Products SEO & GEO Agent Paid Ads Agent Website Agent Conversion Agent
How it works Pricing FAQ Blog Book a call

Google's AI ranking systems: RankBrain, BERT and MUM explained

Updated · 6 min read

Short answer Google ranks pages with several AI systems. RankBrain (2015) links words to concepts, neural matching (2018) matches the idea of a search to the idea of a page, and BERT (2019) reads how words combine. MUM isn't used for general ranking. None can be targeted directly: write clear pages in your customers' words.

Want help applying this to your business? Talk it through on a free 30-minute call.

Book a call

Which AI systems does Google use to rank pages?

Three AI systems help decide where your page lands: RankBrain, neural matching and BERT. A fourth, passage ranking, scores single sections of a page. MUM and Gemini are newer and bigger, but Google says MUM doesn't rank ordinary results. SpamBrain removes spam rather than ranking. Dates come from Google's own blog posts.

SystemStartedIts jobRanks normal results?
RankBrain2015Links the words in a search to real-world conceptsYes
Neural matching2018Matches the overall idea of a search to the overall idea of a pageYes
BERT2019Reads how words in a sentence change each other's meaningYes, in almost every English search
Passage rankingNot dated by GoogleJudges single sections of a page to see how well it answers the searchYes
MUM2021Understands and writes language, across text and imagesNo, only special uses
SpamBrain2018Finds spam sites and pagesRemoves, doesn't rank
Gemini (custom models)2024Writes AI Overviews and AI Mode answersNo, it uses the ranked results

These sit on top of everything else Google weighs: links, freshness, how reliable a source looks. Google's ranking systems guide lists them all and gives no weights.

RankBrain: matching words to meaning

RankBrain was Google's first deep learning system in Search, launched in 2015, per Google's 2022 post on AI in Search. Deep learning means software that learns patterns from huge amounts of examples instead of following hand-written rules. Google describes RankBrain's job this way:

"It means we can better return relevant content even if it doesn't contain all the exact words used in a search, by understanding the content is related to other words and concepts." (Google Search Central: A guide to Google Search ranking systems)

For an owner, this is why a page about "water heater replacement" can show for "no hot water in my house". The page doesn't need every phrase a customer might type.

Neural matching arrived in 2018. Google says it looks at a whole search or a whole page, not single keywords, to grasp the concept behind each and pair them up. A page that clearly covers one service gives it a clean concept to match. A page that mixes six services blurs it.

BERT: how words change each other

BERT stands for Bidirectional Encoder Representations from Transformers. In plain words, it reads every word in light of the words before and after it. Small words like "for", "to" or "without" can flip a meaning, and BERT keeps them.

When Google launched it in October 2019, it said BERT would help it understand "one in 10 searches in the U.S. in English". By 2022 its role had grown:

"Today, BERT plays a critical role in almost every English query." (Google blog: How AI powers great search results)

That post says BERT both ranks pages and helps find them in the first place.

MUM: powerful, but not for everyday ranking

MUM (Multitask Unified Model) was announced in May 2021. Google called it 1,000 times more powerful than BERT, trained across 75 languages, able to read text and images. Google's current guide says it doesn't rank your site:

"It's not currently used for general ranking in Search but rather for some specific applications such as to improve searches for COVID-19 vaccine information and to improve featured snippet callouts we display." (Google Search Central: A guide to Google Search ranking systems)

There is nothing to "optimize for MUM".

SpamBrain: the AI that removes pages

SpamBrain launched in 2018. Google's webspam report for 2021 says it found nearly six times more spam sites that year than in 2020. It catches things like hacked pages, gibberish and pages made only to game rankings. If SpamBrain flags a page, no other system will rescue it.

Do AI Overviews rank pages differently?

No new ranking layer. AI Overviews run on a custom Gemini model, per Google's May 2024 announcement, and AI Mode launched in March 2025 on a custom version of Gemini 2.0. Both pull their sources from the normal results. Google's generative AI guide says these answers rely on "our core Search ranking systems to retrieve relevant, up-to-date web pages". The AI often runs several related searches for one question (Google calls this query fan-out) and reads the pages those return.

"There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." (Google Search Central: AI features and your website)

So a page that ranks well for the questions behind a search is the page AI features can cite. Local detail: AI Overviews for local businesses.

Can you optimize for RankBrain or BERT?

Not directly. These systems exist to understand people, so the only lever is a page people understand. What that means in practice:

Old habitWhat these systems reward instead
Repeating the exact keyword many timesCovering the topic in natural sentences. See keyword density
One page per keyword variationOne strong page per service. Google warns that making pages for every variation, including fan-out queries, can break its scaled content abuse policy
Odd phrases like "plumber Dallas cheap"The words customers actually say, including small words
"Chunking" text or adding an llms.txt fileGoogle says you can ignore these for its Search
Adding special schema so AI picks your pageGoogle says structured data "isn't required for generative AI search"
Chasing bounce rate or time on pageGoogle uses "aggregated and anonymized interaction data" for relevance, and names no bounce rate or time rule

Practical steps for a service page are in on-page SEO for service pages.

Does Google's AI rank AI-written pages lower?

Not for being AI-written. Google's 2023 guidance on AI content says it rewards quality "however it is produced", but using AI to make content mainly to manipulate rankings breaks its spam policies. Its helpful content guide asks whether you use "extensive automation to produce content on many topics". Full answer: does Google penalize AI content?

Example

A plumber with keyword-stuffed pages

The site has "emergency plumber Austin" 25 times on one page and separate pages for "24 hour plumber Austin" and "after hours plumber Austin". RankBrain and neural matching treat all three as one idea, so the pages compete with each other. Merging them into one emergency page, written the way callers describe the problem, gives Google one clear match.

A dentist told to "optimize for MUM"

An agency quotes $1,500 for a "MUM optimization package". Google says MUM isn't used for general ranking. The money is better spent making each treatment page answer the questions patients ask, with the dentist's name and credentials shown.

A bookkeeper asking about AI Overviews

She wants to show in AI Overviews for "do I need a bookkeeper for my LLC". There is no separate AI checklist. A page that ranks for the related questions (cost, what a bookkeeper does, bookkeeper vs accountant) is what Google's AI can read and cite.

FAQ

Is RankBrain still used?

Yes. It is on Google's current list of active ranking systems, last updated in December 2025.

Is BERT a ranking factor?

Not a factor you can score on. It is a system that helps Google understand searches and pages, used in almost every English search.

Does Google use Gemini to rank websites?

Google hasn't said so. It says custom Gemini models write AI Overviews and AI Mode answers, and that those features use the core ranking systems to find pages.

Do these systems work in French or Spanish?

Google said in 2019 that BERT ranking started in US English and would expand to more languages, and MUM was trained on 75 languages. Google doesn't publish a current language list for each system.

Different for you?

Keep learning

NextYandex source code leak: what it reveals about ranking factorsWhat the 2023 Yandex source code leak showed about how a search engine ranks pages, which leaked factors Google also confirms, and which myths to drop.

Do it yourself

SEO checklist: 10 steps to rank higher with the site you haveAn SEO checklist for an existing website: 10 steps in priority order, the free Google tool for each, and how to tell when a step is done.

Ask us one question

One line is enough. We reply by email.

Sources

  1. Google Search Central: A guide to Google Search ranking systems
  2. Google blog: How AI powers great search results (Feb 2022)
  3. Google blog: Understanding searches better than ever before (Oct 2019)
  4. Google blog: MUM, a new AI milestone for understanding information (May 2021)
  5. Google Search Central blog: Webspam report 2021
  6. Google blog: Generative AI in Search (May 2024)
  7. Google blog: Expanding AI Overviews and introducing AI Mode (Mar 2025)
  8. Google Search Central: Optimizing for generative AI search
  9. Google Search Central: AI features and your website
  10. Google Search: How Search works, ranking results
  11. Google Search Central: Creating helpful, reliable, people-first content
  12. Google Search Central blog: Google Search's guidance about AI-generated content (Feb 2023)