Feb. 16, 2021
11 min reading time
11 min reading time
Entity optimization deserves the same amount of press coverage as other topics we SEOs delve into week after week. A guide to understanding entities.
They are by no means a secret, and the role of the entities in the SEO It has been extensively documented – entity optimization simply isn't the trending topic you might see every time you check your Twitter timeline.
We would much rather discuss less impactful concepts, such as whether content within a subfolder ranks better than in a subdomain, or whether it is important for an SEO to learn Python (am I right?).
But entity optimization deserves the same amount of press coverage as the other topics and concepts that we SEOs delve into week after week. I want to help us understand why and how we approach content with entities in mind.
Google defines an entity as "A thing or concept that is singular, unique, well-defined, and distinguishable." An entity can be an event, an idea, a book, a person, a company, a place, a brand, a domain, and much more. You might ask, "Isn't that the definition of a keyword? What's the difference?"
An entity isn't bound to language or spelling, but to a universally understood concept or thing. And the core of an entity is its relationship to other entities. Google uses an illustration of "nodes" and "edges" to explain entities, with entities being called nodes and relationships being called edges. Let's look at a search to see how this works:
A search for "Justin Trudeau" displays a knowledge board where he is listed as "Prime Minister of Canada." And a search for "Prime Minister of Canada" displays a knowledge board for Justin Trudeau. So we know that Justin Trudeau is associated with the Prime Minister of Canada, and vice versa. Trudeau is the current Prime Minister, so what happens when we search for the same entities with different relationships?
Here we see a different set of results, based on a different relationship between the nodes.
We believe that Google uses a model called Word2Vec (referenced in this patent regarding keyword extraction) to break down entities, map them to a graph, and assign them a unique ID. In a sense, Word2Vec transforms language into a mathematical calculation, enabling Google to correctly identify concepts and appropriately map them—regardless of the language—in a way that traditional models simply cannot.
We do not currently know exactly how entities fit into search results, but based on a model introduced in a patent entitled "Ranking of search results based on entity metrics", we know that one of the biggest factors is relatedness.
The relationship is primarily assessed by something called co-occurrence (the linked patent is still pending, but helpful for understanding co-occurrence). Coincidence assesses the strength of the relationships based on how frequently the entities appear together in documents on the web. The more often two entities are mentioned together, and the more authoritative the document in which they are mentioned, the stronger the relationship.
Entities are not necessarily a ranking factor – at least not in the traditional sense. And we don't know exactly how much weight they carry as quality signals. But we do know that there are two key categories of ranking factors (among many others) that are strongly influenced by the entity diagram.
Historically, keywords were a measure of the relevance and quality of content. Keywords aren't dead, but entities give search engines a better understanding of the relationships between words in a search.
Let's look at the search term "Best shoes for basketball in Atlanta," for example. Sure, we could create a post and fill it with the key phrase. But in a world of entity-based indexing, Google looks for semantics surrounding each of these entities and for signals that indicate their relationships.
You may remember the explosion of the «LSIKeywords. Regardless of whether the Google algorithm uses latent semantic indexing or not, this fascination with semantics is rooted in the entities themselves. Every search is now semantic.
In the world of SEO, it's fairly common knowledge that not all links are created equal. Entity-based indexing reinforces this notion. A post aiming for a ranking of the "best shoes for basketball in Atlanta" needs links and references from authoritative sources about shoes, basketball, and the city of Atlanta to truly rank in that SERP.
Patents on entities have been appearing for over a decade, and most people believe that entities have played a role in search algorithms for quite some time. The question is, when did entities become central to indexing?
Cindy Crum of Mobile Moxie wrote a brilliant five-part series on entities. She argues that entities became a strong ranking signal at the same time as Google's introduction of Mobile-First Indexing. In fact, she refers to the entire Entity-First Indexing update as "Entity-First Indexing."
Did BERT have anything to do with entities? Although I believe BERT received somewhat more attention than it probably deserved, its use in the Google algorithm can help us understand the meaning of entities.
BERT (Bidirectional Encoder Representations from Transformers) is a natural language processing model introduced by Google in 2018, with rollout beginning in October 2019. BERT is capable of considering the complete context of a word based on the words preceding or following named entities.
We won't delve too deeply into this, but we'll look at an example from Google that helps us understand what BERT means for search. In a recent post, Google called out the search query "2019 Brazil travelers to the US need a visa." The preposition "to" is crucial here, and even more crucial is its relationship to the entities found before and after it. Before BERT, Google would have returned results about US citizens traveling to Brazil. After BERT, Google can recognize this nuance and return a more relevant and helpful result.
Entities are central to natural language processing models like BERT.
Before we delve into some actionable tips, it's important to understand that entities have a far greater impact than content. Optimizing entities is crucial for building brands, setting up domains, and all kinds of other online endeavors. Beyond that, there's a massive impact on content.
*Quick preface: I've used this approach to rank articles and have seen success, but it's by no means foolproof or battle-tested. I'm not aware of any studies at this time that prove a direct link between such an approach and high rankings. Nevertheless, I believe in it and think that understanding entities can trip up SEOs.
To begin, we need a topic and a key phrase for which we want to create a ranking. We won't delve into how to conduct keyword or topic research, but let's stick with our example above and aim for a ranking of the "best shoes for basketball."
If we want to achieve a ranking for this key phrase, we need to gain insights into which other topics and concepts Google considers related in its entity graph. Where can we gain such insights? In just a few places:
Wikipedia: We know that entities are the foundation of Google's knowledge graphs – and we know that Wikipedia feeds a large part of its knowledge about entities. We can assume that if Google relies on Wikipedia to help them understand topics, the attributes and sources found in Wikipedia could help us orient our content.
Google Images is another goldmine for understanding entities:
Below the search bar, we find entities that Google positively associates with "best shoes for basketball." These aren't the shoes or shoe attributes you need to list in your article, but logically, mentioning these topics would help Google connect your article with them.
"People Also Ask" is another helpful resource for entity optimization. These are the other topics and questions that Google associates with your target keyword phrase:
Identify the two or three articles that best match your target keywords. Now we'll look at how Google interprets the entities found in those articles. We'll use Google's NLP API demo:
This is just a sample demo of their NLP cloud product. Nevertheless, it delivers truly valuable data. Before we dive in, we need to define a key term.
Google's API demo covers a handful of things: nuance, feel, syntax, and categories. In this article, we'll really only focus on the highlights.
Salience is an assessment of how important an entity is within the context of the entire text. The higher the score, the more prominent the entity. We will use salience as a guideline for our content. Here's what to do:
We see that the highest-ranking units are "player," "best basketball shoes," and "basketball shoes." Since Google ranks this page well for our desired keyword phrase, we can conclude that these are units for which we should try to optimize our article.
How can you optimize for these entities? When you begin writing, your goal should be to establish the relationship between the entities you're targeting in your keyphrase and give Google all the context you need to connect your target keywords to its entity graph. This isn't achieved by simply stuffing keywords, but by using some of the language and semantics we've gathered from the sources mentioned above.
Google Images and Wikipedia should help you choose semantically related keywords and the language you use in your article, while People Also Ask can help guide your general topics and headings. Again, the point isn't to cram in keywords, but to have a toolbox of individual words, phrases, languages, and topics to guide our writing in a way that prioritizes our target audience.
Once you've finished writing, run your article through Google's NLP API demo to get a feel for how it stacks up. If the desired entities show low scores, it might be worth going back to the drawing board. At the very least, you can analyze articles that show more success with the entities to gain insight into how Google associates your goals.
Since entity optimization is somewhat more complex than keyword optimization, it's more beneficial to regularly update your content whenever new topics related to your entities emerge. For example, if new basketball shoes are released and Google establishes their place in the entity diagram, adding them to your post would improve the visibility of your entities.
BERT is another great example. If you had a post about natural language processing, Google would expect it to be mentioned, given its massive online popularity.
I myself, and the industry as a whole, still have a lot to learn about entity optimization. And here, too, the implications extend far beyond content optimization.
But I believe the focus on entities has already begun, and the signals will only become more important for Google and other search engines.
© 2012-2025, MIK Group GmbH | General Terms and Conditions | Imprint | Privacy policy
