoalho.fi/writing/hydration-nextjs-seo/)
-   [
    
    Lab 2026 · 05 · 20
    
    Localization engineering: scalable multilingual SEO architecture
    
    Plan multilingual SEO as engineering: hreflang logic, URL topology, CMS fields, canonical rules, loc…
    
    read →](https://nikoalho.fi/writing/localization-engineering/)
-   [
    
    Lab 2026 · 05 · 20
    
    Predictive search analytics: forecast demand before it spikes
    
    Predictive search analytics use Python forecasting to catch demand before competitors react. The Pro…
    
    read →](https://nikoalho.fi/writing/predictive-search/)

[More writing →](https://nikoalho.fi/writing/)

Direct with Niko · 20-min intro, no pitch [Book a slot →](https://nikoalho.fi/book/)

## Structured data

```json
{
  "@context": "https://schema.org",
  "@type": "WebSite",
  "@id": "https://nikoalho.fi/#website",
  "url": "https://nikoalho.fi/",
  "name": "Niko Alho",
  "description": "Agentic SEO and custom AI builds for B2B companies.",
  "inLanguage": "en",
  "publisher": {
    "@id": "https://nikoalho.fi/#person"
  },
  "potentialAction": {
    "@type": "SearchAction",
    "target": {
      "@type": "EntryPoint",
      "urlTemplate": "https://nikoalho.fi/search/?q={search_term_string}"
    },
    "query-input": "required name=search_term_string"
  }
}
```

```json
{
  "@context": "https://schema.org",
  "@type": "Person",
  "@id": "https://nikoalho.fi/#person",
  "name": "Niko Alho",
  "givenName": "Niko",
  "familyName": "Alho",
  "url": "https://nikoalho.fi/about/",
  "image": "https://nikoalho.fi/og/default.png",
  "jobTitle": "Agentic SEO & Custom AI Consultant",
  "email": "mailto:contact@nikoalho.fi",
  "telephone": "+358401539426",
  "address": {
    "@type": "PostalAddress",
    "addressLocality": "Turku",
    "addressCountry": "FI"
  },
  "knowsAbout": [
    "Search Engine Optimization",
    "Agentic SEO",
    "Topical Authority",
    "Retrieval-Augmented Generation",
    "Large Language Models",
    "Custom AI Builds",
    "B2B SaaS Content Strategy",
    "Schema.org Structured Data",
    "Generative Engine Optimization"
  ],
  "knowsLanguage": [
    "en",
    "fi"
  ],
  "worksFor": {
    "@id": "https://nikoalho.fi/#organization"
  },
  "sameAs": [
    "https://www.linkedin.com/in/nikoalho/",
    "https://github.com/alhoniko"
  ]
}
```

```json
{
  "@context": "https://schema.org",
  "@type": "ProfessionalService",
  "@id": "https://nikoalho.fi/#organization",
  "name": "Niko Alho — SEO & AI Automation",
  "alternateName": "Niko Alho",
  "description": "Agentic SEO and custom AI builds for B2B companies.",
  "url": "https://nikoalho.fi/",
  "image": "https://nikoalho.fi/og/default.png",
  "logo": "https://nikoalho.fi/assets/logo-mark.svg",
  "email": "mailto:contact@nikoalho.fi",
  "telephone": "+358401539426",
  "priceRange": "$$$",
  "founder": {
    "@id": "https://nikoalho.fi/#person"
  },
  "employee": {
    "@id": "https://nikoalho.fi/#person"
  },
  "knowsLanguage": [
    "en",
    "fi"
  ],
  "address": {
    "@type": "PostalAddress",
    "addressLocality": "Turku",
    "addressCountry": "FI"
  },
  "areaServed": [
    {
      "@type": "City",
      "name": "Turku"
    },
    {
      "@type": "City",
      "name": "Helsinki"
    },
    {
      "@type": "Country",
      "name": "Finland"
    },
    {
      "@type": "Place",
      "name": "European Union"
    },
    {
      "@type": "Place",
      "name": "Worldwide (remote)"
    }
  ]
}
```

```json
{
  "@context": "https://schema.org",
  "@type": "TechArticle",
  "@id": "https://nikoalho.fi/writing/semantic-distance-modeling/#article",
  "headline": "Semantic distance modeling: grouping keywords for authority",
  "name": "Semantic distance modeling: grouping keywords for authority",
  "description": "Group queries by semantic distance instead of surface overlap. A practical embedding workflow for clustering, validation, and content mapping.",
  "image": "https://nikoalho.fi/og/semantic-distance-modeling.png",
  "url": "https://nikoalho.fi/writing/semantic-distance-modeling/",
  "datePublished": "2026-05-20T00:00:00.000Z",
  "dateModified": "2026-07-18T00:00:00.000Z",
  "inLanguage": "en",
  "isAccessibleForFree": true,
  "wordCount": 1487,
  "articleSection": "Lab",
  "keywords": "semantic distance modeling, semantic clustering, keyword clustering, vector keyword clustering, embedding seo",
  "author": {
    "@id": "https://nikoalho.fi/#person"
  },
  "publisher": {
    "@id": "https://nikoalho.fi/#person"
  },
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://nikoalho.fi/writing/semantic-distance-modeling/"
  },
  "about": {
    "@type": "Thing",
    "name": "Lab"
  },
  "speakable": {
    "@type": "SpeakableSpecification",
    "cssSelector": [
      "h1",
      ".tldr",
      ".article-body > .prose > p:first-of-type"
    ]
  },
  "citation": [
    {
      "@type": "CreativeWork",
      "url": "https://www.linkedin.com/in/nikoalho/"
    }
  ]
}
```

```json
{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    {
      "@type": "ListItem",
      "position": 1,
      "name": "Home",
      "item": "https://nikoalho.fi/"
    },
    {
      "@type": "ListItem",
      "position": 2,
      "name": "Writing",
      "item": "https://nikoalho.fi/writing/"
    },
    {
      "@type": "ListItem",
      "position": 3,
      "name": "Semantic distance modeling: grouping keywords for authority",
      "item": "https://nikoalho.fi/writing/semantic-distance-modeling/"
    }
  ]
}
```

```json
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "How does cosine similarity decide cluster membership?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Embed every candidate page and the topic centroid. Pages with cosine similarity above a threshold (typically 0.78-0.85) belong in the cluster. The rest stay out or form their own cluster."
      }
    },
    {
      "@type": "Question",
      "name": "What's a topic centroid?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "The average vector of pages that definitively belong to a cluster. You compute it once, then measure new candidate pages against it. As the cluster matures, recompute the centroid."
      }
    },
    {
      "@type": "Question",
      "name": "Why is this better than 'hub and spoke' brainstorming?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Brainstorming relies on human guesses about relationships. Vector distance gives you a defensible number — you can audit cluster decisions and prove that page X belongs with page Y."
      }
    },
    {
      "@type": "Question",
      "name": "Does Google literally use cosine similarity?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Google uses neural matching and vector-based retrieval (since RankBrain and BERT), which approximate this behavior. The exact algorithm is proprietary, but the principle aligns with modern information retrieval."
      }
    }
  ]
}
```