As Google continues to embrace artificial intelligence and machine learning in Google Search, one might wonder how these technologies enhance its performance. Since 2015, when Google introduced its first AI system named RankBrain, the company has continually deployed AI systems to better understand language and improve the search results presented to users.
Several months ago, we inquired with Google about how it uses AI in search, including systems like RankBrain, neural matching, BERT, and the latest breakthrough, MUM. We aimed to gain a deeper understanding of when and how these AI tools are utilized, how they work together, their evolution over the years, and what search marketers should know regarding Google’s use of AI in search.
We spoke with Danny Sullivan, Google’s Public Liaison for Search, to clarify these questions. In essence, RankBrain, neural matching, and BERT are integral to Google’s ranking system across many queries. They focus on understanding the language of both the query and the content being ranked. On the other hand, MUM is not currently used for ranking purposes; it is only used for tasks like COVID vaccine naming and enhancing related topics in video results.
It Starts by Writing Content for Humans
Google consistently advises creating content for human readers. In the early days of SEO, when algorithms were simpler, SEO specialists often crafted content tuned for various search engines. Nowadays, with complex algorithms incorporating machine learning and AI, these systems understand language more naturally. As such, writing for human comprehension ensures that the algorithms will also understand the content. This article aims to explain how Google incorporates AI in its search functions rather than offering specific SEO tips for optimizing sites for particular AI systems.
Overview of AI Used in Google Search
RankBrain: Introduced in 2015, RankBrain was Google’s first AI in search, designed to understand how words relate to broader concepts. Initially used in 15% of queries, it is now integrated into queries in all languages and regions. RankBrain aids in ranking search results.
- Year Launched: 2015
- Used For Ranking: Yes
- Languages: All
- Common Usage: Most queries
An example is searching for “what’s the title of the consumer at the highest level of a food chain,” where RankBrain deduces that the concept relates to animals, identifying the answer as “apex predator.”
Neural Matching: Released in 2018 and expanded in 2019, neural matching examines queries in full context. It is widely used across languages and regions for ranking search results.
- Year Launched: 2018
- Used For Ranking: Yes
- Languages: All
- Common Usage: Most queries
For instance, searching “insights how to manage a green” would be interpreted by neural matching to mean management tips based on a color-based personality guide.
BERT: Launched in 2019, BERT (Bidirectional Encoder Representations from Transformers) helps understand word combinations and intents by analyzing word sequences on a page. Initially covering 10% of English queries, it now supports almost all languages and is used in most queries.
- Year Launched: 2019
- Used For Ranking: Yes
- Languages: All
- Common Usage: Most queries
For example, searching “can you get medicine for someone pharmacy” allows BERT to understand the intent behind picking up medicine for someone else.
MUM: The Multitask Unified Model, introduced in 2021, assists in understanding and generating languages. Currently, MUM is not used for ranking but supports all languages and regions.
- Year Launched: 2021
- Used For Ranking: No
- Languages: All
- Common Usage: Limited purposes
MUM enhances searches for COVID-19 vaccine information and is anticipated to improve search functions using a combination of text and images in the future.
AI Used Together in Search and Specialized for Vertical Search
According to Danny Sullivan, these AI systems often collaborate to process and rank queries. Google explained that these AI tools are designed to understand both the query and relevant content. RankBrain, neural matching, and BERT are employed globally in all languages supported by Google Search. Specialized AI systems are used for different search verticals like images, shopping, and local search.
Core Updates and AI Integration
While RankBrain, neural matching, and BERT are prominent in search queries, Google also implements core updates a few times a year, which can be more noticeable. These core updates often work in conjunction with the larger AI systems. Additionally, Google employs various machine learning systems that impact core updates beyond these three major AI systems.
This detailed overview highlights how Google utilizes advanced AI to enhance search functionalities and adapt to the evolving requirements of language comprehension and query processing.
