What Is Semantic Search and How Can it Be Used in SEO?

What Is Semantic Search and How Can it Be Used in SEO?

Since Google has decided to gradually refuse the traditional search by keywords in favor of more advanced algorithms, the concept of semantic search has appeared. What is a semantic search, can it be used for SEO, and what semantic search examples exist? Let’s answer these questions.

What is Semantic Search?

Semantic search is an algorithm based on the search queries of internet users. Does Google use semantic search? The answer is definitely. Let us verify this with an example.

If the search system in my browser would focus only on keywords, then for the query Washington DC, I would only get the selection of links on Wikipedia, the official touristic site of the city, etc. Taking into account that with this query, people are often interested in its area, local time, and population, I can also see this:

Therefore, the Google algorithm takes into account the semantic context  (in other words, the information that users Google most of all with the main keyword). Based on it, the selection of sites is determined for the search result.

Moreover, you might have noticed that when you type in your request, suggestions appear in the search bar automatically. This is an aspect of Google semantic search.

The Appearance of Semantic Search: Background

The semantic search was discussed for the first time in 2013 when Google launched the Hummingbird algorithm that was, in those times,  revolutionary. Its essence was based not on sorting out data but on their indexation.

In comparison to the other search algorithms, which were focused on every particular word of the search request, Hummingbird analyzed the context of all the words. This was necessary to place the pages that correspond to users’ search requests over the pages that correspond to just a couple of words.

Hummingbird’s introduction was intended to make the interaction between a user and the search system more human-oriented, by having an understanding of the connection between keywords. Also, the algorithm placed great emphasis on the content of a page, ensuring the relevance of the search results.

Then, RankBrain was introduced, it was essentially an expansion of the methods used by Hummingbird. Google mentioned RankBrain for the first time in fall 2015. This algorithm was already based on computer-assisted learning.

When a user typed a word or phrase in the search system, the algorithm put forward a suggestion on which words or phrases could have an analogic meaning or similar writing. Meanwhile, the results were filtered accordingly. The accuracy of the search results reached new heights.

Back in those days, Google stated that RankBrain had become one of the three most significant tools in search ranking. This was partially explained by the fact that RankBrain began taking into account articles, conjunctions, prepositions, and other parts of speech that act as linking sentence members between meaningful words.

Also, in some interviews with Google employees, it is mentioned that to process requests, this algorithm uses TPU ASIC – a tensor processing unit utilized for mutual work at the library of computer-assisted learning TensorFlow. Compared to the traditional graphic processors, it is distinguished by increased efficiency.

And finally, in 2019, the algorithm BERT (Bidirectional Encoder Representations from Transformers) was released. It’s based on the methods of natural language processing (NLP). This algorithm was initially thought of in 2018 and it began functioning only a year later. This is also a Google brand development.


Compared to the previous approaches in speech recognition, BERT applies the models of neural networks, such as word2vec and GloVe.

word2vec is an algorithm for integrating words. In essence, this is a double-layer neural network that learns to reconstruct the speech context of words.

In its turn, GloVe is an algorithm of computer-assisted learning without a teacher. It is used to get the vector representation of phrases and, once again, helps find suitable words from the context of the search request.

Currently, BERT can recognize more than 70 languages in Google Search.

Examples of How Semantic Search Works

Here are a couple more examples that can help you understand the definition of semantic search:

If you search the word “weather” in the search bar with an indicated location, you will get a predictable result as shown in the screenshot above. However, if you request something like “weather today”, you will get the weather forecast for the location where you currently are:


Here is one more semantic search example, to make it completely clear. Today, if you type in the search bar the phrase “face mask”, this is what you are going to see in the Google search results:

As we can see, the search results are related to coronavirus and the purchasing of medical masks for the face. Half a year ago, before the pandemic, the SERP offered much more diverse examples of masks, including virtual ones for Instagram and Snapchat.

The Principle of Semantic Search

Let’s talk about how the mathematical model of semantic search works.

It is based on the so-called Google Knowledge Graph. This is the software method of modeling the fields of knowledge that employ the knowledge of specialists in the niche, methods of linking data, as well as the algorithms of computer-assisted learning. Google Knowledge Graph was first introduced in 2012 as the tool that determines the most relevant results for the search results.

The Google Knowledge Graph is built over the existing databases so that it links together all the data within the network combining them as structured, as well as non-structured, data.

Below, we are going to discuss its importance in the semantic search.

The Role of a Google Knowledge Graph in semantic search


A Google Knowledge Graph uses the relations between words and meanings analyzing the context of the user’s request.

The key elements for building this graph are:

  • Classes. Usually, the description of entities contains their classification to the classes’ hierarchy. For instance, for the human resources department, the classes of entities can be an Employee and a Job Position. The concept of class is borrowed from the object-oriented approach to programming where every object belongs to only one class.
  • Types of relationship. Relationships between entities usually define the way they interact with each other. Thus, if you have got a class City, the types of relations that are presented in it can be friendly, relative, colleague, neighbor, etc.
  • Categories. Every entity can be also related to categories. For instance, a book can simultaneously belong to the category of Books about Australia, as well as to the category Best Seller of 2020.
  • Text descriptions. In the same way, every entity may have an ascription that makes the results precise.
  • Ontologies. They define the end-user that researches. This can be a human or an application processing whopping data streams. This ensures the general understanding of the search target and improves the context.

Explanation of the Google Knowledge Graph Structure in Simple Terms

To make this clearer, let us study a frequent example of such a graph (of course, much more compact than Google Knowledge Graph). 

Let’s imagine that we have an accounting system of the city residents for which it is necessary to arrange a semantic search.

In such a case, the highest class in the hierarchy will be the abstract class Resident. It might have such standard attributes like full name, gender, age, place of residence, parental and marital status (in general, all the information contained in the ID).

More specific classes are the Residents of employable age (in this case, additional attributes will include the kind of activity, job place, experience, etc.) or the Resident working as a private entrepreneur (in this case, it will be necessary to specify the license number, name of the enterprise, address, turnover, etc.).

Let’s continue. Certain relations can be built amongst the city residents. As mentioned above, they can be relatives, colleagues, neighbors, friends, lovers, etc.

Meanwhile, some groups of citizens can be united into separate categories, for example, they can study at the same school, work at the same organization, or be listed in the same hospital. The same resident can belong to several categories.

Every resident can have a specific ascription (text description) that won’t be typical of the majority of the remaining residents. For example, the Major of the city or the Chief Architect.

And, finally, ontologies. They define the format that the search results will be provided in. For instance, if in the aforementioned example, if a person typed the search request, Smith, they would just want to see the alphabetic list of people with this family name. If the same information is requested by an application, the same data will be required in the sequence containing zeros and ones.

When the information is presented with a detailed and properly structured graph, the contextual search with it will be easy for the search system and it will bring precise results.

The Influence of Hummingbird, RankBrain, and BERT Algorithms on the Ranking of Pages

Let’s talk about the influence of the semantic search algorithms Hummingbird, RankBrain, and BERT on the ranking of pages in SERP.

Indeed, search optimization has dramatically changed under the influence of the oldest algorithms of the mentioned ones, i.e., Hummingbird. Due to it, web developers and copywriters can finally use natural language when creating content. This means that the necessity of the compulsory use of keywords (for instance, kids’ hats buy Denver) vanished.

After presenting the RankBrain algorithm in 2015 based on computer-assisted learning, the accuracy of understanding the users’ requests has increased considerably. This algorithm has made it possible for content creators to write human-oriented texts while taking into account requests with prepositions, articles, conjunctions, and other parts of speech.

With the appearance of the BERT algorithm, Google learned to recognize specific phrases that are typical in conversational speech. In terms of SEO, it means that even the pages from forums (where the necessary exact keywords are rarely used) have begun to be ranked.

Taking into account the aforementioned, it is possible to conclude that the old-time SEO, typical of the 2000s, is no longer relevant. And the response to the question “Is the Semantic Web dead?” is negative. That is why, if you intend to get to the top of the SERP; create useful, unique, and hot content.

Semantic Search – Effective Alternative to Search by Keywords

Open semantic search has become a worthy substitute for the keywords’ search. Now, Internet users do not have to solve puzzles to get the necessary information from the stream of template keywords, while copywriters do not have to try every trick to work key phrases into almost every sentence of the text.

Let’s discuss the pros and cons of semantic search.

Benefits of semantic search

What are the main advantages of semantic search?

  • It is human-oriented. As we have already mentioned, with the appearance of semantic search systems, more and more readable, but, at the same time, commercial, texts have been appearing on the Internet.
  • SERP is highly accurate. Due to its taking into account the context and the previous search history, the accuracy of SERP results has increased exponentially.
  • Suggestions. Even if you are not sure about the correctness of your request, Google will suggest what you are looking for and it is likely it will be correct.

Problems of semantic search

The contemporary algorithms of Google semantic search have backlogs as well. Despite the use of NLP methods, Google search is still sensitive to some aspects of conversational human speech. This is especially noticeable in the languages with loose word order (for example, in Russian).

The Use of Semantic Search for SEO

Google mainly intended to provide its users with accurate and authoritative search results. When choosing web pages, Google search robots take into account the availability of backlinks from authoritative resources, as well as on the type of content (usually, Google hides adult sites only unless the request is too obvious), and the relevance of the text and media content to the typed key request.

This is why it is necessary to invest in the creation of quality, readable, semantic content that is highly unique and has maximum relevance (for instance, the use of actual statistic data).

6 Recommendations for the Promotion of Site with Semantic Search

We are providing six simple but effective recommendations for SEO that are oriented to work with a semantic web search.

Fewer keywords, more sense, and benefits

It is better to not spam texts with keywords, it is enough to focus on how helpful the content is. A couple of exact matches in an article will be enough for it to be properly ranked.

Be more focused on what your target audience is interested in

Think about the interests of your target audience. For instance, if you sell car tires, there are more chances to get a high volume of traffic writing an article about the comparison of tires by several popular manufacturers (instead of once again writing about the importance of timely replacement of the winter tires). To get a better result, read the article on How to Write a Blog Article.

Focus on the quality of content, not the quantity

In the past, SEO specialists recommended adding new blog articles almost every day. Now, it makes sense to focus on the quality and creating extensive reviews with media content (for example, with video).

Guess the intentions of users

Think about the search topics that can be related to yours. By doing this, you will probably be able to increase the CTR of your site’s pages directly from the SERP. For example, those looking for information on the benefits of osetra caviar will be likely interested in the benefits of cod liver oil. Due to this, the chances to get to the block PAA – People Also Ask increase. Read more in the article Google’s ‘People Also Ask’ Boxes.

Use the markup of structured data

Structured data is the standard format for a Google search than ensures the automatic classification of data on the page for a specific category of content. For example, if on your site, you post recipes, you will need to separately mark the list of ingredients, indicate cooking time, define the calorific value of dishes, etc.

Choose long-tail keywords

Resist using short keywords in favor of long ones, for instance, those with an indication of your target audience location. This will make the content relevant for TA and it will also contribute to its promotion in the local search. You can find out more about this topic in the article Local SEO Checklist.

The introduction to the semantic search technologies has a significant impact on the content that sites offer to their users. Thanks to these technologies, you almost might not bother with keywords and focus on the quality and benefits of content. How do you use semantic search to promote sites? Share with us in the comments.

About author
SeoQuake content manager