From Keywords to Entities: The Evolution of Search Understanding 

“Things, not strings.”

This was Google’s famous quote referring to the launch of its Knowledge Graph back in 2012. 

It meant that, for the first time, Google’s algorithm could comprehend real-world entities (people, places, things, etc.) instead of just matching strings of keywords. 

It was also the birth of entity SEO, which laid the foundation for AI-powered platforms like ChatGPT, Perplexity, and Google’s AI Overviews. 

That’s because knowledge graph association is how machines like large language models (LLMs) can properly identify entities. 

With the prevalence of AI Overviews and AI search tools, marketers need to emphasize entity recognition and entity reinforcement over raw keyword research. 

Why?

Because establishing yourself as a relevant entity in your niche is how you build enough credibility to earn AI citations and brand recommendations. 

This marks a stark shift from the largely keyword-focused days of legacy SEO. 

In the following guide, we’ll cover how search evolved from keywords to entities, including how you can put entity SEO into practice today. 

The Keyword Era (Pre-2013, Early 2000s) 

Let’s start by examining the keyword era of SEO, which is everything that happened before Hummingbird’s introduction in 2013. 

While 2012’s Knowledge Graph was a solid step in the direction of semantic search, Google didn’t make semantic understanding a full-blown part of its algorithm until 2013. 

Before that, the way search algorithms ranked content was pretty basic.

In particular, Google’s algorithm relied on things like exact-match keywords and authority signals coming from backlinks. 

The main takeaway?

Keywords were king, and the more exact-match keywords you could stuff inside your content, the better. 

It was also standard practice to jam keywords inside of meta descriptions and meta tags. 

While professional marketers always focused on content quality to ensure a solid user experience alongside top rankings, the sheer simplicity of the system attracted lots of spammers and ‘black-hat’ practitioners. 

Even if the inclusion of repetitive keywords made content awkward, things like content quality and helpfulness were often an afterthought (if even considered at all). 

The keyword era’s formula for success

The formula went something like this:

1. Conduct keyword research to discover keywords that were trending with your target audience. Let’s use the example keyword ‘dental services St. Petersburg’. 
2. Marketers would then create content containing the exact phrase in strategic areas like the title, first 100 words, and numerous times in the body of the content (like ‘Providing the best dental services St. Petersburg has ever seen,’ etc). 
3. Pair the keyword-heavy content with a large volume of backlinks.

For many years, this recipe produced consistent rankings, even if the content wasn’t very helpful.  

Google was well aware of the rampant rise in spam, which is why it eventually moved toward an entity-based model instead of focusing on keywords. 

In the meantime, Google released the Penguin update as an attempt to curb the amount of manipulative content appearing in its results. 

The first step of the evolution into entity SEO was the introduction of the Knowledge Graph in 2012, but the first major change came the very next year. 

The Rise of Semantic Search (2013 – 2015) 

In 2013, Google released the Hummingbird update, which added semantic understanding to Google’s core ranking system. 

The previously released Knowledge Graph was only a search results feature and didn’t impact the ranking of results. 

It was the first time that Google’s algorithm had any form of natural language processing. Before Hummingbird, words were just words. The algorithm had no way of interpreting the meaning behind the words or the connections the words had to other entities. 

Suddenly, users were able to use longer, more conversational queries on Google, and the algorithm was able to understand what they meant. 

Hummingbird also introduced the concept of search intent, or the ‘why’ behind a user’s search. The algorithm could now tell if a user wanted to buy something, find information, or navigate to a certain website. 

This led to an improved user experience and higher levels of user satisfaction with Google’s search results. 

It was also not as easy to game the system, which made keyword stuffing strategies less effective.

Instead of cramming keywords into content, it was more important to produce content that was actually relevant and helpful.  

AI-Powered Entity-Based Search (2015 – Present) 

2015 saw the release of RankBrain, which officially introduced AI and machine learning to Google’s ranking system. 

While Hummingbird was adept at interpreting most searches, it was thrown off by some types of long-tail queries. RankBrain was introduced as a component inside the broader Hummingbird algorithm to enhance its semantic capabilities. 

It’s also able to learn from user behavior in real-time, and it adapts accordingly. For example, if RankBrain notices that you choose certain types of results for similar queries, it’ll prioritize those pages for future searches. 

In that sense, RankBrain is able to personalize each results page based on user preferences. 

With the Knowledge Graph, Hummingbird, and RankBrain on board, Google was finally starting to look more like an entity-based search engine than a keyword-focused one. 

Yet, the final transition into full-fledged entity SEO didn’t come until the introduction of Google’s AI Overviews (AIOs) in 2024. At this point, AI search tools like ChatGPT already existed for several years, but they paled in comparison to Google’s popularity. 

It wasn’t until AIOs overshadowed the organic results that AI SEO became the new norm. 

Organic traffic began to nosedive, and search marketers had to quickly adapt their strategies or become obsolete. 

Check out this graph showing how AIOs impacted the #1-ranked positions’ click-through rate on Google (source: Ahrefs):

The keyword-based tricks of the past were suddenly rendered useless, and visibility began to hinge on AI citations instead of keyword rankings. 

That’s because the LLMs that power AI search tools use knowledge graphs to identify entities and understand the relationships they share with one another. This directly ties into how they generate answers, pull sources, and weigh credibility. 

In other words, entity recognition is how algorithms and LLMs answer the question, “Which thing do you mean?”

For instance, imagine that an LLM encounters the word Paris in a sentence. Entity-wise, this could refer to several things. 

There’s the city in France, but there’s also the celebrity Paris Hilton, the Greek mythological figure Paris, and other cities named Paris, just to name a few. 

With named entity recognition, LLMs are able to correctly identify which Paris you mean, given any context. 

AI Overviews and Entities 

Google’s AI Overviews are powered by its internal LLM, Gemini. 

What makes Gemini especially powerful is its direct access to Google’s massive Knowledge Graph, which it uses to recognize entities, understand their relationships, and generate answers. 

Yes, that means entity recognition plays a huge role in improving your visibility in AIOs. 

If you want Gemini to cite your content and recommend your brand, you must establish yourself as an authoritative entity in your industry. 

Basically, you want to train Gemini to associate your brand with your niche and your target audience. You want it to view your content as trustworthy and authoritative, and consider your products and services high-quality. 

How do you achieve this?

You have to exhibit certain trust signals, which are similar to ranking factors in traditional SEO. 

Gemini and other LLMs trust content that:

• Has strong entity associations like brand mentions on relevant websites, positive community discussion, and a library of insightful, authoritative content. 
• Contains structured data like semantic HTML and schema markup since they both reinforce entity recognition and make content easier to parse. 
• Clearly answers questions and provides concise definitions without beating around the bush. 
• Contains E-E-A-T (experience, expertise, authoritativeness, and trustworthiness) characteristics. 

You can think of these signals as your training tools. The more you exhibit them, the more you can train LLMs to recognize your brand as a relevant, trustworthy entity. 

How to Optimize for Entities 

Next, it’s time to learn how you can optimize your online presence for entity alignment and reinforcement. 

We’ve already mentioned the tools, but there are marketing techniques that coincide with each one. Here are some of the most effective entity SEO tactics. 

Creating topical clusters

Instead of domain authority, marketers should focus on building topical authority instead. Topical authority is where algorithms and LLMs view you as a ‘go-to’ source for information on a particular topic. 

For example, if your blog has a ton of content covering various fitness routines, it can establish topical authority for fitness-related prompts. 

One of the best ways to build topical authority is by creating content clusters. These are interrelated blogs that cover similar topics. 

A content cluster consists of a pillar piece that introduces the main topic, and a series of sub-pieces that delve deeper into each subtopic. 

An example content cluster could be:

1. Pillar piece: A Detailed Overview of Digital Marketing 
2. Cluster page: The Basics of SEO and Social Media Marketing 
3. Cluster page: Why Email Marketing Still Yields Results 
4. Cluster page: Top 10 Digital PR Tactics for 2025 

As you can see, each cluster page covers a different aspect of digital marketing, which is the main topic. 

If you want algorithms and LLMs to pick up on your content clusters, they must be interlinked, so don’t forget to include internal links on each page. 

Including structured data 

Semantic HTML and schema markup are both extremely important for aiding entity recognition. Semantic HTML simply refers to HTML elements that contain descriptors, like . Nonsemantic HTML does not describe what it contains, like