RankBrain Explained: How Google Started Using AI in Search

TL;DR

• RankBrain was the first step. Google says RankBrain, launched in 2015, was “the first deep learning system deployed in Search”. It helps Google connect words to concepts.
• Then came neural matching and BERT. Neural matching (2018) matches concepts in queries and pages. BERT (2019) reads how word order and small words change meaning.
• MUM added generation. Announced in 2021, MUM can both understand and generate language, but Google says it is not used for general ranking.
• AI Overviews and AI Mode changed the page itself. Since 2024 and 2025, Google often writes an answer above or instead of the classic list of links.
• The trade-off is clicks. Searchers get answers faster. Publishers get fewer visits per search, according to independent research.

Contents

1. Why Google needed AI in the first place
2. A timeline of AI in Google Search
3. RankBrain, neural matching and BERT in plain language
4. From understanding to answering: MUM, AI Overviews and AI Mode
5. What changed for searchers
6. What changed for publishers
7. FAQ

For most of its history, Google Search matched the words you typed to the words on a page. That works until people type the way they think: vaguely, with half-remembered phrases. RankBrain was Google’s first public answer to that problem. It was not a secret master algorithm. It was the start of a decade in which machine learning moved from helping rank links to writing the answer itself.

Why Google needed AI in the first place

Google describes its ranking as a set of “automated ranking systems that look at many factors and signals about hundreds of billions of web pages and other content in our Search index”, according to its guide to Google Search ranking systems. RankBrain, BERT and the others are parts of that larger machine, not the whole of it.

The core difficulty is language. A query like “what’s the title of the consumer at the highest level of a food chain” lacks the phrase a biologist would use. Google used this example in a 2022 post by Pandu Nayak, explaining that its systems needed to connect the question to the concept of an “apex predator”. Keyword matching alone cannot do that. A system that has learned how words relate to ideas can.

A timeline of AI in Google Search

RankBrain became public in October 2015, when Bloomberg reported that Google was handing a growing part of web search to AI, with the summary line “In tests, system beats company’s experts at page selection”.

RankBrain, neural matching and BERT in plain language

These three systems are still listed as active in Google’s ranking systems guide. They overlap, but each solves a slightly different problem.

The examples come from Google’s 2022 explainer. They are unpolished on purpose. People do not search in tidy sentences.

BERT deserves a note of its own. When Google announced it in October 2019, it said BERT “will help Search better understand one in 10 searches in the U.S. in English”. Its showcase example was “2019 brazil traveler to usa need a visa”, where the word “to” decides who is travelling where. Older systems tended to ignore such words. BERT does not.

From understanding to answering: MUM, AI Overviews and AI Mode

Everything above is about understanding: reading the query and the page well enough to rank links. The next stage was about generating text.

MUM, announced in May 2021, was described by Google as “1,000 times more powerful than BERT”, trained across “75 different languages” and “multimodal”, meaning it handles text and images together. Its role in ranking is limited. The ranking systems guide says MUM is “not currently used for general ranking in Search but rather for some specific applications”, such as improving featured snippet callouts.

The visible shift came with generative answers. In May 2024, Google announced that “AI Overviews will begin rolling out to everyone in the U.S., with more countries coming soon”. In March 2025, it introduced AI Mode as “a new generative AI experiment in Google Search”, first in Labs, and said AI Overviews were “now used by more than a billion people”. On 20 May 2025, Google said AI Mode “will start to roll out for everyone in the U.S. today”.

AI Mode works differently from a classic results page. Google says it uses a “query fan-out” technique, running many related searches across subtopics and sources and combining the results into one response. The search engine now searches on your behalf and writes up what it found. We look at how those sources get picked in how AI search chooses sources.

What changed for searchers

For people searching, the change is mostly convenience. Vague or long questions now get reasonable results more often, because systems like RankBrain and BERT read meaning rather than keywords. With AI Overviews and AI Mode, many questions get a written answer without opening any page.

The cost is less visible. A summary is one interpretation of several sources, and checking where a claim came from takes an extra step most people skip. Google itself notes in its documentation on AI features that AI Overviews and AI Mode may use different models and techniques, so the same question can produce different links in each.

What changed for publishers

The first AI wave (RankBrain, neural matching, BERT) mostly favoured pages that answer a question clearly over pages stuffed with exact keywords.

The second wave is harder. When the answer appears on Google’s page, fewer people need to click. A Pew Research Center study of 900 U.S. adults’ browsing in March 2025 found that people clicked far less often when an AI summary appeared:

Google’s position, stated in its AI features documentation, is that “there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”, and that clicks from AI Overviews are “higher quality”. Google does not publish data for that second claim on the page. Publishers are asked to take it on trust. Generous.

The practical result: being a cited source matters more, and raw ranking position matters a little less, than it did when RankBrain arrived.

FAQ

About the Author

Vyara Stefcheva

Vyara Stefcheva is the founder and editor of Egorithms and an independent marketing expert. She follows how technology, AI and algorithms change the way people find, choose and buy, and writes about what that means for businesses and for the people who use them.

She has spent her career building marketing and business strategies for more than 300 brands across three continents. She built and exited a marketing agency and now works independently. She has taught more than 20,000 people through lectures, courses and corporate training, and speaks at conferences about AI and the future of the internet.

On her YouTube channel she tests new AI and marketing tools as they launch and gives each an independent verdict: adopt, watch or ignore.

See author's posts