Could MUM be Google’s next step towards becoming a semantic search engine? Discover how MUM may affect SEO in the future
Multitask Unified Model is a new technology for Google launched in May 2021.
In fall 2021, the forthcoming rollout was announced, and the technology was explained in more detail. Using artificial intelligence or natural language understanding and processing, MUM answers complex search queries with multimodal data.
MUM can process information from a variety of media formats in order to answer questions. MUM understands images, videos, and audio files in addition to text.
What is the MUM process?
In May 2021, Google introduced MUM as a 1000 times more powerful evolution of BERT. Both technologies utilize natural language processing. But MUM goes beyond natural language processing.
Using MUM, Google searches will become even more semantic and context-based. Google wants to answer complex search queries with MUM, which cannot be answered with a normal SERP snippet.
MUM was given the following tasks:
- MUM provides a deep understanding of world knowledge.
- To understand up to 75 languages, MUM will be trained simultaneously in up to 75 languages. Previously, each language was trained in its own language model.
- MUM should also be able to comprehend texts, images, audio, and video.
Google MUM is Google’s next major paradigm shift toward more performance and user-centricity
After Hummingbird, RankBrain, and BERT, MUM is the next major paradigm shift for Google search. The previous innovations based on machine learning used several trained models for different tasks, but at MUM, the goal is to use just one model for all indexing, retrieval, and ranking tasks.
By focusing on only one language model, the consideration of different languages for semantic interpretation becomes obsolete. The algorithms are trained using English-language search queries and documents. Moreover, they can be applied to all other languages – a significant advantage from a performance and semantic perspective. Natural language processing makes English easier to interpret than grammatically more complex languages such as German.
Google has always prioritized English as its primary language. The first translations of English-language documents appeared in knowledge panels as early as 2019.
This represents a significant performance improvement. If resources can be used efficiently, machine learning or natural language processing can be used. Reducing the number of processes running simultaneously is a prerequisite for this. Focusing on querying information from just one model for each search query results in improved efficiency and performance.
Research published by Google in 2020 titled “Multitask Mixture of Sequential Experts for User Activity Streams” describes a technology called MoSE that is similar to MUM in many ways.
In a data model based on clicks and search history, MoSE can summarize very efficiently. In a similar manner to classic search engines, it operates as a market research tool, starting with the search engine user rather than the data indexed by the engine. The focus is on the user intention, and Google can predict which questions and corresponding answers the user will need during his search.
The SERPs can compile all the information required to accompany the user seamlessly through the customer journey.
MUM offers new opportunities for Google shopping along the customer journey
Google has lost ground to Amazon and smaller e-commerce platforms when it comes to product-based searches. In the preference phase of the customer journey, many users look for products directly on Amazon. From an economic standpoint, this is challenging for Google since these users or commercial searches account for the majority of clicks.
Google users use it for information-oriented searches during the awareness phase. During the preference phase, Google is losing many users to its competitors.
Google wants to provide users with valuable information in the early stages of the customer journey (awareness and consideration). Google aims to inspire users, provide a comprehensive overview, and help them make a purchase by redesigning the SERPs and shopping search.
In the preference phase, Google has given up the direct fight for product searches and is concentrating on its core competencies. The organization and processing of the world’s knowledge in an easy-to-use format. E-commerce platforms of this size are unable to keep up.
What SEOs can learn from the future of Google search
With MUM, Google is completing its path to a semantic search engine that continuously improves the context of search queries and content. Thus, the relevance of content and content passages to match search intent (more about Google’s steps to a semantic search engine in my article Google’s way to a semantic search engine).
Until a usable quantum computer is developed, Google will need to use effective technologies such as MUM to access the currently lacking computing power for big-scale machine learning. Google can develop its own search systems more quickly this way, without having to worry about hardware performance issues. One could argue that software development now outpaces hardware development.
It is predicted that commercial quantum computers will become available in 2029. It is therefore safe to assume that Google search will be a semantic search engine by then. If this happens, keyword text matches in Google search will no longer exist.
Google MUM: SEO must change
The question must also be asked at this point as to what role Google will play as a traffic provider in the future. In addition, SEOs will still have an impact on ranking positions.
In the same way that Panda and Penguin brought changes to the industry, BERT and MUM did as well. Hummingbird and Knowledge Graph are powered by natural language processing to speed up semantic search. Instead of focusing on keywords, SEOs should think about entities and topics related to E-A-T.
It is imperative to ensure that search-relevant content can be crawled and indexed. But technology does not make content relevant, nor does it create authority or expertise. There are a few small levers to intervene when it comes to trust (https) and user experience (page experience). These levers, however, do not guarantee top position. Using natural language processing, Google’s understanding of structured information will become less and less necessary, reducing the need for technical tasks such as marking up.
Content and links remain the most influential factors. Other critical factors also contribute to authority. Search queries and content (text, video, audio and images) co-occurring are significant trust and authority signals. Google has broader access to data sources and information through MUM. Furthermore, Google can use language-independent data mining to collect and merge all information about entities and topics around the world. Old data silos are being opened.
As a result, Google can answer questions even better and impart really deep knowledge.
Managers of content should pay less attention to keywords in their content and consider the perspectives from which a topic should be approached. TF-IDF analysis remains one of the most effective ways to identify important terms in a keyword corpus.
The content provides answers to questions. However, producing content alone will not be sufficient in the future. In order to transfer product-related traffic to their own shopping world, Google would like to answer questions throughout the entire customer journey. They want to regain market share.
As SEO continues to become more relevant, it is becoming increasingly important for the person responsible for the content to provide content marketing along the customer journey. This is in order to provide as many content touchpoints as possible during the customer journey.
Depending on their level of knowledge, users undergo a research process over a shorter or longer period of time. Users confront a variety of challenges and questions in their search for solutions with a growing knowledge base.
Someone new to the topic of search engine optimization is more likely to ask the question, “What is SEO?” Next, they ask, “How does SEO work?” only to realize that the topic is quite complex, and they are more likely to ask “Who offers SEO services?” On this journey, companies should provide the answers.
User-centric content must anticipate needs and questions throughout the customer journey, like Google does with MUM. An in-depth SERP analysis can predict future search intentions.
SEO also applies to media formats other than text in Google MUM
SEOs are primarily concerned with text content. As Google becomes better at understanding video, images, audio and text and placing them in context, MUM makes the SERPs significantly more diverse in terms of media formats. If you look at the automated marking of places in YouTube videos, or the classification of images in the image search, you can already see it.
Google visitors through MUM to decline?
Google wants to display more answers in the SERPs without the user having to click on the source again with innovations like MUM and BERT. The concern is that Google will turn off the traffic tap and display as much information as possible in its own world.
In this case, the interests of Google and the content publisher may diverge, and Google may use the appropriate content passages without the publisher’s consent. That’s up to Google and how they balance the interests.
Google relies on current content to answer current and future user questions. Google is a technology group that can index information algorithmically and prepare it in a user-friendly manner.
Despite this, it will probably be impossible to independently build up in-depth specialist knowledge and display it independently of publisher content. One can therefore trust that Google will keep rewarding good content with traffic.