How to write for the web in the age of Google BERT?

Nicolas Nguyen, SEO expert and co-founder of Semji, presented in his latest webinar the Google BERT update and its impacts on writing web content. We’ve chosen to transcribe it so you can discover how to write for SEO in the age of Google BERT.

What is BERT?

On December 9, Google announced the launch of BERT in French, a few weeks after the English announcement. What lies behind the name BERT? The acronym “Bidirectional Encoder Representations from Transformers.” It’s the name of a language model in the Natural Language Processing community.

If we had to remember a few key points about BERT:

The first is that it’s a language model, so an NLP model (see definition below) created by Google itself. This model is Open Source, used since 2018, i.e. 1 year before it became an update. Google announced this update as the biggest update in 5 years, notably since Rank Brain.

Google also announced that BERT would impact 10% of queries. Having now dug into these impacts, they affect long-tail keywords more.

The point to remember: Google BERT is both a language model and a Google update.

From a purely language-model standpoint, BERT made headlines within the community of Natural Language Processing experts. Why? Because it achieved better scores than all other known algorithms.

Today, BERT will:

• better understand written content
• know how to answer questions
• be better at sentiment analysis
• also manage to detect whether two sentences have the same or a similar meaning.
• be able to assess the logical continuation of a sentence. (For example: “I say a first sentence, I write a second one, but is the second sentence really logical with the first.”)
• generate content

What’s important to take into account today is that BERT is part of a long line of language-model evolution. Indeed, in 2013, Google had already created WORD2VEC. In recent years, we can see that there have been many developments and a priori this won’t stop.

Why all these developments around language models? Large companies like Google, Microsoft, Facebook, Apple, etc., are racing for algorithms and language models with a single goal: to understand content like humans.

The other part that will concern us much more: the Google update part. What’s important is that BERT is one more update among the long series of Google updates around content, always with the aim of highlighting content that is relevant to an internet user’s query: that remains Google’s mantra and its number 1 value proposition. It’s very much in our interest to improve this understanding of content in order to highlight relevant content.

During its communication, Google showed us several examples that are quite interesting. In this example, users typed a query like: “brazilian traveler 2019 to usa need a visa” and we can see the before / after. Before, Google returned content, not for Brazilian travelers but rather for Americans, which wasn’t relevant. Whereas today, Google will highlight content from the US embassy in Brazil.

Another example with the keyword: “Can you get medicine for someone else at the pharmacy?”. Before, you could find content for filling the prescription, and today we have content that specifically answers the question. We can see that in these 2 examples, we’re dealing with long-tail keywords and notably a question-type keyword.

An interesting point to note: when Google announced the launch of BERT, they repeated a figure we’ve known for several years now: Google records billions of queries every day, and 15% of these queries are keywords that have never been recorded in the database. That’s why we need algorithms that make it possible to highlight more relevant content even when we don’t even know the query.

Another point, Google tells us that thanks to BERT, we manage to have more relevant content for diverse and varied long-tail queries. Google hopes that users will gradually get used to typing more and more long-tail (this can be typed on the keyboard but also searched by voice). Moreover, from an SEO standpoint, we tend to think of main keywords, generic queries; however, this is an important subject to challenge. Indeed, today, according to this Ahrefs study,

92% of keywords are long-tail keywords and not first-level generic keywords.

Overall, what impact can we imagine on the user side? The long tail, thanks to these updates and the BERT update, will become more relevant: so the user will search for more and more long-tail queries.

For advertisers who practice SEO, what does this mean? From a content-strategy standpoint and an SEO standpoint, if we anticipate, we can imagine 2 important challenges: the first is that it’s very much in our interest to quantify our audience’s queries and questions as much as possible, since we have the chance to have access to keywords and search volumes that allow us to better know our market and our audience. The other point, when doing Content Marketing and SEO, is that you have to be present as a brand at every stage of the customer journey. Thanks to identifying our audience’s queries and questions, it’s all the more relevant to be visible and to provide them with an answer, which is possible thanks to a content strategy.

When you write content, there’s a point that’s very important: the traffic curve, which is one of the key KPIs at creation.

When content is published, the traffic curve to avoid is the one where there’s a spike at the beginning tied to social media and newsletters, and then afterward there’s nothing left, so your content is read fairly little because there’s no SEO traffic.

According to an Ahrefs study, this is a case that happens very often, 9 times out of 10 to be exact. Semji was developed to reverse this trend. Thanks to our technology, 80% of the content optimized and created in the platform captures SEO traffic.

How to write content optimized for SEO?

The method is broken down into 9 steps where we’ll have to:

• Choose the right keyword
• Define the angle
• Define the title
• Define the content length
• Have an effective outline
• Write compelling content.
• Proofread and enhance the content
• Link other content to the content we’re creating
• Regularly update the content

Step one: choose a keyword with potential.

When setting up your editorial calendar, it’s important to always define, for each subject, a keyword and also base it on the corresponding search volume. This search volume will allow you to validate and quantify your audience in your market that’s interested in the subject. Thanks to certain tools, such as Ahrefs, you can do a more in-depth analysis of the various content already ranking on the keyword, as well as the traffic it captures each month.

Step two: define the right angle.

This means that we’re going to, for each keyword, always look at the top 3 and the subject of the content that’s already ranking.

Example on the query around coconut oil: we can see that Google will favor content that helps me understand the uses around coconut oil and its benefits. Moreover, we’ll notice that the third result is a mix of informational content and transactional content, since it’s an advice sheet with the offer to buy products.

Once the main angle of the content is defined, we can move on to the next step.

Step three: find a catchy title.

What is a catchy title? If I type “coconut oil” into Google, I’ll see content like: “12 unsuspected magic uses,” “Coconut oil: how to make good use of it?,” “Why is coconut oil a miracle product?”… These are positive examples of hooks. So think about polishing the title. Also think about polishing the meta description.

Which titles will generate the most interest and engagement? A study by SEMRush shows us that “list”-type titles are the ones that “perform” best. For example: “10 tips for using coconut oil well.” Next, we can find “guide”-type content, such as: “Everything you need to know about coconut oil,” but also “how”-type titles, such as: “How to use coconut oil well?”.

Step four: define the content length

To define its length, here’s our method: I type the keyword, I look at the length of the first three pieces of content, and I write more.

For example, if the second Google search result is 1250 words, I’ll create content that’s at least 1300 words. It’s a quantitative shortcut, to say that I’ll create content that covers the topic more in depth than my competitors. Of course, this isn’t length just for the sake of length.

It’s a criterion that can be striking. We often have this thought: “Internet users want to get the answer to their questions very quickly.” Which is true. On the other hand, when internet users look up coconut oil, they’ll look for several pieces of information, not just one. They have several search intents. Choosing between a scenario where they’ll search for several keywords, where they’ll make several clicks, and a scenario where they’ll click on one result and scroll, well, they’ll prefer to scroll.

For example, for the same level of information, between “I scroll 4 screens” and “I access 4 screens via clicks,” it’s 4 times faster thanks to scrolling.

Long content works better at every level, not just from an SEO standpoint. According to a SEMRush study, 3,000-word content will generate 3 times more multichannel traffic than 1,000-word content. This is the first point covered.

Second point, it will acquire 3.5 times more links, and 4 times more shares on social media.

We can therefore observe that long content will help from an SEO standpoint, but will also help in general to improve engagement.

“Moreover, a little tip that’s essential in my opinion; when you write long content, always remember to set up a table of contents. We see it today in the results that tend to rank better; very often, tables of contents are present.”

Why set up a table of contents? First of all, it will allow your audience to very quickly visualize what the different ideas covered are, and they’ll be able to click on them to go directly to the various subjects that interest them. From a purely SEO standpoint, the table of contents will trigger sitelinks in the SERP, so it’s a CTR gain.

Step five: define the content outline.

It’s important to keep in mind that, when creating content, the goal is to make it better than competing content, which is why the outline is important.

We can use the following metaphor: if competitors build apartment buildings, my goal is to build a 60-story skyscraper to surpass, in quality, competing content and for my content to become the reference in order to aim for the first position in Google.

To build a 60-story skyscraper, it’s in our interest to make a plan beforehand. It’s the same for content!

The other important point is that behind a keyword there are often several intents. The keyword “coconut oil” is searched 50,000 times a month. Behind these 50,000 searches, there are users asking one or more different questions, such as: “why use coconut oil?,” “how to use it well?,” or “is it good for your health?,” … Users will ask one or more questions that we’ll need to answer in our content.

The first technique is that of the 5Ws. The 5Ws are the interrogative adverbs in English: “What,” “Why,” “Who” … We can add “How?,” which is the most important. Each time, starting from our blank page, we’ll ask ourselves these 6 questions: the why, the who, the where, the when, the how, etc. This will allow us to identify the ideas to cover when creating our outline.

For example: “what is coconut oil?,” “why use it?,” “where can you find coconut oil?,” … It’s a first approach that has the merit of being simple and effective.

The other approach, which is very important, is to quantify internet users’ expectations. We have more and more keywords, we have more and more queries from internet users, recently we have questions from internet users, and now there’s BERT that will make the long tail explode. It’s in our interest to analyze this data to understand our audience, our personas, understand their barriers, and thus produce content that will stand out and better address their needs.

This goes through keyword analysis, since there’s a large quantity of them.

For example, on “coconut oil,” there are more than 13,000 keywords; it’s a step that quickly becomes time-consuming. However, it’s this step that allows us to identify that certain themes are important for our content. We can see: “coconut oil hair” is searched 10,000 times, so it’s in my interest to cover this theme in my content: “Organic,” “face,” “food,” “skin,” “benefit,” “virgin,” … All of these are indicators that can help me identify themes to address in my content, because it’s interesting for my audience. This is a key point. The second key point is that voice search and question-type keywords will continue to gain search volume. It’s then in our interest to write “question/answer” content, i.e. the heading is a question that we answer in our paragraph.

Around “coconut oil,” when we push the analysis, we distinguish more than 700 questions. From these 700 questions, we can identify questions that come up regularly, such as “Where to buy coconut oil?,” “How to use coconut oil?,” “How to make it yourself?,” …

Third step to make the outline of our famous skyscraper: analyzing the outline of the top 3 competitors so we can do better than them.

Thanks to identifying competing content, other ideas, relevant to our audience, can be added to our outline.

Once we have this data, it’s easier to form an outline that will allow me to create the best content on the market on the subject.

Step six: write content that will stand out

The important point around writing content is to understand how an internet user behaves when reading content.

Generally, the internet user will focus on the top-left area of their screen. This means we have to spark their interest from the very first lines, even before the fold, to make them want to scroll. So we’ll try to put a compelling title right at the start, an introduction that awakens their curiosity, and make the famous table of contents visible very quickly.

When writing, the technique to adopt is that of the inverted pyramid. This consists of putting the most important information at the top and the information that’s more of the detail kind at the bottom.

You have to work well on your hook and summarize the very essence of your content in it, so that the internet user understands the subject and thus scrolls to go see the content in more detail.

The other fundamental point is to think mobile. The screen is even smaller, so it’s in our interest to polish the hook to make the mobile user want to scroll down.

“ Moreover, between thinking desktop and thinking mobile, I recommend thinking mobile because today 52% of internet traffic comes from mobile. It’s not for nothing that Google switched to mobile first.”

The impact of “thinking mobile” has several levels. It has a level of content readability. The font must be large enough to be legible, but as a writer, remember to write short sentences. The more short paragraphs there are to lighten up the content, the more easily readers will be able to navigate this content.

Another tip: “I get asked the question: “should I write for Google or should I write for the audience?” 5/10 years ago I would have told you for Google, whereas today, we write for the audience precisely because Google has made many advances around language understanding. BERT is just one more step. To take a concrete example: Google understands very well that the phrases “stop a car insurance” and “cancel an auto insurance” are exactly the same thing. So, no more “keyword stuffing,” no more “keyword density.” The goal is to create content first and foremost for the audience that answers the audience ”.

“ A few little tips to conclude on this writing part: To improve the readability of content, always remember to put text in bold. Also think about bulleted lists that let the reader have a synthesis anchor point. Bulleted lists also help from an SEO standpoint, since certain queries, in the answer engine, favor them. Finally, when you write content, think about pointing and making external links to reference content such as studies, institutional content, to highlight these references in order to support your points. This particularly concerns the YMYL (Your Money, Your Life) sectors ”.

Step seven: proofread and enhance.

We know we have to do it but we don’t necessarily do it. Since our goal is always to stand out from competitors, it’s imperative to go through this step.

In your content process, as soon as you have a second person who rereads and enhances your content, you’ll get a head start. You’ll have new ideas for your content, the phrasing will be improved, the content structure too, typos will be corrected. It’s a plus for your audience, and therefore a plus for SEO.

Step eight: establish internal linking.

Once our content is ready, we’re going to send it power by doing internal linking. A link from written content prevails over a link in the header, the menu, or the footer.

Today in all the content success stories, there’s always internal linking. Internal linking is the fact of having several URLs pointing to your content, which allows you to show Google that your page and your content are important. It also transfers PageRank to it, which means this page is more likely to generate SEO traffic.

Step nine: update the content.

It’s preferable to maintain content because internet users are more interested in up-to-date content. Google has understood this well, which is why its algorithms favor recent content. Moreover, by optimizing your existing content, you’ll have enormous SEO traffic gains as a result. To show you an example of content where the client added only 40% more content writing, the traffic tripled.

To conclude, the important point, if we step back over all these recent years, is that artificial intelligence has become a “buzzword” but also a reality. If we look at the analyses around AI, we realize that it’s a booming market. Necessarily, AI and natural language processing are intimately linked. This means that since BERT was launched, as a language model at the end of 2018, it has already seen many advances thanks to derived algorithms.

Moreover, the GAFA (Google, Amazon, Facebook, Apple) are investing enormously on this subject. It’s not just Google. It’s the arms race of language models. BERT is just one more step in the future Google updates around language understanding. BERT still has limits. We can then largely anticipate, without taking any risk, that there will be other updates around content algorithms.

If we want to write content thinking long-term and thinking SEO performance, we’ll remember these key points:

The first: it’s that we’re lucky to have data to understand the audience. We must use this data, these keywords, these search volumes, in order to better know our audience and its needs and then produce “user centric” content, i.e. content centered on the user.

Last point, which is confirmed over time, is to always create quality content. That’s what Google says. It means creating content better than that of the competitors who are already ranking on the keywords we’re targeting.

Here are the questions asked by participants during this webinar:

“ Regarding the content update, apart from adding content, should you also refresh the publication date of the content or keep the initial go-live date?”

Nicolas Nguyen’s answer : “That’s an important point! In general, you’ll have two types of content: dated content, and undated content. If your content is dated—and moreover it’s a preference of internet users, but the content still has to be updated—the recommendation is to highlight the update date of your content. What you can do: either you have a single date and you put the last update date, or you have two dates: the initial publication one and the update one on such-and-such date. Both are good practices. But, in any case, if you update content, highlight the update date as soon as it’s dated content because internet users prefer up-to-date content.”

“Isn’t long content penalizing on mobile?”

Nicolas Nguyen’s answer: “Penalizing, I imagine from a UX standpoint. Maybe the question was asked before I covered the subject on the different practices. Overall, as we were saying, a user prefers to scroll rather than click, like on Facebook, Twitter, … Today, from a UX standpoint, it’s internet users’ preferred practice. On the other hand, to facilitate navigation, remember to set up a table of contents above your long content to facilitate both understanding of the ideas covered and your internet user’s navigation. So in my opinion, they’re not penalizing at all, quite the contrary.”

“Isn’t there a limit to the system of the number of words to write to beat competitors?”

Nicolas Nguyen’s answer: “Yes, of course there’s a limit. The limit is that if tomorrow you write 10,000 words in Lorem Ipsum, it won’t work. The number of words is just one criterion among many. It’s a quantitative criterion but afterward there are necessarily all the qualitative aspects. What’s important is to apply the whole set of criteria of the method I’m proposing. If you only do the content-length part, it won’t work. The shortcut is to say: when I create content, I ask myself the question of “what are all my audience’s questions on the subject so I can answer them as much as possible?” That’s the idea.”

“To rank and get traffic, do you advise publishing content that precisely answers a given query or publishing a more generic and more complete document?”

Nicolas Nguyen’s answer: “So the subject is, for example: I have a long-tail keyword, like a question. The question is “do I make a pillar content where the answer to this question is 5% of my pillar content, i.e. I have my coconut oil content and then the answer to the question occupies 5% of the space, or do I make a specific page to answer this question?” Ideally, you have to choose both, because here we’re getting into subjects more oriented toward the semantic cocoon, so I invite you to go see our webinar which is accessible. The subject is: I make a pillar page that gives an overall vision to the person who types “coconut oil,” and then if the person who types “coconut oil” wants to know more about “how to use coconut oil for my hair?,” they have a short answer in the pillar content but if they want to go into more detail, they have an answer that points them to a child page that’s more specific.

So pillar page, child page, on one side the first-level generic keywords and on the other side the long tail. That’s the best approach in my opinion. The other element of the answer will be that it necessarily depends on the competition opposite, since on long-tail keywords, we often come across specific pages.”

“How often is it recommended to publish new content on a blog?”

Nicolas Nguyen’s answer : “In content marketing, frequency is an important subject, rhythm is important, potentially the day is important. It’s a bit like radio shows where you have a radio show title on such-and-such days at such-and-such time; it creates an appointment effect for your audience. So, often in content marketing, we repeat it, rhythm is a criterion. For example, the fact of sending a newsletter every month, promoting content on social media, all of this will pace your communication with your audience. I want to tell you that it’s in your interest to have a constant rhythm, generally on a monthly basis. From a purely SEO standpoint, what I see with those who perform best in content, they set themselves a monthly content goal.

Of course, it’s not just quantity, there’s also quality. It’s better to make less content but of very good quality that brings a lot of SEO traffic, than lots of content that makes 0 SEO traffic.”

“Do you advise deleting obsolete content and publishing a new one, or updating the original content?”

Nicolas Nguyen’s answer:“To answer this, I’m going to rather paraphrase recommendations made directly by Google. I take the example of all these pages that live off seasonal queries, typically Black Friday, which happens every year. The subject is “do I create a new page every year?” The good practice recommended by Google is to keep the same URL, and update the content so that it’s up to date every year for your audience. So above all do not create a Black Friday page every year, otherwise it’s a catastrophe, and if possible, let’s avoid changing the URL. If we’re forced to change the URL, we put a 301 redirect. Why? Because what justifies it and is fairly logical is that when you create a new URL, Google takes several months to transfer all the authority to this new URL. Google’s inertia is long. The second point is, if you create several pages, on the same subject, with obsolete content, you dilute the SEO power toward lots of different content. If the content is obsolete, let’s try to update it, and keep the same URL. On the other hand, we can modify the angle so that it’s more suitable.”

“You talk a lot about Ahrefs, what do you think of SEMRush? Are they tools complementary to Semji?”

Nicolas Nguyen’s answer: “If I have to choose between Ahrefs, SEMRush, Semji,

necessarily, I recommend you take Semji. However, beyond that, Ahrefs and SEMRush are tools that, for me, can be complementary. It’s like any tool, each has advantages and disadvantages. For me, Ahrefs has one key advantage, it’s the quality of the data, which is, in my opinion, more advanced than SEMRush. It’s also the UX, the navigation speed of the tool, so the productivity I’ll have with the tool. SEMRush will have other advantages, but sometimes I’ll find the data a bit less relevant. So it all depends on your uses. The king’s choice is to have all three.”

“How do you put a link to another site, to cite a source for example? Should you put a nofollow link to keep Google on our site, or should you put a dofollow link to show that we have qualified sources? ”

Nicolas Nguyen’s answer : “Nofollow, originally, why was it invented? If I remember my history lessons correctly … It was invented because of spam linking. Notably from the era when, with a button, you could click and publish 10,000 comments on 10,000 WordPress sites to get 10,000 links. There Google said “whoa,” this isn’t great for webmasters. So, the nofollow attribute is an attribute to allow webmasters to say that this link doesn’t transfer juice. So when you cite a source in your content, you don’t need to bother with nofollow, because you won’t have a spammed link, and one of the bases of Google’s algorithm still remains links, so you might as well cite sources with dofollow, if only to congratulate our dear sources.”

“What do you think of Yoast for evaluating content on WordPress?”

Nicolas Nguyen’s answer : “Excellent question. 3 years ago, I would have told you it’s not bad because it’s directly in WordPress; now I’ll tell you that the Semji platform is 100 times better. Why? Because Yoast, even premium, unless it’s changed since, doesn’t analyze the SERP, doesn’t analyze competitors, doesn’t analyze keywords and search volumes, which means you don’t have the right recommendations.”

“What tools let you know the volume of a keyword?”

Nicolas Nguyen’s answer: “So there I had listed several tools in one of the slides. You have Semji, Ahrefs, SEMRush, for the big tools, let’s say. You also have Ubersuggest

, which has a free approach, and Keywords Everywhere.”

“I didn’t hear about inserting videos or infographics into content; do you think these media can help rank our content, notably from a BERT standpoint?”

Nicolas Nguyen’s answer: “That’s an excellent question, which indeed I didn’t answer. I’m completely convinced that inserting images, videos, infographics helps you make content that’s more relevant for your audience, and more indirectly, helps you get more SEO traffic. I’ll give you an example that’s very simple and very concrete:

Today, who sends links to your content? It’s influencers. Influencers take as data everything that’s a study, statistic and infographics. So if you make a statistical study with an infographic, you have all the influencers taking it up and making links to your website. That alone favors SEO. Video has other advantages for your audience; you have people who, depending on their uses or the context at the moment they consume the content, will prefer to read the content while others will prefer to watch the video, so if we offer them both, so much the better. The advantage of video too is that as soon as you optimize it from an SEO standpoint, it can itself send traffic to platforms such as YouTube, and thus increase your notoriety. Everything that’s videos, infographics and so on are formats that are important. They’re formats I was going to say of the future, but that are already the present.

This webinar was more on the writing part, but in your content, add these new forms of media.”

Closing glossary:

BERT (“Bidirectional Encoder Representations from Transformers”) is the name of a natural language processing model developed by Google in order to better understand internet users’ search intent.

NLP (Natural Language Processing), or TAL (Traitement Automatique du langage) in French, represents “the set of techniques that allow a machine interface to automatically analyze and process an individual’s written and spoken words and to express the necessary responses.” ( Source)

SERP (Search Engine Result Page) refers to the search engines’ results page.

RankBrain: is a Google algorithm aimed at better understanding internet users’ search intent.

PageRank: “Google’s search engine developed the PageRank algorithm (registered trademark) to rate the popularity of each Internet page based on the links it receives and the popularity of the sites that point to it. Each page and each website indexed by the search engine is indexed, among other criteria, according to their PageRank score. Popularity is even one of the most important criteria in organic search.” ( Source)

User centric is an English term that means placing the user at the center

CTR (click through rate) is the click-through rate. It’s the proportion of internet users who see an advertiser’s ad and click on it. CTR is a very important performance indicator (KPI) for advertisers seeking performance. This is particularly true in SEA, where the best-performing ads in terms of clicks are favored by Google Adwords algorithms. ( Source)

AI: According to one of its designers, Marvin Lee Minsky, AI (Artificial Intelligence) is defined as “the construction of computer programs that engage in tasks that are, for now, accomplished more satisfactorily by human beings because they require high-level mental processes such as: perceptual learning, memory organization and critical reasoning”

UX (User Experience), or expérience utilisateur in French, refers to the experience lived by the internet user on a web interface.

YMYL (Your Money Your Life) is a Google term to characterize pages that directly impact internet users’ lives, such as health or bank pages.

About Semji:

Semji is a SaaS platform that maximizes the ROI of companies’ SEO and Content Marketing content. The solution allows marketing and SEO teams to exceed their goals and gain performance on the key aspects of their job: the search for visibility, the quest for ROI and productivity gains.

Want to know more about Semji? Request your personalized demo now!