For months, our internal linking process for Photovoltaik.info felt efficient. We had clear rules: identify primary keywords in a new article and link them to existing pages targeting the same terms. It was a logical, repeatable system, but I started noticing a limitation. The system was connecting words, not ideas. It could link an article about „solar panel costs“ to another about „inverter costs“ because they both contained the keyword „costs,“ but it couldn’t grasp the deeper user journey—that someone researching costs might next be interested in installation steps or government subsidies, even if the keywords didn’t perfectly match.
Our system was technically correct but contextually blind. It was building a library, but it wasn’t building understanding. This realization prompted a new experiment: could we teach our system to link based on user intent rather than simple keyword repetition?
From Matching Words to Understanding Purpose
The problem with keyword-based linking is that it only operates on the surface of the content. It’s a relic from an older version of the web, when search engines were less sophisticated. Today, that approach can lead to what I call „hollow connections“—links that look relevant to a machine but offer little real value to a human reader. Research backs this up; while experts like Google’s John Mueller have confirmed that internal linking is „one of the biggest things“ for SEO, many automated tools still rely on simple keyword matching, which can create irrelevant or over-optimized anchor text clouds.
We saw this in our own analytics. Users would arrive on a page, but the internal links we provided weren’t leading them to the next logical step in their research. Click-through rates on these links were lower than we’d hoped, and user paths through the site felt more random than guided. The system was creating a web of links, but not a clear pathway for learning. We needed to evolve from a simple matching engine into a contextual reasoning engine.
Designing a System That Connects Intent
The goal was to create an automated linking layer that thought more like an editor and less like a database. We decided to build our logic around a semantic model, focusing on the meaning and contextual relationship between pages. This required moving from asking, „Does Page B contain the keyword from Page A?“ to asking, „Does Page B answer the next logical question for a user reading Page A?“
This shift became the core of our new system, which we built on two foundational elements:
-
Structured Data as the Language of Intent: Before the automation could work, we had to teach the system what each piece of content was about. This is where our approach to structured data proved essential. Every article was tagged not just with keywords but with its core user intent (e.g., „informational – cost analysis,“ „transactional – find installer,“ „navigational – brand comparison“). This gave the system a much richer dataset for understanding each page’s purpose.
-
Semantic Similarity Scoring: Instead of just matching keywords, our new engine analyzed the entire text of two articles to generate a „semantic similarity“ score. Using models that understand context, it could determine that an article about „solar panel efficiency“ was highly relevant to one about „best solar panels for cloudy regions,“ even if their primary keywords differed. This allowed for more nuanced and genuinely helpful connections.
This combination let us build more intelligent logic. The system could now identify that a user reading an introductory article on „how solar panels work“ (Intent: Foundational Knowledge) would likely benefit from a link to „calculating your home’s solar needs“ (Intent: Practical Application) far more than another basic definitional article. It was the difference between keeping a user in a loop and guiding them forward.
Insight: Smart Systems Are Built on Constraints
This project reinforced a core principle of system design: intelligence often emerges from well-defined constraints, not from absolute freedom. By constraining our system to prioritize user intent over simple keyword availability, we forced it to make smarter, more editorially sound decisions. It stopped looking for every possible keyword match and started searching for the best possible connection to serve the user on their learning journey.
The result was a system that began to build a truly cohesive user experience on its own. It was a foundational step in building a scalable content system that doesn’t just grow, but grows smarter. The ultimate insight is that the best automation doesn’t just replicate manual tasks; it elevates the logic behind them. We didn’t just automate linking; we automated a user-centric editorial strategy.
Frequently Asked Questions
What is internal linking, and why is it important?
Internal linking is the practice of connecting one page on your website to another page on the same site. It’s important for two main reasons: it helps search engines like Google discover all of your pages and understand the relationships between them, and it guides your visitors to other relevant content, keeping them engaged and helping them find what they need. As some studies have shown, a strategic approach can boost traffic significantly.
Why isn’t keyword-based linking enough anymore?
While not useless, relying only on matching keywords is an outdated strategy. Search engines have become much better at understanding the overall topic and intent of a page, not just the keywords it contains. An intent-based approach is more aligned with how modern search works and, more importantly, it creates a better, more logical experience for your human readers by anticipating their next question.
Can a system truly understand „user intent“?
In a way, yes. A system can’t understand intent in the human sense of having consciousness or feelings. However, by using structured data and semantic analysis, we can program it to recognize patterns that correspond to user intent. By categorizing content (e.g., „how-to guide,“ „product review,“ „cost breakdown“) and analyzing the language used, the system can make highly accurate predictions about what a user reading that page is likely to want next.
