February 3, 2021
BERT and… SEO (You were expecting Ernie?)
In the olden days of a few years ago, everyone was stressing
about keywords when looking at Search Engine Optimization (SEO).
What keywords are you targeting? How often do they appear
and how prominent are they in your content?
Keywords provided not only a focus for the content, but also
acted as beacons to search engine bots looking to index your website.
Some SEO experts recommended bloating your content with
singular and plural versions of your targeted keywords. Still others instructed
using common misspellings, too, just in case. Anything to use the system to
But then BERT entered the metaphorical room. BERT (Bidirectional Encoder
Representations from Transformers) is changing (for the better) how we
think of and act on SEO.
In short, it changes our mindset from “how search engines
look at our content” to the more important “how users (i.e., potential
customers) look at our content.”
What is BERT?
BERT is a newish, open-source AI “neural network-based technique for natural
language processing (NLP) pre-training” that Google uses for its searches. In 2019, BERT was responsible for about 10% of all Google searches. One year later, as announced during Google’s Search
On event, BERT is handling almost 100% of
BERT is optimized for “for longer, more conversational queries,” as opposed to just quick search terms. In
very simplistic terms: rather than just analyzing content based on locating keywords
within it, BERT is looking at how the words interact with each other in a
sentence, both from left to right and right to left. Or, in other words, what
the sentence is actually about.
the context of content, BERT can better evaluate the quality of
the content, especially when considering incoming search queries. What this
means is, search engine results pages (SERPs) are getting more and more
accurate. And when SERPs are more accurate, users are happier.
understanding BERT important to SEO?
Basically, BERT was designed and trained to digest
content the same way a person would. And people don’t tend to ask each other
one- or two-word questions. You don’t look at your friends, for example, and
say, “Best restaurant.” Instead, you ask, “Where is the best restaurant in this
area for getting some delicious tacos?”
Before BERT, searches would take your “best restaurant” and
deliver results based on that short-tail keyword. If the permissions were there
to collect other information (e.g., your location), search would also absorb
any other data about you before displaying SERPs.
But now that BERT’s on the scene, it can handle more complex and
conversational searches. Using the previous example, BERT parses the question,
“Where is the best restaurant in this area for getting some delicious tacos?”
and understands the relationships between “best restaurant,” “in this area,”
and “tacos.” It knows the search is for all of these put together, compares the
search criteria with all its learned knowledge of web-based content, and
delivers better and more precise SERPs.
could say that [BERT’s] kind of going in the direction of decreasing the
importance of exact matches for your keywords in your content…. But it’s not
that that is the goal of these algorithms. But rather our goal is to understand
the big vast amount of content out there a little bit better so that we can
show the right versions to users when they ask.”
– John Mueller, Senior Webmaster Trends Analyst at
In essence, BERT is added AI “brain” power to support the
digital assistant revolution. Siri, Alexa, Cortana, Google Assistant—every
major platform seems to have its own speech-powered assistant. Voice search is
being used more and more. Microsoft’s 2019
Voice Report found that 72% of respondents have used digital assistants for
search. It’s a good bet, too, to expect the future of search to eventually become
fully voice driven.
Leveraging BERT’s strengths in your favor
Adapting your SEO plan for the BERT revolution is actually
pretty easy. We break it down into three simple steps:
1. Hunt long-tails.
Keyword research is still
important. You need to know what terms your users are searching, so you can
craft your content more appropriately and strategically. But when deciding what
terms to focus on, look to long-tail keywords.
These phrases are more in
tune with how voice search users search for information AND for how BERT
processes content. Long-tail keywords are also inherently more specific, which
means users who find your content will be better matches for consumers of your content.
2. Write like you speak.
Since BERT analyzes the
relationships of words in your content to determine meaning, it only makes
sense to create content that is more conversational. But conversational doesn’t
just mean including lots of slang or funny expressions—it means language that is
more accessible than formal academic writing. Content that is okay having short
sentences. Why not? Content that can end in a preposition, because that might
be exactly what users are searching for. Contact that can start with “and” or
“because.” Because conversational content is simply clear writing that anyone
3. Share your content on
Consuming content is often a
social activity. Indeed, a report from Backlinko shows
that voice search results have a large amount of social media engagement. So,
sharing your content on social media is no longer just about enticing users to
visit and providing added value—BERT likes it too. An interesting side fact
from that report: users are also eating up long form content, which almost
seems contradictory to social media-sized brevity. The overall verdict seems to
be: write quality content, write a lot of it (both in quantity and length), and
share it on social media.
Since BERT is different from how search engine algorithms
have worked in the past, there’s really no more gaming the system. Just think
of BERT as looking to deliver the best, most-applicable, and
easiest-to-understand-and-use content to users. Which is exactly what your goal
should be too. Create glorious content and BERT will make sure the right people
find it and you.