Results from marketing data company Sistrix suggests reasons why Google’s March 2019 Core Algorithm Update feels like a rollback. The data also suggests an explanation of why so many publishers feel like this is a minor update despite event though Google is reported to have said this is one of the biggest updates in years.
March 2019 Core Algorithm Update Feels Like a Rollback
Brett Tabke, founder of WebmasterWorld and PubCon was given advance notice of the update. He was told that this update would be among the biggest in years.
When asked what he thought about the update so far, he indicated that his impression was that it looked like a rollback of previous algorithms. What he meant was that there were many reports of previously penalized websites regaining traffic and SERP positions, as if previous update had been rolled back.
This is what Brett Tabke observed:
“I think we may be seeing a rollback of a few of the last updates.”
Why Google’s Update Feels Like a Rollback
The data that Sistrix was looking at was based on UK winners and losers. This isn’t USA data. Nevertheless, the information gives insight into why the update feels like a rollback.
What’s super interesting is that Sistrix’s data shows that 75% of the winners were previous losers. That means that 75% of the websites that improved in rankings in this update were sites that lost rankings in the previous updates of 2018.
Because so many previous losers appear to be winning, it gives the impression that this update is a rollback. I don’t believe Google rolls back updates. What I have been told in the past by search engineers is that Google fine tunes their algorithm.
I believe that in a major update they improve how sites are ranked. I believe this is the case, with the side effect of positively affecting sites that previously lost rankings.
So although this may look like a rollback, it’s highly unlikely. Whatever changes were made appear like a rollback.
If 75% of the winners consist of losers from previous updates, then Brett Tabke’s observation is correct. The March 2019 Google update looks like a rollback. But it likely is not a rollback.
Anecdotal evidence and actual data from Sistrix suggests that up to 75% of the sites that improved rankings were sites that lost rankings in previous updates. This gives the update the impression of a rollback.
Why the March 2019 Update Feels Minor
Sistrix observed that their data indicated that sites that were losers tended to have lost long tail positions and not huge positions. This means that the amount of traffic associated with the loss of rankings was relatively softer than if the loss had been due to a loss of more important rankings that represented higher amounts of traffic.
This coincides with the anecdotal observations that this doesn’t “feel” like a major update.
More importantly, 70% of the sites that were losers were sites that were previously hit by previous algorithms. If this data is correct and extrapolates to other countries, that means that most of the damage was sustained by sites that had already lost ranking positions. That may be what’s contributing to the sense that this update isn’t that big.
This is what Sistrix reported on what their data suggested:
“The top losers in our data aren’t as strongly hit (in terms of percentage) as the winners. … among the losers there are many domains (70%) that were affected by previous core updates.”
What is the March 2019 Update?
There have been many advances in information retrieval technology in the past year, it’s difficult to point at one and say this is what the update is about. The update could be one addition or more than one.
For example, Google recently published a research paper titled, Non-delusional Q-learning and value-iteration.
Fixing Bias in Reinforcement Learning
This research paper notes that there can be a bias in “reinforcement learning,” a fundamental aspect of machine learning. (More information about Q-Learning here)
The Google research paper states:
“We identify a fundamental source of error in Q-learning and other forms of dynamic programming with function approximation. Delusional bias arises when the approximation architecture limits the class of expressible greedy policies. …inconsistent or even conflicting Q-value estimates can result, leading to pathological behavior such as over/under-estimation, instability and even divergence.”
I am not saying that Google has introduced a more accurate version of machine learning, one that reduces or eliminates a built-in error or bias. I am simply citing one research paper as an example of the many research papers published by Google that may give a clue to what is going on.
A New Relevance Signal for Ranking
Another research paper introduces a new way of ranking web pages. It’s called, Learning Groupwise Scoring Functions Using Deep Neural Networks.
Here is what the research paper proposes:
Consider a search scenario where a user is searching for a name of a musical artist. If all the results returned by the query (e.g., calvin harris) are recent, the user may be interested in the latest news or tour information.
If, on the other hand, most of the query results are older (e.g., frank sinatra), it is more likely that the user wants to learn about artist discography or biography. Thus, the relevance of each document depends on the distribution of the whole list. Second, user interaction with search results shows strong comparison patterns.
What that means is that the age of the web pages that are relevant to a search query can sometimes give context to what a user may want. Secondly, the history of user preferences show in the search engine results pages (SERPs) can also help reinforce the age of the document relevance clue.
Ranking Algorithms Evolve
I am not saying the above research papers are behind the recent Google Update. I am showing what two recent advances are to illustrate what the state of the art is today. Too many people still believe in 200 ranking factors and that updates are a matter of “targeting” low quality sites. That’s an incorrect way to understand Google updates.
As you can see from the above two examples, Google’s algorithm is far more complex than 200 ranking signals.
Google Explanation of the March 2019 Update
I expect that Google may at some point in the future explain what was introduced. If the change was related to something similar to the above kinds of algorithm, then Google might seek to obfuscate what the algorithm is and speak of it in terms of what the results of the algorithm are.
It’s one thing to make observations about what kinds of sites were affected. Drawing conclusions from those observations is frequently a poor approach. The so-called Medic Update was named that because of the observation that many medical related sites were affected.
Yet the fact was that Google was not targeting medical sites. Unfortunately, this poor analysis resulted in confusion. The false idea that Google was targeting medical sites persists. This highlights the danger in reaching conclusions based on limited observations. It’s too bad that the Medic Update was so named. It’s a source of great confusion.
I believe it’s best to wait for information from Google, find parallels with patents and research, and then begin to unravel what happened. That’s the most prudent way to understand Google updates.
This data from Sistrix goes a long way to explaining why Google’s update feels like a rollback to many publishers. The data is not evidence of a rollback. The data only confirms why it feels like a rollback. I do not believe this update is a rollback.
Read More about Google’s March 2019 Broad Core Update
March 2019 Core Update: What’s Changed? Early Insights & Reaction
Insights from Brett Tabke, SEMRush and from around the web
Google Update Florida 2: March 2019 Core Update Is a Big One
The original announcement, with details about how a Googler leaked that this update would be important.
Images by Shutterstock, Modified by Author
Screenshots by Author, Modified by Author