Expansion, Machine Learning & Core Updates

Google’s John Mueller was asked a series of questions around the product reviews update on last Friday’s video hangout. In short, it will expand to other languages and countries, it probably does use machine learning and eventually it might be incorporated into the larger core updates.

As a reminder, we had two product reviews updates in 2021, one in April 2021 and the other in December 2021.

These series of questions came up at the 46:21 mark in this video, here it is if you want to play it, otherwise the transcript is below:

Expanding The Product Reviews Update

John basically said Google does plan to expand the product reviews update beyond just English, as we covered before – but he has no timeline for when that might happen.

Here is what was said:

AUDIENCE: So I have a few questions about product reviews, which is my domain. So I noticed that recently you made some moves in the US to promote product reviews, where there were real testing instead of just comparison. Do you have any time estimates of when we could expect these upgrades to be, like, in the market stack in French or German?

JOHN MUELLER: I don’t know, with most things, it’s something, where we tend not to pre-announce them. So it’s hard to kind of say ahead of time. And for some of these updates, the team moves very fast in kind of, like, rolling things out globally. And for other types of updates, it’s very slow. And sometimes there are also policy and legal reasons kind of that make things a little bit harder. So it’s always very tricky to give an estimate.

Re: the Product Reviews Update expanding to other languages Via @johnmu: We have nothing to pre-announce. For some updates, the team works fast to roll things out to other languages. For others, it’s slower. Policy/legal reasons can also impact things: https://t.co/RRk1Q67bzW pic.twitter.com/KTisK4pWDr

— Glenn Gabe (@glenngabe) January 16, 2022

Machine Learning & Product Reviews Update

Then he was asked if the product reviews update uses human input or machine learning? John basically said no to human input but likely to machine learning. John said this takes an “algorithmic approach” and not a human editing approach, like most of what Google does. He isn’t sure if this specific update uses machine learning, but he said it probably does “to some extent.”

Here is what was said:

AUDIENCE: I’m curious to just learn a little more in general how could Google train, like, an algorithm to understand where there were real testing or not. Is it machine learning, or is it also human reviewers?

JOHN MUELLER: For Search, for these kind of general ranking updates, it’s not something, where we would have human reviewers in the loop, because there’s just so much content out there. We can’t review it all. And a lot of times, we also don’t have a clear yes or no understanding, where someone can go through a website and say, oh, this page is good. This page is bad. It’s just not scalable.

So these are all essentially algorithmic approaches that we take. And we do use a lot of machine learning. Whether we use that for this particular update, I don’t know. My guess is probably also, at least to some extent. But it’s hard to say which exact technology was used for which update.

More: Is Google using machine learning to train algos for the PRU? Via @johnmu: For updates like this, it’s not something where human reviewers would be in the loop. And we do use a lot of machine learning overall, so it’s probably used to some extent: https://t.co/7wx6l8IPw5 pic.twitter.com/0GFLS92Rwv

— Glenn Gabe (@glenngabe) January 16, 2022

Core Updates & Product Reviews Update

Finally, he asked if eventually will the product reviews update be incorporated into the core update like some other algorithms have been in the past. John said maybe, it depends on a lot of things. He said it also depends on how you define core algorithms.

This is what was said:

AUDIENCE: OK, and do you plan to integrate this kind of understanding of a web page into the core Google algorithm?

JOHN MUELLER: I mean, it’s always tricky to define what is a part of the core update– the core algorithm. So I don’t think I have an answer for that. Because usually, what happens is when we see that some algorithm is working really well, then we’ll kind of like just keep using that for a longer period of time. And at some point, you could say, well, it’s a part of the core algorithm, even if it’s not in a file that is called Core Algorithm or something like that.

And more: Will the PRU be baked into Google’s core ranking algo? Via @johnmu: It’s tricky to define the “core algo”. If an algorithm works very well, we might continue using it for a long time. So you could say it’s part of the “core algo” at that point https://t.co/TXPnwJmGny pic.twitter.com/j6AjNpgakA

— Glenn Gabe (@glenngabe) January 16, 2022

Forum discussion at Twitter.