Two Things Everyone Missed about BERT

UPDATE: Since this article was released, Danny Sullivan, Google’s Search Liaison, said on Twitter that the slide used by John Mueller during the Google Webmasters Conference in Zurich was outdated, and the BERT technique is in fact used for search queries and featured snippets alike.

https://twitter.com/dannysullivan/status/1208052794785067008

On October 25th, Google officially announced that they started using BERT in Google Search. Since then, it’s been one of the hottest topics in SEO.

a Google Trends chart showing the popularity of "google bert" and "google eat" keywords in Google Search

And it’s not surprising – Google said that this change would be the most significant since RankBrain was introduced, particularly in terms of how it understands the more ambiguous and unique queries.

a quote from Google's article announcing BERT's implementation in search

What’s BERT, anyway?

If you’re like me and reading the definition of BERT makes you question your life choices, just power through to see the examples of how it works in practice!

BERT stands for Bidirectional Encoder Representations from Transformers. It’s a neural network-based technique for natural language processing pre-training. Developed by Googlers back in 2018, it can be used to power a state-of-the-art question-answering machine.

Here’s one thing we need to state clearly: BERT is not a ranking algorithm. The ONLY role of BERT is to make sure that Google understands queries better, so there’s nothing to optimize for from the SEO perspective!

All BERT does is allow Google to improve its search feature to understand the intent behind a query better. By considering the full context of a given word, Google took its search results to the next level – particularly for longer queries where one word may change the meaning of the whole query.

Within the announcement, Google gave some great examples of how BERT can improve their understanding of complex search queries:

Example #1

An example of search results for the same query before and after BERT was implemented in Google Search

Example #2

Another example of search results for the same query before and after BERT was implemented in Google Search

But what we learned last week is that some SEOs might have completely misinterpreted the impact of BERT on their websites’ rankings.

With that, let’s dive into two points that we think that SEOs and website owners alike should acknowledge regarding BERT:

1. You CAN’T optimize for BERT.

Googlers couldn’t have been more transparent about that, and the SEO community has attempted to debunk it multiple times, but optimizing content for BERT is a myth that’s still floating around on the web.

https://twitter.com/dannysullivan/status/1188689288915050498

Still, if you look it up on Google, you’ll be surprised how many SEOs jumped the gun and started giving out advice on optimizing your content for BERT.

Let’s think about it for a second. Can BERT affect some of your rankings, particularly for long-tail keywords? Yes, definitely.

Is there anything you should do about that? Absolutely not.

With an improved understanding of user intent behind some more complicated queries, Google is able to serve more accurate search results. That means that some websites may no longer rank for some keywords and lose traffic as a consequence – but this is not the traffic you want anyway.

If your page got some traffic as a result of being surfaced in an inaccurate search results page, you shouldn’t be happy about that! These visitors will almost always bounce, and even worse – they will remember being unsatisfied with your website.

Now, even if you wanted to, how could you possibly optimize your content for that? Just focus on the quality of your content, and Google will push the right traffic your way.

2. BERT only affects search queries in English!

BERT was first fully implemented for search queries in English, but in early December, Google announced that it’s now also being used for over 70 other languages.

BERT, our new way for Google Search to better understand language, is now rolling out to over 70 languages worldwide. It initially launched in Oct. for US English. You can read more about BERT below & a full list of languages is in this thread…. https://t.co/NuKVdg6HYM

— Google SearchLiaison (@searchliaison) December 9, 2019

BERT is rolling out for: Afrikaans, Albanian, Amharic, Arabic, Armenian, Azeri, Basque, Belarusian, Bulgarian, Catalan, Chinese (Simplified & Taiwan), Croatian, Czech, Danish, Dutch, English, Estonian, Farsi, Finnish, French, Galician, Georgian, German, Greek, Gujarati…. (MORE)

— Google SearchLiaison (@searchliaison) December 9, 2019

However, last week at the Google Webmasters Conference in Zurich, Switzerland, it was revealed that BERT is currently only used for generating featured snippets in every language other than English.

John Mueller destroying SEOs who argued that BERT ruined their non-English websites' rankings.

While this will most likely change in the future, at the moment Google is only using BERT to better understand search queries in English. So whatever ranking volatility was recently noticed for pages in other languages, it wasn’t because of BERT!

Wrapping up

If you’re an SEO or a website owner working on a page in a language other than English, and you feel like your content hasn’t been performing well recently, I hate to break it to you:

It wasn’t BERT.

If you are struggling to diagnose the problem though, we are here to help!