Machine Learning and SEO (plus how it may feed into the Helpful Content Updates)

Well , it's a branch of artificial intelligence that aims to imitate the way humans learn from experience and how we refine our knowledge as we gain experience . In traditional programming , developers have to manually provide specific instructions to a computer to cover all scenarios of a problem they're trying to solve .

So if the scenario or problem changes , then the developer needs to go and manually update the code . It works great if you've got like a very specific control circumstance , but it's hard to scale and manage as scenarios and data become more complex .

In machine learning we reverse the process , so instead we feed data to the computer and then let it come up with the solution itself without having to explicitly instruct it how to do so . Let's look at a simple example that I hope will make it a bit more obvious what I mean .

So imagine we want to write some code that we can give an image of a piece of clothing and it can then tell us if that piece of clothing is a shoe . In traditional programming , we'd have to manually specify all the possible types of shoe that could possibly exist , from trainers to formal shoes , high heels , walking boots , etc .

And what the exact specifications and attributes of all those different types of shoe were in detail , such as lengths , colours , sizes . It's an almost impossible task , and even if you achieved it , you'd be constantly having to update the code as new shoes were released all the time by countless manufacturers around the world . So it's like an impossible task really .

However , using machine learning , we can simplify this task hugely . So instead of having to go into all that detail , all we need to do is gather a set of training data to train a machine learning model with .

So in this case , we'd get a set of as many example images of shoes we can find maybe hundreds or a couple of thousands of different images of shoes and we feed them into the model and we say to the model these are images of shoes , and the machine learning model itself would start to recognise the characteristics that embody what a shoe is , and the more

examples we show it , the better it will be , just like with humans , the more examples we see of something , the better we get at recognising the patterns that embody those examples . So , once trained like this , we can then show new images to the model and say , is this image a shoe ?

And the model can then use its pattern recognition that is built up from the training data to identify if the image does or does not contain a shoe .

So this means that we don't need to specify all the different types of shoes ever produced , just enough examples for it to learn what a shoe should look like , just like we , as humans , can recognise shoes we've never seen before as being shoes , because We've learned over time to recognize those that make up what is a shoe .

Okay , so that simple example I've just shared should Now give you just a brief overview of how machine learning essentially works . That is , we give a set of data to a machine learning model to train it in a certain task , and then it can Perform that task and even learn as it goes along with that , as it Executes that task .

Now , how do we know that Google uses machine learning ? Well , because there's lots of information about it , and they've been researching artificial intelligence and machine learning since before the you know , the 2010 .

They founded Google Brain in 2011 , which rolled into Google and then has become Google DeepMind , where they do lots of advanced machine learning for all sorts of applications .

They've released lots of software open source software , such as TensorFlow , which allows neural networks to be used by the public , and it's used within a countless AI projects used TensorFlow , so we know that it's like a real core part of their research and development and the Products that they've produced with it .

Since we also know of several machine learning AI algorithms the Google use within their search , such as RankBit and one . I'm not going to go into detail what they do now , but they are declared machine learning algorithms that Google talk about , that they that they use to rank and evaluate sites .

There's other machine learning Examples , such as if you do the , if you ever seen where Google says , hum part of a tune and they'll find the Record , the single . That that is related to that again is machine learning . That's where the machine can recognize , for example , songs , which song you're trying to hum .

And also if you use camera or image search , where it can identify objects , places , from images . So we know that , that it's embedded deep within all of their algorithms when it comes to search .

So going back to the original example of how machine learning is based , on feeding a set of training data Into a machine learning model and then allowing that model to learn from it and then giving it new examples and it that model can then score whether the thing that's been presented now matches something in that model .

Now , this is where this gets a little bit more contentious . There's because this is not something that Google have explicitly stated that they do . That I'm going to discuss now . But looking at the things that they are capable of doing and looking at the data sets they're building , and this is something that we could potentially say happens .

Now this is going back to the helpful content update recently that has just confused so many people , where you know Google are saying if we find your site to not be helpful , then we're going to demote it basically , we're going to dump . And how you come up in the research results ?

Rather frustratingly , for most people have been hit by the helpful content update . The only advice that Google really give is to create content for people first and not for search engines . They do give some examples of what they mean by people first , but it's all very vague .

It's not a case of going and saying you must have an author bio or you must format your content like this , or you must lay out your pages in a certain way . They don't give that level of detail , they just say it's got to be helpful for people . But how to Google define what's helpful for people ?

I don't think they actually have a specific definition that they work to . That is encoded by a developer into a system . I think there is a machine learning model that has been trained to identify what types of site , what types of content , what types of layout , what types of page elements are helpful for people . Now where does this training data come from ?

Well , if we look at Google quality raters , that is where I think this training data comes from . If you've not heard of Google quality raters , this is a set of I think they say it's about 16,000 people worldwide whose job essentially is to rate Google search results and rate the sites that Google is surfacing .

So Google's used quality raters for quite a long time . I think in the early days it was a more simple , giving them sort of A-B tests of different search results .

So the people would then just rate which search results of two options were the best , and that's how Google would train its and test and train its algorithms in terms of making sure that they were giving the right answers and not coming up with spurious results . It was a way of testing new updates to their algorithm .

But if you go and look at the Google quality rater guidelines and I'll put a link to that in the show notes most of what people are being asked to do with these raters is to actually rate the content on websites on a whole variety of factors , but the way the rater is from less helpful to more helpful .

So there's this big data set that Google have been creating , a training set that allows that . It's all about what is helpful and what is not . So they've now got a huge set of different web pages and different sites graded by real people as to whether they're helpful or not helpful .

This is the basis , I believe , of a model where they can then decide to run pages through this model and then get a score as to how helpful these pages are , and I think that's possibly the basis of what the helpful content update is .

There's no doubt that the helpful content updates are machine learning powered , because Google themselves mentioned it in their Google Search Central blog around the first I think it was the first Google helpful content update in August 2022 . And their exact words are this classifier process is entirely automated using machine learning model .

So we know that the helpful content update is a machine learning powered update , so which means there must be a training set of data that that machine learning model is built upon , and by far and away , I believe the most obvious training set is that that's created by the Google quality raters .

So obviously now , if you want to understand how to make the most helpful content , then it's really understanding those Google quality rater guidelines that will help us evaluate the content that we're producing and how we should be producing content to try and be classified as helpful rather than unhelpful , even if it's not the helpful content update .

Google is definitely only collecting that data for a reason that it's got , and that reason is gonna be based around ranking somewhere . So even if it doesn't solve a helpful content update , it's gonna solve some other problem for you .

So those Google quality rater guidelines and understanding of them is really , really key , and that's what we'll look at in future episodes . So hopefully now I've given you a good overview of what machine learning is at its basic level and how it's used by Google . Thanks for listening . I really appreciate it . Please subscribe and share . It really helps .

Seo is not that hard . It's brought to you by keywordspeopleusecom , the solution to finding the questions people ask online . See where thousands of people use this every day . Try it today for free at keywordspeopleusecom . If you want to get in touch or have any questions , I'd love to hear from you . I'm at channel five on Twitter .

You can email me at podcast at keywordspeopleusecom . Bye for now and see you in the next episode of SEO . It's not that hard .