How BERT will affect content in the future

Dec 10th, 2019

Search is about understanding language…

Google must constantly work to improve its technology and make their search results as relevant for users as possible. To do this, they need to ensure that their systems can make sense of language in the most complex way, just like humans do. While you might understand that ‘surfing’ can refer to a sport or an action that you do on the web, a machine could struggle to see the difference. This means that Google must figure out which version is being referred to by the searcher. 

Introducing BERT. BERT was announced at the end of October and was created so that Google can better understand the intent of a query. It’s their job to fathom what the user is asking and provide the best articles on the web that can give them an accurate answer, “no matter how you spell or combine the words in your query”. In their article, Google admitted that they still don’t get this right and it’s one of the reasons they’ve worked so hard to improve the language understanding capabilities. 

BERT stands for Bidirectional Encoder Representations from Transformers. In simple terms, this refers to the way in which the system interprets sentences. Unlike Google’s previous systems that understand the words by looking at each one individually, BERT can look at the whole sentence and then try to interpret the context of a word. For example:

Input: The man went to the [MASK]1. He bought a [MASK]2 of milk. 

Labels: [MASK]1 = store; {MASK]2 = gallon

This is how Google trained BERT to understand more about the English language. Google have stated that BERT will help Search better understand one in 10 searches, making it the biggest update since RankBrain was released in 2015. 

BERT and the impact on content

Danny Sullivan, Google’s public search liaison, explicitly stated that you cannot optimise your site for BERT. Rather than a change to the ranking factors, it’s a change to how the algorithm views the intent of a search query and allows Google to better understand what’s being asked. If you’ve previously written content that’s made for algorithms, you may have seen a drop in visibility since BERT. However, sites that have written content directly to help the user will likely see an improvement. 

So if you can’t ‘optimise’ for BERT, how could it have an impact on your content in the future?

You should continue to create good quality content, not necessarily because a search engine wants you to but because your users will find it beneficial. There’s less pressure to write for an algorithm. Instead, focus all of your efforts in writing excellent content for real people with legitimate problems or queries. If you’re writing for authentic users, you’re already ‘optimising’ for BERT and therefore future-proofing your content. 

Think about whether your content is actually helpful or whether you’ve stuffed keywords onto a page in order to make it rank. Google is getting much better at understanding what kind of content should be on a page in relation to a query. For example, if someone searches ‘how to tie a tie’, there isn’t a single product page ranking within the first 10 pages of the SERPs. It’s because the search is an informational query, whereas your product page is transactional, not informational. In this example, Google accepts that searchers don’t need to buy a tie – it’s probable that they already have one. Instead, create a blog post that demonstrates how to tie a tie and add a few internal links to your tie product pages. It is much more likely to rank this way.

Could BERT affect voice search?

If Google has purposefully made their machines better able to understand language, it could be that they’re doing so to tailor for the increase in voice searches. Voice search has increased year-on-year as more people utilise voice assistants, such as Google Home and Alexa, to find out information. 

Before BERT, Google might not have been able to understand spoken searches as much as it could understand typed phrases. This is because of language complexities that Google just couldn’t fathom. However, when users type a phrase, they use ‘keyword-ese’. According to Google, this is when the user types a string of words in a way that they think a search engine will understand. It may not necessarily reflect how a typical human would talk. The human language can be a complex thing for a machine. Now, however, BERT can more accurately understand and process the language and is more likely to present relevant results.

According to Jon Earnshaw, owner and co-founder of PiDatametrics, voice search is becoming increasingly accurate and users are more satisfied with their voice search results than ever before. He found that voice search queries were nearly 30 per cent more likely to satisfy the intent of searchers than typed searches. Earnshaw believes that BERT is becoming more conversational so that it can directly improve the results for users searching with voice. 

Content in the future

In order to satisfy BERT, you should write good content that’s designed for people. This means providing helpful content that’s easy to read. You can do this by using: 

  • Short paragraphs and frequent headings to make the content scannable
  • Clear language
  • Visuals, such as images and videos
  • Definitions of jargon or hard-to-understand words
  • Conversational tone of voice

If you’re writing to help a user, you should explore different ways to display your information, like images, audio, videos, interactive pages, so that every user’s information preferences are catered to. 

For example, a piece titled ‘How to make a small bathroom look bigger’ that has 110 searches a month might include an interactive asset. The user could select the shape of their bathroom, where the windows and doors are and whether they want a bath, shower or both. Then, based on the selections, the asset can produce the best layout for the room. 

This kind of content might be far more valuable to a user than a 1,000 word article on the topic. Just remember that the page might still need a little bit of content, as Google cannot see or understand some images or interactive assets.

Avoid throwing lots of content onto your website because you believe it’s what Google wants to see. It’s better to have less content that’s good quality and updated regularly instead of a large quantity of pages that only have 200-300 words and aren’t ever updated. 

BERT is trying to tempt us away from writing for search engines and is convincing us to write for the user, however this doesn’t mean that you need to do away with all of your SEO strategies. Lots of ranking factors are already in place to optimise the experience for the user, such as page speed, redirects, alt text and friendly URLs that explain what the page is about. These things are favoured by Google because they improve the user experience and you still need to apply these things to give your content the best chance of ranking. 

Click’s expert team have worked with brands across a multitude of industry sectors. If you’re looking to improve your brand’s search visibility get in touch with us today to discuss how we can work together.

Facebook Twitter Instagram Linkedin Youtube