How Google Search uses AI

Are your SEO efforts not delivering the results you expect, and you can’t figure out why?

Traditional SEO tactics are becoming less effective by the day. While you’re focusing on keywords and backlinks, Google’s AI is evolving rapidly, fundamentally changing how search results are ranked.

This shift is happening behind the scenes, making it increasingly difficult to understand why your content isn’t performing as well as it should.

Understanding how Google’s AI systems work is key to adapting your SEO strategy. This article explores the evolution of Google’s AI – RankBrain, neural matching, BERT and MUM – and explains how these advancements are reshaping search.

By grasping these concepts, you’ll be better equipped to create content that aligns with Google’s AI-driven approach, improving your chances of ranking higher in search results.

Google’s AI systems

Google has been using some form of AI to identify, weigh and order URLs since around 2015, with its first AI system called RankBrain.

Three years later, Ben Gomes, Google’s Senior Vice President of Learning and Education and former Head of Search, called AI the “next chapter of Search.”

Gomes explained that AI will allow Google to realize a better user experience, not isolated to just the query. He said AI will create “three fundamental shifts” in how search works:

• From answers to journeys: “To help you resume tasks where you left off and learn new interests and hobbies, we’re bringing new features to Search that help you with ongoing information needs.”
• From queries to providing a queryless way to get to information: “We can surface relevant information related to your interests, even when you don’t have a specific query in mind.”
• From text to a more visual way of finding information: “We’re bringing more visual content to Search and completely redesigning Google Images to help you find information more easily.”

This shift started with RankBrain. 

RankBrain (2015)

The RankBrain system was the first step to help the search engine to “understand how words relate to concepts.”

Understanding the connection a word has to a concept is an intelligent activity and Google’s first step in understanding content like a human.

For example, if you search “What’s the color of the sky?” the AI could understand what “sky” is and that it has a perceived color. So Google could return a result that didn’t have the exact words but did answer the query.

A few years later, Google made more progress in connecting words to concepts with neural matching. 

Neural matching (2018)

This system/sub-system was created to help Google understand how “queries relate to pages” for concepts that are more difficult to understand.

Let’s say you search “tie my laces,” which could mean multiple things. With neural matching, Google could understand that “laces” means shoe laces and return results on ways to tie them.

BERT (2019)

BERT stands for Bidirectional Encoder Representations from Transformers and was considered a “breakthrough.”

Think about BERT as the evolution of RankBrain and neural matching, so now Google could understand how multiple words in a sentence relate to multiple words on the page and the concepts behind them.

BERT seems to be important for entity recognition. This can help google understand a brand name, who a person is and maybe even what their expertise is in a given topic.

This is the type of AI model that makes generative AI and AI Overviews possible. Google has been using it since 2019. 

• Related to BERT is a “deep learning system” called DeepRank. As we learned from Panda Nayuk’s testimony during the DOJ trial, essentially DeepRank is BERT when BERT is used for ranking.
• DeepRank also replaced much of RankBrain.

MUM (2021)

Google claims that the Multitask Unified Model (MUM) is “1,000 times more powerful than BERT.”

If BERT understands language, then MUM generates it. And it can also understand both text and images and maybe video by now.

Pandu Nayak, Google’s Chief Scientist, Search and former VP of Search, explained MUM like this:

“Take the question about hiking Mt. Fuji: MUM could understand you’re comparing two mountains, so elevation and trail information may be relevant. It could also be understood that, in the context of hiking, to “prepare” could include things like fitness training as well as finding the right gear. 

Since MUM can surface insights based on its deep knowledge of the world, it could highlight that while both mountains are roughly the same elevation, fall is the rainy season on Mt. Fuji so you might need a waterproof jacket.”

However, MUM’s application to improve search results around COVID-19 vaccine information highlights how powerful this system is.

Nayak said MUM helps to differentiate the different vaccine brand names and provide the “latest trustworthy information about the vaccine.”

MUM highlights that Google can improve search results faster than in the past.

Harnessing AI for SEO: What’s possible?

What you can do with generative AI, Google can do with the AI in their ranking system. Let that sink in.

ChatGPT may have an IQ of up to 155, so it’s fair to assume that Google’s AI can vet sources like a human to a degree. 

A human vetting the qualify and relevance of a page to their intent might ask these questions:

• Are you an experienced expert in the subject you’re writing or talking about?
• Are other experienced experts talking about you and your expertise?
• Do you have a bad reputation for spamming Google to rank higher?
• How does what you say about a topic relate to other experts in the field?
• Is this the best product for what I’m searching for?

But remember that Gomes said AI will move “From answers to journeys.” This is very important, indicating that Google can track how you and your audience are engaging with or creating content about your brand or internal experts.

With this, then Google could answer much more relevant questions:

• Do people benefit from your product or service?
• Is one website/company affiliated with another or different, with customers that use both?
• Are customers sharing information about your product and then searching for it on Google?

It’s time to stop thinking about SEO in terms of ranking signals and focus on how humans search for information and why.