How does natural language processing affect SEO? You may have heard the term before. With foundations dating back to 2011, this development has an increasing influence on search updates, which I will explore here.
Research has continued since 2011. In 2015, as neural network models became more popular, research and development into NLP-based language models accelerated.
Google did not achieve its breakthrough until 2019. BERT (Bidirectional Encoder Representations from Transformers) was the first NLP system developed by Google and successfully implemented in its search engine. It uses Google's own Transformer NLP model, based on a neural network architecture.
The announcement was significant: Google said BERT would immediately affect 10% of global queries. Two years after implementation, in 2021, Google announced that it supported 99% of all English search results.
After BERT, Google announced SMITH (Siamese Multi-depth Transformer-based Hierarchical) in 2020, another NLP-based model refined beyond BERT.
- Google SMITH: Its advantage is understanding passages within documents as BERT understands words and sentences, allowing longer documents to be understood. SMITH is a new model intended to understand entire documents. BERT was developed to understand words in sentence context; SMITH, in simplified terms, understands passages in document context. BERT is trained to predict randomly masked words from sentence context, while SMITH is trained to predict the next sentence block.
Compared with BERT, SMITH offers better processing speed and understanding of long-form content, further helping Google improve result quality and build datasets.
Later that year, Google refreshed its open-source, transformer-based NLP model to launch GTP-3 (Generative Pre-trained Transformer 3), trained through deep learning to produce human-like text.
- Generative Pre-trained Transformer 3, or GPT-3, is an autoregressive language model using deep learning to produce content similar to human writing.
In 2020, Google made another announcement signalling further NLP research and development. It subsequently announced LaMDA (Language Model for Dialogue Applications), another NLP algorithm using multiple language models, including BERT and GPT-3.
LaMDA is presented as 1,000 times faster than BERT. Trained on dialogue, it can conduct natural conversations, and Google highlights its ability to begin open-ended exchanges.
This means it is trained to read and understand many words or an entire paragraph, infer context from their relationships and predict the words that should follow.
How Does the BERT Update Work?
Historically, language models read text sequentially either left to right or right to left, but not both simultaneously.
BERT is designed to read in both directions at once. This bidirectional capability supports pretraining on two related NLP tasks:
Masked Language Modelling
A masked language model (MLM) predicts a hidden word from its context within a sentence.
Next Sentence Prediction
This training objective predicts whether two given sentences have a logical connection or are randomly associated.
With the BERT update, natural language processing became a core component of every SEO effort.
You may have noticed Google's recent focus on frequent broad core updates. Their impact is greater than other releases, such as the latest Link Spam and Page Experience updates.
According to Google's official blog, being affected by a broad core update does not necessarily mean a website has significant SEO problems. Google advises these sites to focus on improving content quality.
As parts of Google's core algorithm, BERT and SMITH continually learn from datasets users supply every second and suggest the kinds of content people want to consume.
These suggestions may concern intent behind a long-tail query or voice search.
Google says sites with more relevant content gain an advantage through broad core updates. This suggests closer scrutiny of queries and the intent of highly ranked content.
When BERT launched in 2019, its impact was limited to 10% of queries. That was only the beginning of a major update.
As a machine learning algorithm, BERT has self-learning abilities that improve as more datasets arrive.
Google has confirmed that 15% of the queries it encounters are new and have never been used before. This is particularly relevant to voice search, where people do not use predictive search.
According to Google, BERT is now present throughout search and determines 99% of English results. With more datasets produced over two years, it has improved. BERT can be described as the most significant development in Google's linguistic AI capabilities.
How Does Google's NLP Algorithm Shape Search Results?
Google rightly sees NLP as its future because understanding intent matters more than ever. Webmasters and content developers therefore need to focus on what users actually want.
Thinking about content only through keywords is becoming less important. Specific, descriptive content answering the immediate questions people have when searching is increasingly valued.
Put simply, Google does not look merely for individual phrases. It tries to understand sentence context and meaning to determine whether your content is better than existing top results.
More interestingly, BERT can even understand the context of links within an article, once again making quality backlinks an important part of ranking.
Optimising a page for just one keyword is not the way forward. Optimise for related topics instead and include supporting content.
With that in mind, make the content as informative as its subject requires and, above all, answer users' critical questions.
NLP's Effect on Backlinks
Content quality and topical depth matter greatly. That does not mean internal and external links no longer matter.
NLP and BERT help Google understand link quality. The NLP algorithm can determine whether link structure and placement are natural and understand anchor text and contextual relevance.
If the sentiment around an anchor is negative, the effect may be negative. If a link is placed in an irrelevant paragraph merely to obtain backlink value, Google is now equipped to ignore it.
You cannot manipulate a ranking factor simply by placing a link on any website. Google's NLP algorithm determines whether the link appears on a relevant site publishing relevant content and within a natural context.
Google aims for links to improve user experience and provide access to additional information people need.
With advances in NLP, Google can monitor such attempts at ranking manipulation. Keep these factors in mind before spending your budget the next time you build backlinks.
Comments