What Is Semantic Search?
Semantic search is how search engines use meaning and context to understand search queries and provide results that match intent.
It’s how search engines understand natural, human language.
And that means Google doesn’t need an exact match keyword to deliver the right results for what you’re searching for.
So, if you search for “the boy who lived,” Google can tell you’re looking for results about Harry Potter. Even though the search query doesn’t contain the words “Harry Potter”:
Similarly, Google can still serve the correct results if users misspell keywords. Like so:
It’s also the reason you see nearby restaurants when you search for “restaurants near me.”
In the article, we’ll cover how semantic search works, the evolution of Google’s algorithm, and how to optimize your content for semantic search.
First, let’s dive into its history.
History of Semantic Search
The Google algorithm has evolved to understand semantics. It started with the Knowledge Graph, which debuted in 2012.
Knowledge Graph (2012)
The Knowledge Graph is Google’s database of people, places, and things (also called entities) and how they’re connected.
Over 500 billion entities, in fact.
Google describes the relationship between entities as “things, not strings.”
In other words, when someone enters a “string” into a search bar, Google now understands that string as a “thing”. It has meaning and context. And Google knows that it shares a relationship with other “things.”
So, if a user searches for “Apple,” Google can also tell you that its CEO is Tim Cook. Or other information, like the company’s current stock price.
Google displays information from the Knowledge Graph through knowledge panels that appear on the search engine results page (SERP).
Here’s a knowledge panel example using the “Apple” search query:
Hummingbird (2013)
The Hummingbird update in 2013 prioritized natural language processing (NLP).
That’s a machine’s ability to read, understand, and derive meaning from human language.
This update happened around the time that people started making searches with their voice—so it was important to keep up and provide them with accurate results.
Together with the Knowledge Graph, Hummingbird improved on NLP, which laid the groundwork for semantic search.
Google adopted NLP to better match pages to their meanings. This allowed Google to understand more conversational searches.
RankBrain (2015)
RankBrain is a system that helps Google better understand the intent behind users’ searches.
It built upon Hummingbird by helping Google understand search queries that contained words or phrases it didn’t know.
RankBrain is a machine learning system. This means RankBrain continues to learn and analyze based on best-performing search results.
And it’s also a confirmed ranking factor.
However, there’s no surefire way to optimize for it. Your best is to provide as much information about a page as possible (which you already should be doing) and let Google do the rest.
BERT (2019)
BERT is a language processing technique implemented by Google in 2019. It stands for Bidirectional Encoder Representations from Transformers.
In simpler terms: Google doubled down on understanding conversational searches.
BERT considers the full context of a keyword, including the words that come before and after it—like so:
In the above example, Google shows pre- and post-BERT examples of how it interprets the keyword “2019 brazil traveler to usa need a visa.”
The word “to” in this keyword is particularly important, as it implies that a Brazilian traveling to the U.S.
Previously, Google’s algorithm didn’t understand the connection. After BERT, it was able to better understand what the searcher meant.
MUM (2021)
The Multitask Unified Model (MUM) is a language processing framework implemented by Google in 2021.
Google describes MUM as being 1,000 times more powerful than BERT.
MUM can understand images, video, and audio files. It’s also multilingual. So it can find information related to your search query even when that information is in a different language.
Google uses the example of a voice search (“can I use these to hike Mt. Fuji”) paired with a photo of hiking boots.
Because of MUM, Google can process both the search and the picture at the same time.
So it can let you know that these boots would indeed work well for hiking. And may even take the next step by showing results for recommended gear.
As you can see, Google has made big advancements in a short span of time. It will only continue to improve to better understand searchers and serve more accurate results.
How Does Semantic Search Work?
Google uses machine learning to figure out what people are looking for based on context and search intent.
Search intent is the reason why a user is looking something up. Are they comparing products? Trying to actually buy something? Google will serve different results depending on the intent.
But how does Google know how to do this?
Semantics.
In other words, the ability to understand the meaning of keywords and their relationship to other keywords.
Which makes search results more accurate. And more relevant to the searcher.
For example, here’s what the SERP looks like when you type in “lo mein”:
The SERP does include recipes, but it also has a Map Pack.
Meaning it can infer it’s possible the user wants to order takeout even though words like “order” or “delivery” weren’t used.
Another example would be if you searched for a broad term like “wedding anniversary.” Google is able to make assumptions about the content people may want. So it may show related pages about wedding anniversary cards and gifts.
As semantic search is influenced by context, search results can vary for the same keyword and are influenced by a variety of things, including:
- Your location
- Your search history
- Current news and events
- Trends
All of those factors provide context to semantic search. For example, say you search for the soccer player Lionel Messi.
When this article was written, the results for “Messi” were mostly about his career in general.
As soon as the World Cup starts, those results will become Messi’s latest stats, achievements, and game highlights.
How to Optimize Your Content for Semantic Search
In order to optimize content for semantic search, you need to put the user first.
Here are a few ways to do that:
- Better understand users’ search intent
- Focus on topics rather than keywords
- Use structured data to enhance search results
- Connect related content with internal links
- Use semantic HTML
In short, optimizing your content for semantic search means creating content that doesn’t just match keywords, but also reflects the way that users write and speak.
Here are five ways you can do that:
1. Understand & Optimize for Search Intent
Search intent refers to a users’ main goal when they type something into Google.
This ties into semantic search because Google is trying to close the gap between what a person types and what they actually want to know.
In order to work with the algorithm, SEOs should prioritize user intent in their strategy.
So when you create content, your goal should be to match what the user is looking for and anticipate their follow-up questions.
For example, if someone searches for “how much house can i afford,” they probably also want to know about mortgages.
Keywords can fall under four search intent categories:
- Informational (learning about a topic)
- Navigational (looking for something specific)
- Commercial (investigating products, services, or brands)
- Transactional (intending to make a purchase)
Understanding the category of intent will help you shape your content.
You can decipher keyword intent using various Semrush tools. Here’s what it looks like in the Keyword Magic Tool:
Here’s an example.
If you want to learn more about the benefits of weighted blankets, Google isn’t going to serve transactional results because it can tell that your intent is informational.
On the other hand, if you search for “best weighted blankets,” the SERP will be different.
Because of semantic search, Google knows that “best” usually means the user is trying to decide which blanket they should purchase:
This means that you need to create different types of content to satisfy different needs.
A good way to approach this is to create content pillars. Start out with one broad topic, then come up with content that addresses more specific subtopics.
Be sure to link these pieces of content together so Google can tell they are related and better understand their relationship.
This can help you create content that meets and anticipates your users’ needs—which is great for optimizing for semantic search.
2. Focus on Topics, Not Keywords
Because Google now attempts to process information like a human might, it’s important to focus on broader topics rather than specific keywords.
This is because people search more conversationally these days. And Google does its best to match these conversational queries with what it thinks is the best answer.
Standard keyword research is important, but creating quality content is much more than just including keywords a certain number of times.
If you exhaustively cover one topic, chances are that your page will rank for a variety of related long-tail keywords (i.e., more specific keywords that have lower search volume but higher click-through rate).
You can quickly find semantically related keywords with our SEO Content Template tool.
It analyzes content from your top competitors, then gives you a list of relevant keywords that appear on their pages.
Let’s target the keywords “SEO,” “what is SEO,” “SEO tips,” and “SEO for beginners.”
The tool looks at pieces that are currently ranking and suggests keywords related to the ones we want to rank for.
A few suggested keywords are “blog post,” “keyword research,” and “organic research.”
Take the time to understand how keywords fit together to form topics. The goal is to understand user intent rather than match a specific set of keywords on a page.
Keep in mind that the goal isn’t necessarily to rank for every single sub-topic (though that would be great).
It’s about finding sub-topics all your competitors cover in their articles.
For example, if all of your competitors talk about keyword research in their article on SEO for beginners, you should too.
Doing so will satisfy the user as well as help Google identify relationships between similar topics.
3. Use Structured Data
Structured data is an organized set of data that helps search engines understand your content.
Structured data markup also increases your chances of triggering rich snippets—search results that display extra information like ratings or reviews:
Here’s what structured data looks like behind the scenes:
Markup (i.e., how the code is written) is important because it helps Google understand how to categorize content. And this is how Google gets data to show rich snippets.
Just look at how much of the SERP is dominated by rich snippets:
Here’s how you can make the most of structured data or schema (i.e., the standardized way of creating structured data):
- Visit schema.org to browse a wide range of schema markup templates understood by all major search engines.
- Use Google’s Structured Data Markup Helper to assist in marking up your content.
- Merkle’s Schema Markup Generator is another great option.
- Use Google’s Rich Results Test to check that your markup is correct.
You can also use Semrush Site Audit tool to see how structured data markup has been implemented on your site.
Click on your project report within the Site Audit tool. Then, click on “View details” in the markup section:
You’ll see a report telling you if your structured data markup has any problems.
Read our guide to schema and structured data to learn more about implementing it on your site.
4. Build Links That Demonstrate Relevance
Both internal links and backlinks demonstrate topical relevance and help Google understand your content better.
It can take time and patience to secure external backlinks. Executing a solid internal link building strategy, on the other hand, takes far less effort as you have the power to make the changes on your own.
Internal links can be just as important as backlinks from other sites, especially when demonstrating the topical connection between two pages.
Let’s use the example of content pillars again. Your main pillar page should link to related sub-topics. Perhaps those sub-topics also have another layer of topics, like this:
Adding internal links to these pages will show Google that your pages are related.
Ideally, if you comprehensively cover a broad topic, your site will have the answers to users’ questions when they make a related semantic search.
5. Use Semantic HTML
Semantic HTML is made up of elements that clearly describe their meaning to search engines.
For example, you could write text in HTML that looks like a heading on your page. But unless it’s tagged with a specific heading HTML code, Google might not know it’s actually a heading.
In the example below, the non-semantic HTML uses unspecified <div> and <span> tags to create content.
But on the semantic HTML side, clearly defined tags organize the content in a way that communicates their meaning to Google:
Semantic HTML tags include <header>, <footer> or <article>.
Header tags denote a header, footer tags indicate a footer, and so on.
You can read our semantic HTML guide to learn how to start marking up your code this way.
Use Semrush to Optimize for Semantic Search
As Google continues to evolve and learn how people use search, it will become more and more important to optimize your content for search intent.
To learn more about the history of the Google algorithm, read our in-depth guide.
