MUM Algorithm: Google's Multimodal Search Understanding Engine

What is MUM

MUM (Multitask Unified Model) is an AI-based search understanding model announced at Google I/O in May 2021. It succeeds BERT and exceeds it in three core capabilities.

1. Multilingual: Understands and processes 75+ languages simultaneously without translation
2. Multimodal: Understands text and images together
3. Multitask: Performs complex reasoning and multi-step tasks beyond simple keyword matching

Google announced MUM is 1,000x more powerful than BERT — a comparison based on complex information processing ability, not parameter count alone.

Complex queries MUM handles

Traditional search worked well for simple fact lookups ("What is Seoul's population?"). MUM handles complex questions requiring multi-step reasoning.

Example of a question requiring 8 search steps before MUM:

"I've climbed Mount Fuji before. How should I prepare differently to climb Kilimanjaro in autumn?"

Fully understanding this requires:

• Grasping Fuji altitude, difficulty, and weather conditions
• Grasping Kilimanjaro altitude, difficulty, and autumn weather
• Comparing differences between the two mountains
• Reflecting autumn season conditions
• Integrating additional equipment, fitness, and altitude sickness information

MUM can process all of this in a single query.

MUM's technical structure

MUM is based on the T5 (Text-to-Text Transfer Transformer) architecture. While BERT is encoder-only, MUM has an encoder-decoder structure with bidirectional ability to understand and generate text.

MUM search application examples

1. Multimodal search (Lens + MUM)

When a user photographs hiking boots with Google Lens and asks "Are these shoes suitable for climbing Mount Fuji?", MUM processes image and text simultaneously.

2. Cross-language information integration

If the most detailed information about Kilimanjaro is in Swahili, MUM can understand it directly without translation and deliver it to English and Korean users.

3. Topic refinement

Searching "Olympic gymnastics" lets MUM identify related topics (floor, horizontal bar, rhythmic gymnastics, etc.) and suggest refined exploration options.

4. Foundation for AI Overviews

2023's Google SGE (Search Generative Experience), now AI Overviews, combines MUM and Gemini to provide AI-generated answers to complex queries.

SEO strategy in the MUM era

What changed in SEO strategy with MUM:

Shift to topical authority

MUM evaluates site-wide topic coverage, not single-page keywords. Building a deep content ecosystem around specific topics became important.

Strengthen multimodal content

Pages with related images, infographics, and video alongside text are advantaged in MUM environments. Use descriptive alt text and filenames for images.

Proactively address complex user questions

Cover complex questions users actually ask in FAQs and content — not just simple keywords. Write content answering comparison and conditional questions like "difference between A and B" and "how to do Y in situation X."

Value of multilingual content

MUM integrates information across languages, so high-quality content in one language may be used for global queries. Conversely, niches lacking English content offer first-mover opportunities for content in other languages.

MUM and AI Overviews relationship

At Google I/O 2024, Google expanded AI Overviews globally. AI Overviews show AI-generated summary answers at the top of search results.

MUM's role in AI Overviews:

• Grasp complex query intent
• Integrate information across languages and sources
• Combine image and text interpretation

Actual generation is handled by Gemini-family models. To get cited in AI Overviews, content needs topical authority MUM evaluates and the clear facts, sources, and structure Gemini prefers.

MUM impact in local markets

Korean is among the 75+ languages MUM supports. AI Overviews for complex Korean queries on Google are gradually expanding.

Naver is developing in a similar direction with its own AI HyperCLOVA X, powering Naver AI search.

Frequently asked questions

Q. What concrete search result changes occurred after MUM?
A. Richer information began appearing for complex comparison queries and deep topic searches in travel, cooking, sports, etc. Google Lens integration and AI Overviews are the most visible changes from MUM adoption.

Q. Is short content still competitive in the MUM era?
A. Concise content with accurate facts remains effective for specific queries. However, in-depth long-form content is more advantaged for complex comparison and reasoning queries in MUM environments.

Q. Did image SEO become more important after MUM?
A. Yes. MUM treats images equally with text, so accurate alt text and meaningful filenames on relevant images became more important.

Q. What is the difference between MUM and ChatGPT?
A. MUM is Google's internal AI for understanding search results; ChatGPT is an independent conversational AI service. Both are transformer-based, but MUM integrates with search indexes and specializes in real-time web information.

Q. What is the most important thing to do now for MUM optimization?
A. Strengthening topical authority is key. Build content clusters covering related questions in your expertise area, add multimodal elements like images and video, and write FAQs answering complex comparison and conditional questions.

Related sources

• Nayak, P. (2021). MUM: A new AI milestone for understanding information. Google Blog. https://blog.google/products/search/introducing-mum/
• Google Search Central (2024). AI Overviews Help. https://support.google.com/websearch/answer/14901683
• Google I/O 2021. MUM: A new milestone in search. https://events.google.com/io/