On Monday, Bloomberg published a story that reached every SEO forum within hours. Google confirmed that it has been using a machine learning system called RankBrain to help process search results, and one of its research scientists, Greg Corrado, stated that it has become the third most important signal in ranking.
This is the first time "machine learning" and "SEO" have appeared in the same sentence outside a research paper. We read the news on Nusa Ceningan, between boat trips, three months after the dive centre opened. The announcement can be read in two ways: as a major development or as a piece of Google public relations.
Our assessment is that it is both, and that its significance is greater than it appears this week.
What does RankBrain actually do?
RankBrain helps Google interpret searches, particularly those it has never seen before. Its role is narrower than the headlines suggest. According to Google, about 15 percent of the searches it receives each day are entirely new: unusual phrasings, long and wandering questions, typos, local expressions. Until now, such a query was matched mainly on the words it contained. A page that did not use those words had little chance of appearing.
Based on Google's statements, RankBrain converts words and phrases into mathematical representations, called vectors, so the system can infer that an unfamiliar query means roughly the same thing as a familiar one. It learns these connections from patterns in very large amounts of text, rather than from a list written by hand at Google. The system has already been running for several months, and nobody noticed.
That last detail is informative. For now, the public facts fit in a few paragraphs, and the RankBrain entry on Wikipedia is a useful reference to follow as more information becomes available. Claims of more detailed knowledge should be treated with caution.
Nobody outside Google can observe what the system does, which is already frustrating many SEO practitioners. It cannot be tested or tuned for in the way keyword density has been tuned for years. There is no checklist.
Why is meaning starting to matter more than exact keywords?
Because the system aims to understand what a query means, not only which words it contains. For a small business, this is largely good news, even if the effects may take time to appear.
Consider how people actually search for a dive trip. Some type "diving Nusa Ceningan." Others type "where can I see manta rays near Bali that isn't too crowded," or "beginner scuba course small island Indonesia." The former approach was to build a separate page filled with each phrasing. If RankBrain works as Google describes, one clear page that genuinely answers these questions has a real chance of ranking for all of them.
Keyword stuffing has been declining since Panda in 2011 and Penguin in 2012. Hummingbird, two years ago, moved Google further toward reading complete questions. RankBrain does not eliminate stuffing outright, but it makes the approach even less useful. If the system understands what a page is about, repeating a phrase fifteen times does not help; it only makes the page harder to read.
What kind of page performs best?
A clear, specific page. The practical recommendations are straightforward:
- Dedicate each page to a single topic.
- State the subject plainly in the title and the first paragraph.
- Answer the questions real customers ask, in the words they use.
- Include the specifics: the location, the price, the dates and what is included.
Vague pages about "our passion for excellence" give a machine nothing to work with, and they offer a human reader very little as well.
Where is this trend heading?
RankBrain appears to be the beginning of a longer development rather than a one-time update that can be waited out.
Google has been acquiring and building machine learning capabilities for some time. It acquired DeepMind last year. Its voice search is becoming remarkably accurate at understanding indistinct questions. RankBrain is the first time Google has publicly stated that a learning system makes ranking decisions, and significant ones. The trend points toward a growing share of searches being handled by systems of this kind each year, and toward these systems becoming much better at reading full sentences the way a person does.
If this trend continues, several consequences follow:
- Pages that answer real questions clearly will continue to perform well, and likely better than today.
- Specific facts, such as an address, hours, prices and services, will become more important, because a machine can identify and rely on them.
- Tricks will become harder to apply and less rewarding each year.
The pace of this change is uncertain; the direction is clear.
How should a small business respond?
By writing for the reader. A system that cannot be observed cannot be gamed. This is the point we emphasize most with small business owners. There is strong demand for the latest technique, and there is always someone offering one. After a story like this, a wave of "RankBrain optimization" offers can be expected. The appeal is understandable, since it suggests that a secret exists.
In practice, there is very little secret. RankBrain cannot be examined from the outside. What a business can observe is its customers: the questions they ask, the words they use and what they need to know before they call. Writing that down clearly, on a page that loads quickly on a phone, covers most of what matters. For businesses unsure whether their site meets this standard, an objective SEO audit is a better investment than any technique.
This recommendation may seem less technical than the services being promoted this week. It remains, in our assessment, the most reliable approach.