MUM is an acronym for Multitask Unified Model, an AI-based technology created by Google. It was first announced in May 2021 by Pandu Nayak.
In September 2021, Google said that it would be integrating MUM into Google Search in the following months.
But what exactly is MUM, and what does it imply for the future of Google Search and for SEO? Let’s take a look at Google’s history with advanced technologies and see how that has brought us to the age of MUM.
Google’s Evolution
In the beginning, Google was a relatively rudimentary search engine that returned web pages based purely on how many words on the page matched words in your search query. As there was no initial means of comparing the quality of the different pages that matched the query, this early model was quickly gamed by website owners, leading to a lot of spammy and irrelevant results for searchers. This was slowly tamed by Google over the years as their algorithm adapted to the rapid expansion of the internet, and they developed a much more sophisticated process for curating and ranking pages.
Google’s next, and much larger, project for improving their search results was to get the algorithm to understand syntax and semantics so that it could refine search results to match searcher intent that much more closely.
Google’s History with AI Technology
In 2010, the official Google blog published a post about helping computers to understand language. The opening paragraph reads:
“We can write a computer program to beat the very best human chess players, but we can’t write a program to identify objects in a photo or understand a sentence with anywhere near the precision of even a child.”
This is a fascinating excerpt from a little over a decade ago, and MUM represents the biggest stride forward on these kinds of understanding. Let’s take a look at how Google evolved from here to today.
2015: RankBrain
Google continued to refine their machine-learning abilities in the following years, and in 2015 officially announced RankBrain, their first attempt at using AI in Google Search. RankBrain changed Google’s process for refining search results. It advanced Google from analyzing literal strings of characters and words into analyzing the overall subject matter or intent behind search queries.
Unrecognizable words or phrases RankBrain encountered helped it determine meaning based on the surrounding contextual information and similar language.
To learn more, check out this Moz article on RankBrain.
2018: Neural Matching
In 2018, Google began making use of neural matching across 30% of terms. This was later rolled out as part of the process to generate local search results in 2019. Neural matching helped Google better relate the words used in queries to larger concepts.
The use of neural matching means that Google can do a better job going beyond the exact words in business name or description to understand conceptually how it might be related to the words searchers use and their intents…
— Google SearchLiaison (@searchliaison) December 2, 2019
Danny Sullivan, aka Google’s Search Liaison, described neural matching as a “super-synonym” system:
“For example, neural matching helps us understand that a search for ‘why does my TV look strange’ is related to the concept of ‘the soap opera effect.’ We can then return pages about the soap opera effect, even if the exact words aren’t used…”
2019: BERT
BERT is a transformer-based machine learning technique for natural language processing pre-training. This helped Google strengthen their algorithm’s understanding of natural language used by searchers.
Google’s machines learned that prepositions like ‘for’ aren’t always disposable ‘little words’, because there are some cases in which the meaning of the larger query hinges on them. It then used this framework to learn context and thus meaning by tracking relationships in sequential data.
For example, if you search for “can you get tickets for someone box office,” BERT understands that you’re trying to figure out if you can pick up tickets for someone else. Before BERT, the short preposition was taken for granted, mostly surfacing results about how to get tickets at a box office for yourself.
BERT was also able to learn from language improvements in English, and apply them to other language models. This expansion into other languages helped Google improve entity identification, and relationships between entities, with higher precision than ever before, paving the way for MUM.
2021: MUM
MUM uses the T5 text-to-text framework, and is 1,000 times more powerful than BERT.
T5 uses transformer-based architecture, just like BERT, but instead uses a text-to-text approach. What that means is with T5, the input (query) and output (result) are always text strings, in contrast to BERT-style models that can only output either a classification label or the span of the input into a question and answer format. This means that the output with BERT, while undeniably impressive, was in comparison to MUM still rather abstract.
The T5 text-to-text process includes more in-depth machine translation, document summarization, question answering, and classification tasks (e.g., sentiment analysis).
How Will MUM Affect Google Search?
MUM aims to eliminate the need to search multiple times around a topic to gather the information that you need by delivering results centered around the complete answer the first time around.
The distilled way that we have learned to interact with a search engine might evolve and become more advanced, and look more like the way we would interact with another human.
MUM Understands Language
One of the most groundbreaking features of MUM is overcoming language barriers, understanding and even generating content in 75 different languages. This means that MUM can learn from sources that are written in another language, and help bring that information to you from relevant results written in your language.
MUM Is Multimodal
MUM aims to apply its technology across multiple media types including image, video and audio. This means that if you have a specific question, and the best answer is a video or audio result, MUM can surface the relevant portion of that piece of media for you in search results.
MUM and Visual Search
MUM is especially powerful when it comes to visual-based search.
With Multisearch, a Google Lens feature, you can (for example) carry out a visual search for a shirt, and refine the search with additional text asking questions about it, such as its availability in another color, or whether the pattern can be found on other pieces of clothing. Google will then return results that answer these questions. This might be a shopping experience, more images, or simply more information to help you on your journey.
This Google article explains how you can even take a photo of your hiking boots and ask questions outside of the realm of style, asking something like “can I use these to hike Mt. Fuji?”. MUM would understand the image and connect it with your question to let you know your boots would work just fine. It could then point you to a blog with a list of recommended gear.
You can also take photographs and simply ask Google ‘what is this?’ or ‘how do I fix this?’, and MUM will decipher the image, and return the most suitable results. A complete game changer for home DIY.
Things to Know
With MUM also comes the introduction of new features in the SERP.
With ‘Things to know’, if you search for any given topic, Google will understand how people typically explore said topic, and show the aspects people are likely to look at first.
In the example shown at left, we can see how the topic of ‘acrylic painting’ generates a ‘Things to know’ panel containing step by step information, styles, tips, how to clean, and so forth.
Broaden and Refine Search
The ‘Broaden’ and ‘Refine’ features provide the user the ability to zoom in and out of any given topic.
In the example shown at right, you can see how the user is presented with options to narrow down the topic of acrylic painting with ideas, techniques and online courses, or broaden into the realm of painting styles and famous painters.
Visual Inspiration Upgrade
Another feature is designed to satisfy searchers looking for visual inspiration.
Search queries like “Halloween decorating ideas,” or “indoor vertical garden ideas,” will generate a visually rich page full of ideas with articles, images, videos and more.
Related Topics for Video
Based on its advanced understanding of information, MUM can show related topics for videos, even when the topic isn’t explicitly mentioned in a video.
A Greater Search Experience
It’s clear to see that Google has surpassed the basic input/output model — receiving search queries and returning web documents that seem to match — and is becoming an expansive discovery and learning tool. While this is great for end users, and especially great for Google (since it helps keep users within the walls of their ecosystem), it all arguably comes at the expense of hard-working publishers and brands who create the content Google curates.
For some website owners these enhancements are a present pain-point — stealing the click, so to speak — and MUM might make things even worse for them. Let’s take a look at how this might affect online retailers specifically.
The Messy Middle
The ‘messy middle’ is defined as the space between awareness and purchase where shoppers move fluidly between two mental modes — exploration and evaluation — as they learn more about their options.
The graphic below is a visual representation of the Messy Middle model.
As you can see, the exploration and evaluation phase is representative of an eternal loop that the user finds themselves in as they research before eventually making a purchase.
The argument can be made that Google is responsible for keeping users inside the exploration and evaluation loop by presenting content from websites in the SERP through featured snippets, People Also Ask, and their own features like carousels or Related Topics panels.
There are multiple facets available to filter and refine the search process beyond even what the searcher initially may have intended.
For example, if you search for “two person outdoor hot tubs”, Google immediately presents options to narrow that search to nearby results, a specific hot tub shape, or reviews.
Upon clicking the ‘Square’ option, we can see how visually overwhelming the search results are:
The bottom half of the above screenshot is Google’s own Shopping experience. Upon selection of any given option, the user is presented with a list of retailers that provide the product.
Where a simpler SERP might have been an asset in driving the user from discovery to purchase, this dramatically busier SERP might keep users inside this messy middle for longer.
Google shows no signs of slowing down with the enhancement of the search experience on their platform, and we must prepare for what will likely only become a busier, more complex SERP, which might force us all to spend more time in the messy middle on more and more kinds of queries.
What Does MUM Mean for SEO?
At its core, MUM is about building deeper connections between concepts and learning how people search for information within topics. But the goal of search engine optimization remains the same: to earn the click to our website. With MUM, the search results are going to become more diverse and competitive, and laden with more features to keep searchers trapped in the messy middle. If your content strategy isn’t oriented to this new reality, it may result in fewer organic visitors over time.
So here’s the million-dollar question: how can you continue earning those clicks in a post-MUM world? How can you make MUM work for your website instead of against it?
Create Content Across Many Media
Harness the multimodal aspects that MUM aims to highlight. Aim to understand the connections between entities of any given topic, and incorporate those findings into a comprehensive content strategy that spans different media.
Create videos. When you write an article, include or create a YouTube video to supplement it. A video doesn’t have to be a major production. It can be a simple point-and-shoot reading of the article that you have written already. This allows you to create clear chapters that Google can surface in video results in addition to the article. There are many tutorials explaining how to make a YouTube video.
Use your own images. While stock photography can be a nice quick shortcut to getting your article finished, an original authentic photograph or screenshot is likely far more useful to the subject in question.
Continue to follow SEO best practices for these media. For images, choose high-quality photographs that contain succinct alt text, along with descriptive titles, captions and file names. Also, make sure that the image files you use are responsive and properly sized.
For videos, add subtitles and closed captions, and take care to observe technical video best practices.
Leverage Long-tail Keyword Research
Keyword research is still important. However, we should aim to look beyond optimizing a page for just a single keyword, and instead understand the bigger picture that your ideal target keywords fall under.
MUM will make conversational search possible. Conversations naturally evolve and elaborate, so we should be mindful of that when creating content moving forward if we want our content to stay competitive.
Understanding the long-tail keywords being used currently around the topics will help drive this initiative.
- Think about what larger purpose the searcher is trying to fulfill with their searches
- Ask how you can anticipate the next steps that they will take
With MUM aiming to provide the most complete answer to the user, we may see an increased presence of forums like Reddit, FAQs, and blog posts with comments, where long-form, in-depth questions are being asked and answered.
All of these areas are great places to research and understand the questions that people are asking about topics in your industry. Leverage these conversations and create comprehensive content around them.
Build an Experience On Your Website
With MUM making the ‘messy middle’ even messier, and broadening the window of results to more competition, it places more emphasis on making your site as sticky as possible.
Try to avoid making your content just part of the journey and do what you can to retain visitors, helping them exit the ‘exploration and evaluation’ cycle.
Aim to give visitors a reason to stay on your site. Keep them engaged with relevant content or with interactive elements like videos, forums, and tools.
RevZilla continues to be my go-to example of a website that has a fantastic ecosystem of motorsport content, products and conversation.
RevZilla uses images, videos and articles to keep the user engaged. Furthermore they do a great job of blending educational content with marketing their products.
Take an article like ‘What’s the best beginner motorcycle?’ as an example.
The topic itself is poised for an open conversation beyond the opinions of the author, especially on a website built for motorcycle enthusiasts.
Fortunately RevZilla recognized this and kept the conversation going in the comments of the article. This is the type of content and user interaction that will help an article like this surface for those long-tail search terms.
Invest in Brand Awareness
Create content that builds brand recognition and customer loyalty. Much like RevZilla has done, work towards making your brand synonymous with a product, service or topic.
Cover niche topics that your visitors would appreciate, and foster a welcoming environment that would make them want to return for more content and user experience like it in the future.
Use Structured Data
Add semantic markup to entities across your site to help Google identify them.
One of the best ways to avoid ambiguous guesswork for a robot is to talk their language. As Google’s own documentation states: “Structured data is a standardized format for providing information about a page and classifying the page content; for example, on a recipe page, what are the ingredients, the cooking time and temperature, the calories, and so on.”
All of these aspects fall directly in line with the new ‘Things to know’ feature that MUM will be presenting to searchers.
Structure your data accordingly using relevant headings, and make use of unordered and ordered lists where appropriate. Seize the opportunities presently available to enhance your search presence and explore schema.org to see if there are other markup types that help best define your content.
Strive for Expertise, Authoritativeness and Trustworthiness
People will still rely on authoritative sources, not just the opinions of others. So work on demonstrating your expertise, authority and trustworthiness.
- Respond to comments on blog posts, and user reviews on other platforms. Let customers and others know that you are an active, caring brand.
- Participate in Q&A’s, podcasts and webinars of influencers in your industry.
- Develop a knowledge sharing system or knowledge base that not only helps visitors ask and answer questions, but also provides you with insight into the questions that your users are asking that you can create content about.
- Show that you are a reliable source of information for your chosen topic or industry through accolades, partnerships, or participation in local meetups and other events.
Be the Wikipedia of Your Industry
All of these concepts have been around for quite some time and nothing should come as a big surprise. Ultimately Google will continue to reward truly useful, quality content, and that should remain the focus as they roll out new AI technologies and core updates.
I often say to clients: be the Wikipedia of your industry. Create comprehensive detailed content that gives visitors no room to leave to find the answer to their next question. Great content is hard to ignore for both users and search engines. So focus on creating something that is truly valuable, and unique. Ask yourself how you can present the things that your users want to see in a way that is impossible not to be shared by others.
