Google is now using machine learning and AI to deliver its search results. But what does this mean for SEO?
Learn how to train the AI machines with Jess Peck, Genie Jones, Myriam Jessier and Marco Giordano.
Watch On-demand
Listen to the Podcast
Transcript
Dixon Jones
Hi everyone, and welcome to another episode of Old Guard New Blood, Majestic’s monthly or periodic podcast. And this is issue 34. And today we’re talking about how to train the AI machines. So Googlers for many, many years being into machine learning and a huge number of things have happened and I’ve got a brilliant cast with me, cast with me as normal. And I’m gonna ask them all to introduce themselves in a second. We’ve already got one question in the chat, which we thought we’d start off with because Windy City Parrot seemed to come up with the questions, so we might bring that up straight away. But before we do, why don’t we ask everybody to introduce themselves? Jess, why don’t you start, Tell us about who you are and where do you come from.
Jess Peck
All right, thank you Dixon. I’m Jess. I am a recovering technical seo and I work as a machine learning operations engineer at Local SEO guide. And I live in Boston, but I have an English accent because I grew up in England.
Dixon Jones
And for you, Halloween goes for all of November
Jess Peck
Until Christmas, maybe beyond.
Dixon Jones
Excellent, excellent. Thanks Myriam, who are you and where do you come from?
Myriam Jessier
Oh, that’s a complicated one. So I’m originally from France, but I grew up in Hawaii and currently I’m located in Italy.
Dixon Jones
That sounds very, very cool. I would grown up in Hawaii. Genie, how about you?
Genie Jones
Yeah, not as cool as everyone else. I born in Luton, nice big up. And now I live in London so I’m the knowledge graph manager for in links. And yeah,
Dixon Jones
Thanks for coming along. And Marco, how about you? How are, where are you and where do you come from?
Marco Giordano
Hello. So of course my name is Marco. I come from Sicily, South Italy now I live in Zurich, so German, Switzerland. And of course my focus is on content for B2C as well. B2B sometimes and analytics.
Dixon Jones
Excellent. Thanks a lot for coming along guys. And just before we dive into the sessions itself, I just wanna bring in our producer David. David, have I missed anything this morning, this afternoon, this evening? Depending on where you are in the world
David Bain
Indeed. Not that I can think of. I just wanna say we get plenty of listeners on Spotify, on Apple Podcasts and Google Podcasts and it would be good just to get a few more people watching along live and actually interacting live because it’s a great opportunity to ask questions about the particular topic that we’re covering. So if that sounds like you, if you’re listening to this on some kind of podcast platform, sign up at majestic.com/webinars next time and then hopefully we’ll see you live and with us for the next episode.
Dixon Jones
So before we get into everything and start with Mitch Rezman’s question, what I like to do at the top of every show is for anybody that doesn’t have time to go through the 45 minutes or so of the podcast or just wants to have a quick takeaway tip, if you are trying to influence the AI within Google in any way, shape or form, have you got some kind of tip that you can give people that they can take away with and use? And Marco, why don’t I start with you for an idea?
Marco Giordano
Yeah, actually one of the best tips that is underrated in my opinion is checking your sentences. What I mean is that machines are able to understand no how your sentence is structured. This is a subject, this is an object, this is a predicated. So if you have a page let’s say about gaming laptops, a great idea would be to use gaming laptops as a subject or in central positions rather than barely mentioning it at the end of a sentence. So it’s not just about keywords but also about the position of what you’re telling me, especially for machines.
Dixon Jones
That’s a brilliant idea. We might come back to some of the tools that you might use to help you do that a little later on. That’s a good one. Jeanie, what about you thought for you?
Genie Jones
So I guess my biggest tip will be using that same as tag in schema markup. So big one for Entity SEO and marking up entities to tell Google what you’re talking about always is a brilliant idea And I think there’s so much you can do with that in terms of machine learning and it, it’s like the ultimate influencer in a way because it’s just speaking Google’s language,
Dixon Jones
Telling the machines, telling the machine what it should be, understanding
Genie Jones
What it should be understanding. Look at these bits, not these bits. I mean brilliant. You can be as subtle and as obvious as you like with machine learning I think in a lot of times and brilliant to different effects.
Dixon Jones
Thank you. Myriam, what about you Got a thought for you?
Myriam Jessier
Are you asking me? I’m sorry I didn’t, Okay. Yeah, so I do have a tip but it’s an unusual one. Be me worthy because humans function like this, right?
Dixon Jones
Be sorry, be what? Worthy,
Myriam Jessier
Be me worthy, dominate Reddit with whatever you want. And that will send a very, very strong message. One of the reasons I know that Betty White was older than sliced bread is because Reddit really loved this.
Dixon Jones
That means I gotta go and do Reddit and I hate Reddit cuz they hate me. Okay, brilliant tip. Jess, what about you?
Jess Peck
Make sure that you check your content for being understandable on a machine basis and being understandable when you use different, make sure it’s accessible to machines as well as people. Make sure that your content is clear and make sure that that feeds into everything else because if you don’t have that fundamental right, nothing else will matter,
Dixon Jones
Right? Okay. And that’s kind of similar but different to Marco but it kind of makes the point that badly written content is not just about writing about the wrong things, it’s about not writing in a way that a, that machines can understand because machines are like little kids when they’re trying to understand this code, they, it’s like simple sentences a machine will understand complicated senses, not so much. So there’s tools out there those would you say Grammarly and Hemingway, they’re two sort of structured tools and things. Is that helpful for machine machine learning, helping the machines learn? Any other tools that you use?
Jess Peck
I think those are some good tools both for that and for understanding the ways that machines can fall short in understanding like sentence structure. I’m sure everyone’s used Grammarly and had it flag a word that makes sense and is the word you want to use and it tries to replace it with something different. And so being able to see the kind of gaps in your content where you are making sense to a person because people can make better connections the machines can, and using things like Grammarly can help with that. But also if you use a tool like OpenAI’s like GPT-3, if you put your content in there and ask it to summarize that, ask it to summarize that content, you can see what a machine would take away. A very sophisticated natural language machine could take away from the same content that you are writing and see if it aligns with what you actually want.
Dixon Jones
So for the audience that don’t know GPT-3, oh well Myriam, when you talk explain what GPT-3 is as well. Yeah,
Myriam Jessier
Well before I do that, I just wanted to say that you don’t necessarily need an expensive tool because right now Google Docs does this for you as well. You can just copy paste your content in there and click on the top left saying Hey, give me the summary and if the summary is disappointing, something’s off.
Dixon Jones
That’s actually, that’s a good way to do it as well. Absolutely brilliant. And so for the folks out there that haven’t heard of GPT-3 or GPT-4, it’s a system where it just writes, it automatically generates text out of a few ideas really, or in this case reverses that and takes a lot of text and scintillates it down to a few ideas. Okay, so let’s talk on Mitch Rezman dived in before we started. None of us know who Rich Rezman is. So Hi Rich, we looked you up and we windy city parrot.com. Big shout out to you. I wasn’t allowed to actually see it cuz he wanted to check to see if I was a human. But anyway, I guess you miss rank Brain and hunting bird updates and we smiled at that and thought that’s memory lane. So perhaps it’s kind of where AI for most of us all started really, isn’t it? The machine learning. So does somebody wanna summarize what RankBrain and Hummingbird did? Who wants to take that one on?
Myriam Jessier
I just wanted to say that it terrorized SEOs back in the day. Every single new thing that Google brings about.
Dixon Jones
That’s true. That’s true. RankBrain, Hummingbird, anyone wanna summarize? Do you wanna me to go in there? Go on Jess.
Jess Peck
So Rank Brain and Hummingbird are pretty early implementations of machine learning in machine learning in such the Hummingbird was a very early language NLP kind of idea,
Dixon Jones
But the moves, but the move worlds moved on, right?
Jess Peck
Yeah. But I think the fundamental, what they’re trying to do is the same, which is Google wants to rely on ML to not have to rely on the stuff that SEOs can manipulate as much. Yeah,
Myriam Jessier
This is a big thing. I remember when they rolled out, people were like, what can I do to optimize for this? And Google kept saying, there’s nothing that you can do. Yes. And here we are today talking about it.
Dixon Jones
Yeah, yeah. Well I mean Hummingbird was even before they bought Meta web, so before they bought Freebase, which is that their initial knowledge graph was pretty much they bought freebase and built on it. So it was quite early. But now of course I think a lot of people see the machine learning all about training Google’s knowledge graph and training the underlying entities and the connections between the ideas and the entities. So mean, does anyone wanna jump in and say in simple terms how machine learning works and how mean, for example, how does Google recognize difference between a cat and a dog? Just for those that, how would you go go about that? I looked something up before I jumped on so that I had an answer. If none of you could run straight to mind.
Jess Peck
Can I take this one too? I feel like I’m dominating early on. So machine learning is the basically can you make it programs without a computer being explicitly programmed? So traditional programming you say print this, it prints this machine learning is trying to get computers to get to print this without telling it to. And so the things you mostly need to know about machine learning are there’s supervised and unsupervised machine learning. Supervised is where the data gets labeled. So that’s a lot of the machine learning that traditionally happened was you get a bunch of photos of dogs and you label them all dogs and you feed it into an algorithm and the algorithm learns this is a dog and then there’s classification, which is and which is and prediction and things like that. And then unsupervised machine learning is where the program learns without the labeling. So you put data in and it clusters stuff together and you get those clusters out and that’s kind of become more the vogue stuff. Like stable diffusion is a lot more of unsupervised because it’s more about learning patterns and then replicating those patterns rather than learning with the labels.
Dixon Jones
So Myriam, just dive in
Myriam Jessier
I’d love to dive in because one of the things that I do is train people. So I didn’t say that at the start of the podcast, but I’m an SEO trainer and I get asked to explain the nuances between machine learning and artificial intelligence quite a lot. And what I like with what Jess is bringing about is let’s take a step back and consider machines from a different standpoint. I have a wiener dog and my wiener dog is a little machine. I got her when she was a very young puppy and we did some supervised training and she still managed to learn what I would consider the wrong things. So lemme get on with that. We decided to train her to pee outside and what she learned was every time I bend over I get a treat. So 15 fake peas later we had to go down together to one at the front with a treat and one at the back checking if this is truly happening because we were running out of treats. This is literally what we want to discuss today. And when it comes to unsupervised, while the dog quickly learned, because I travel a lot, every time the luggage comes out it means that we are on the move. So now she sits in the luggage saying, do not forget me, I’m coming with you. So while these things are super cute, you can see the parallel that we have going on with machines. This is what we are trying to get to.
Dixon Jones
And Genie, you are involved in effectively the supervisor supervising of the learning in Knowledge Graph. So what kind of things do you have to do? What happens when the machine goes wrong? How do you fix the machine?
Genie Jones
Oh well okay. Fixing the machine is a whole different thing. I was gonna to go back onto what we were just talking about. I suppose what I was thinking is what was saying is how human I have to be to fix the machine. How human I have to empathize, I have to empathize with the machine because it’s trying to see things like a human word. And often, I mean I wrote a blog post about this, how basically what we’re trying to do within the InLinks NLP and the knowledge graph and stuff is train it to be as human as possible. Which is obviously the game, aiming the game for a lot of machine learning. So there’s a thing called schema and I’m not talking about tech SEO schema, I’m talking about actually how you as a human view the world. And the thing that I’ve really learned about machine learning is how similar, we’re just trying to copy and paste comprehension, almost like art, the very basics of how we perceive the world. So when I’m trying to do fix the knowledge graph, if things go wrong, I have to look at context. So if it’s getting confused between cat and dog, I have to look at why the context around cats is maybe overlapping with dog. And in a way often it does because there’s the context that they, they’re both pets or they’re both these things.
Dixon Jones
Yeah, they’ve both got four feet.
Genie Jones
Yeah, yeah, exactly. And it’s not so much even that it might be in one piece of text, cat and dog are the same vague entity because they fall under pet. And that’s the complicated thing that you need to be really patient with in machine learning because it’s something that actually, until you are about five years old anyway, you’re not gonna understand the difference. So I guess there is a sort of element of going this is never gonna be perfect because actually understanding this in a human context is not even that perfect all the time. So I guess that’s kind of how I’m viewing this. It’s just how human it is.
Dixon Jones
Marco? Well Marco, do you wanna come in with anything?
Marco Giordano
Yeah, I want to say one thing that we often forget and it’s about what you don’t mention because we are used to adding entities or adding keywords, just adding stuff, but never about removing stuff. No. Much like search intent. When a page is too long, maybe the intent is not to buy. Could be. It’s possible if it is a product page, there are some elements that hint to machines. This is a product page most likely the same goes for, Yeah, I mean if you mention some words, I don’t know, bull those or tire, they are not related to cats, to the animals. So a machine is able to, okay then this is not an animal is it? So that’s what they just wanted to point out. Yeah,
Dixon Jones
Yeah. Excellent. Myriam, were you dying to jump in with something there?
Myriam Jessier
Yes, one of the comments Jack was saying that basically we are describing the premise of the matrix and terminate Terminator franchises. And I just want to say that Jack, if I remember correctly, the first matrix movie, the machines wanted to fix you and all of us.
Dixon Jones
Well this is always the double edged sword with trading the machines, isn’t it? They might learn but I think certainly the matrix or the flickers in the matrix is kind of an interesting thing cause it’s where knowledge graphs and where machine learning goes wrong in some of the areas and where before we came on one of Myriam, were talking about the fact that machine learning can very easily decide that all monarchs are kings for example, when clearly that’s a lie. And so bias gets into the system very dramatically, doesn’t it? If are not very careful with the machine learning.
Myriam Jessier
And one of the things that we discussed early on with Jess when we were talking about this podcast is quickly a chat bot made by Microsoft can turn into a Nazi took list and 24 hours. And we also face certain problems rather often we see where the machines fail and we either can choose to take it for granted and say it is what it is and it’s sad. Or we can simply say, hey, it’s our responsibility first and foremost as SEOs to try and fix this. Or as human beings to say, hey we disagree with this, which ties into an old school thing. So I don’t know if you remember Google bombings, not everyone does. So back in the day you could say French victories and humans gamed Google. So he would recommend, Did you mean French defeats? Yeah, and it’s patently false. I have to set the record straight. The French army is known to have won a lot of things. I know Jess, I know we can disagree about this but I mean we
Dixon Jones
Only remember, remember Waterloo and that’s it,
Myriam Jessier
But truly it still creeps in and this is why I’m here today. I really don’t like the fact that we have situations where, for example Lidia Infante was sharing on Twitter saying hey, I type in SEO expert and in terms of entities like this is what pops up. There’s like this answer box that’s listing all this and there’s all these men, no women to be seen, no non-binary folks to be seen. Apparently an SEO expert is just a dude and we got to talking about how would you fix that. So for example, if you disagree with that list, how would you do it? And I can pass it on because this is a really nice discussion. So Marco, how would you start?
Marco Giordano
Okay, so first of all, I mean let’s start with the basics. Writing the actual article. So first step, having some material, doing actually doing it. Second of course the usual stuff. So building back links, of course you need some links someway. So posting on social media, having some social buts to give signals even though they’re not direct ranking factors, you need something to show Google that you have some sort of consensus even though we don’t know for sure, I would do it because it’s free. There is no risk ensuring something to social media. Once I have that, of course I would probably complete also, no, this can also be done before whenever read the article, like implement some schema, have details about some of these people or link to their websites like to some sources where I can say for instance, Lily Ray is an SEO consultant based in this, she did this and that maybe not boring like that more juicy, more interesting.
But the details will be there with some numbers, some awards, some external sources to feed Google and say look, these are other links, other websites, other domains where you can get this information. So one article of course would not be enough. It’s not like you pick a fresh domain and you rank tomorrow. So probably I would just do it on a website that is not super aged but relevant to the topic. So relevant to seo, specifically to SEO services or recommendations. I mean it’s not very hard in my opinion. The only objection that people can do is that, but if the consensus for Google is white dudes or Americans, because usually it’s not even white, it’s American or English or sometimes Indian, then how would you convince Google? So this is a mouse objection and I would say that you do it by creating buzz and by having even more in your opinion, even more than one domain, even though they are puppets for your main website because it works. I know it’s not recommended, but if you have three domains, you put the same people, the same amenities, Google at some point will tell, look, this gala has to be reliable.
Myriam Jessier
You are describing algorithmic propaganda.
Marco Giordano
Yeah,
Dixon Jones
This is where I think the knowledge graphs are very, they mirror society so very, very well. I mean it’s not as if propaganda hasn’t been a massive great big thing, not just in recent elections from the Brexits and the US elections and these things, but that propaganda goes all the way back to the Middle ages. I mean I can’t believe when I look at my history books, I can’t believe that the English look back on Henry VIII fondly. I mean he was a nightmare. He said there he sacked the monasteries, he killed half his wives, he decided to create his own church and yet the Judds are seen as something nostalgic. It’s a terrible indictment of society. But that’s the way that knowledge graphs grow. Sorry, Myriam or
Myriam Jessier
Jess? Oh no, Jess and I were discussing something similar to this. One of the elements we would add to Marco’s recommendations is get controversial because it’s an attention economy. So if you get controversial, people will respond. So that signals quite a few things.
Dixon Jones
But getting controversial and getting your facts wrong are in danger of changing the truth because truth is only a perception anyway.
Jess Peck
I think that’s the really interesting thing about Google is Google has this place between they want to provide the truth but they also want to make their users happy. And the truth isn’t something that necessarily is going to make users happy. One thing with the Lydia and Fonte situation that I thought was interesting is the list of dudes that came up was sourced from a very specific blog article where someone had basically sourced other lists and then put his himself as the number one best seo and that was what Google was using as a source for this featured snippet. And I wouldn’t necessarily say it was well SEO’d as an article, but it definitely, it was very clear it was ordered, it had pretty good html structure. And I think the other thing is it included stuff from other lists. So Google thought it was building on the consensus rather than breaking away from it. So if I wanted to include some women in my Tech SEO list, I would probably include a sublist of these are the people you will always see and then add extra because saying that let’s Google kind of understand I’m not here without two out their ideas, but I have some new stuff and you want to diversify your results so promote me.
Dixon Jones
So I want to come on and ask about what kind of data sets Google trusts then in a second. But I’ll just say thanks to Yusuf who’s in the group for saying basically AI is not perfect and Monste is in there pointing out that technology is made by humans and it’s bound to have mistakes and biased. And Jack we’ve already mentioned in there as well. So thanks guys for jumping in and putting you, putting your thoughts in there. But given that we’ve talked about bias so much, I think it’s probably a good place to jump in and ask about data sets that, well firstly data sets that Google may trust or may not trust and then data sets that you trust or may not trust because do you trust the ones that the Google trusts? I mean the one that I know that Google trusts is Wikipedia because it used Wikipedia or stated that they used Wikipedia as a training set when it started to build its knowledge graph. And I still believe that they do that today. The reason I think it’s quite a good data source is because it’s automatically human moderated. So it’s effectively self supervising but it still has huge amounts of bias in there. If you go for famous peers, you don’t have the Ethiopia Addis Ababa Pier, which is a massive great big pier and huge one, but it’s got Brighton Pier, which is a tiny little town. So there’s a big difference from there. Myriam.
Myriam Jessier
I have to staunchly disagree about Wikipedia for one very big reason. There is a genuine problem going on with history being erased and rewritten based on sources that are included and seem legitimate in Wikipedia but are not. If you go down that rabbit hole and you actually follow each reference and you end up on the end document, you’re like this is far right not cpop propaganda. So there a lot are quite a few entries in Wikipedia today regarding World War II and the Holocaust that are actively being rewritten and every time you try to intervene or provide proper documentation and a historical sources being erased it, it’s a genuine problem that we see. So when we’re talking about how to train AI and how to get machines to hear us, quite a few humans have figured out how that works. Quite a few humans are doing it down low.
Dixon Jones
But then that brings you back to the question of what data sets do you think Google should be using?
Marco Giordano
Yeah, so of course the basic Wikipedia, but not just Wikipedia because actually IMDB or these big aggregators information databases of course have a lot of useful data for search changing that. That’s how I would do, I would just pick aggregators as Myriam said. Unfortunately Wikipedia is not the only one. There are usually missing information, mistake, some info doesn’t exist on the web. I have an example, if you look from some forms of art, even movies or some super initial stuff, there are no Wikipedia pages. Maybe IMDB is complete. How do you fix that? You need some human to tell. Look, in the seventies there was this movie with this attributes and you are missing them. So usually Google uses mdb, Wikipedia or this sort of websites I’m quite sure as seed sites. And I also think that if you’re in a domain, let’s say laptops or something like that, tech, of course they will have a list because we know from the patents of course, I’m not saying that the patents are the absolute Bible, that there are some stuff is not implemented, but for sure they have a list or something like that where they say with these words in this context, these websites have more trust.
I consider, I don’t know this website tech me to be more reputable compared to your new domain for this and this reason. So maybe we don’t understand
Dixon Jones
And I think that trust, I think you mentioned where you said it and I actually agree that that trust is in context. There’s no way that Bill Gates is trustworthy when it comes to pop music. That is not his forte. Whereas, Sorry, yeah. Myriam
Myriam Jessier
Jess. No, I’m obsessed with, I want Bill Gates to give his top 10 pop songs. But what I find interesting about this, and I need to provide some context to this, I went to school for not seo, it was not a diploma, it was not something taught in schools when I got started. And what I was interested in was sociology. So I think about these things quite a lot and how they relate because first and foremost, I think Genie had a very good point with entities as human beings, one of the things that sets us apart from many other species is that if I tell you the word table right now, you will have an understanding of what a table is. Okay? Right there, you get it? Okay. That’s what we are trying to teach to machines. But when it comes to data sources, one of the theories that I really, really like in sociology is the two step flow of communication.
So the two step flow of communication is you automatically think that a big fire hose of information coming to you, mass media is going to go straight to people’s brains. But that’s not the truth. Most people will look to opinion leaders to people they trust to filter this information and relay it to them. That’s why we have an epidemic of people going, well I know this and you’re like Baha, do you know who told you? And how did they learn that? And there’s no information. It goes because of the two step flow. We choose to trust certain opinion leaders. And when it comes to data sources, I see it the same way. I mean personally I wouldn’t trust imdb. A lot of people lie on their ages on there quite a lot. But this is how we operate and this is how we are teaching machines to operate as well. This is also a bias we have to be aware of.
Dixon Jones
Go on Jess
Jess Peck
And to kind of build off that, I think there is no, Google cannot trust one data source by itself is the fundamental thing. Google can’t trust Wikipedia by itself. I was thinking about a large part of the Scot’s Wikipedia a few years ago was discovered to have been written by an American teenager who didn’t know any Scot’s like it. They can’t trust it alone. They can’t trust any data sources alone because all of them have biases, blind spots, things that they’re gonna get wrong. There’s a logical fallacy about a newspaper reading where you read an article about something really well and you’re like, this is complete nonsense, these people don’t know what they’re talking about. And you turn the page and read something about particle physics and you’re like, oh this is really interesting. And it’s because no one can get that clarity across all things, which is why Google uses ml because they can cluster and discard outliers and as long as the things they are clustering don’t all have the same biases or the same blind spots, they’re going to get more consistent results.
But the second a hundred of their sources say the moon is made of Swiss cheese, Google starts to think the moon is made of Swiss cheese. And I didn’t experiment to try and generate a featured snippet on a question that didn’t make sense with an answer that didn’t make sense. And Google doesn’t, because Google didn’t have enough information to cluster about it, it could only rely on the sources that had that question and answer paired. And that was me and a bunch of sites that I’d created that were all lying about it collectively. And this is why thinking about language and information in terms of clustering can be useful when you’re doing SEO is because if you know the cluster you’re trying to inhabit, you can produce what’s more trust.
Dixon Jones
So how does Google cluster that information then? So how do they decide the difference between the one, the one that in my head at the moment is a Mustang car and a Mustang horse, What is the important cluster? What’s the overriding cluster that Google might say, well this is talking about a horse and this one talking about a car.
Jess Peck
I mean I think this is a good Genie question, but it’s entities, it’s Knowledge Panels, it’s taking that content and going, well these are the terms that exist mostly here. This is the kind of knowledge graph information that leads into this. Your horses are gonna have a lot of stuff about stables and saddles and your mustangs are gonna have a lot of stuff about going real fast. I guess they’re both have real fast, is that
Dixon Jones
Yeah. And then they have
Jess Peck
Price
Dixon Jones
Car. Yeah, I chose a bad one cuz this is the subject of my talk next week and so I shouldn’t have done that. Choose another example that Genie. But I mean when you see a text of page that’s got something out of context, does that, what tends to happen?
Genie Jones
Yeah, I mean I was reading an article about this today about someone, I think it was literally on the news, the football player (Haaland) and the place in
Dixon Jones
Yes, Halland place, a
Genie Jones
Place
Dixon Jones
In a place in the Netherlands. Yeah. It’s not
Genie Jones
So funny.
Dixon Jones
Yeah, yeah. The guy from Visit Halland Yeah,
Genie Jones
Yeah. Got completely mixed up and now even when you are typing, but they got mixed up because people were typing the wrong things even though they had the intention of looking for the football player or no, even if they have, So they were getting the mixed up by one letter. Also the context of the fact that the World cup is coming up is not helping. So it’s looking like they’re indistinguishable to Google when actually what Google is doing in my opinion, I think is taking into account the weight of the context of not only what it knows and its data set within the difference between how likely is it that they would be looking for these things. But also it, it’s so constantly updating itself on trends that are going on at the minute. So the fact that the world cup is coming up means that the context of this is completely different.
And one thing I was also gonna say is in terms of this, I think you said something, the moon is made of cheese and Swiss cheese. As soon as somebody says that the whole thing goes out the, I think part of machine learning is annoyingly that it’s, we are just going to have to wait for ages until we fix these data sets because then every time you have a bad connection, it still has a more meaningful connection now that we’ve talked about it. So we, we’ve gone, this is a lie and then we can think about, okay, now that entity has a whole separate cluster of entities where you go like oh people are talking about lies. And that’s a connection in itself that Google has to take into account and that’s gonna come over time from the things that we are talking about on these podcasts and stuff as like we create these new things and layers of understanding. So yeah, it’s just a really difficult one. So I guess back to how it goes, you can tell the difference between Mustang a horse Mustang. Yeah
Dixon Jones
Lisa Stansbury asked on that example, aren’t they looking at adjacent words in the sentence to create context yet re Mustang horse and car. And I don’t think that’s what used to happen and certainly they’ve got this Burt and they read things forwards and packs and stuff, but to my mind now and everyone can disagree with me, I think it’s much more about the concepts that are used with on that page of content or even that paragraph of content will,
Genie Jones
Or even that website of content because we are using internal links now to add context. So it does, it’s not limited to just that site, sorry, that page anymore. It’s limited to, okay, every time this person is talking about a car they’re linking to this Mustang page, which therefore is actually training it in a way to figure out that that’s about a car. If you were talking about Mustang and then it didn’t quite get it from that actual page, but all the internal links going to it from the other parts of your page were about horses. I think it would make a significant difference
Dixon Jones
To the outside. So if there’s a lawyer’s site, sorry Jess, I’ll bring you in a second. So you’re saying if there’s a lawyer’s site and the site is all about lawyers, then when the word marriage comes in, it’s got a completely different context to when it’s about Exactly. My husband and I. Yeah. Okay Jess,
Jess Peck
I just wanted to add to that and say I think even beyond internal links, if a million bonds sites linked to a page about a Mustang car, I think that would confuse the entity he understanding of that page because you’ve got a lot of very powerful signals saying this is something to do with barns, this is something to do with horses, horse shoes. And that’s sometimes not even something you can control
Dixon Jones
And that could be a dangerous thing. So if you buy a domain that has had a history then and you 301 you may doing yourself a complete disservice because of some of the legacy links. If it’s different, go
Genie Jones
On, have a quick question. It’s almost like I wonder how long it will take, or even if someone can enlighten us about if they’re already happening of Google, having your data anywhere of what things you tend to look at a lot. And then that also being a factor in topical authority. So in not even hard coding it with a link, but actually just memorizing your history, figuring out what authority you might be looking for and then recording that and it’s all
Dixon Jones
Part ways. They definitely do that. And a good example of that is if you’ve got an Android phone in Google Discover, so Google Discover is showing you things about hiking because you went hiking and those kind of things. So yeah, Google is
Genie Jones
Searching news, but I mean more than that in terms of actually it being becoming part of the footprint of your actual site and the authority that it has, that’s kind of put the onus on the visitor to get those kinds of things. But I wonder if, I dunno, it’s just an interesting thing to think about of,
Dixon Jones
I personally think that Google spends a lot of time looking at using your search history if you’ve given permission using your search history to help you to help understand what you mean when you type in a few words in the query. But I know other SEOs completely disagree with me on that one, so I’m not going to ram it down through anyone’s throats here today. I wanna talk about this for hours now, but cuz it’s one of my favorite talks and actually we spent a lot of time on good bias in the knowledge graph, which is a pet favorite conversation for me. But we are pretty much close to time already, so I don’t want to, before we wrap up, maybe I can go around the room and get a second tip. We had a number of questions we didn’t get to today. If there’s something else that you think you should share with everybody before we go, I think that would be a good thing cuz I don’t want to put words in people’s mouths. Marco, can you come up with something else about machine learning that we haven’t discussed so far?
Marco Giordano
Yeah, okay. Actually maybe it’s not probably, it’s not about machine learning, it’s more about analytics and systems. No, for your workflow. So especially for content websites, in my opinion, when you are describing a huge topic or a very broad topic like laptops, super generic one, sometimes scraping Wikipedia or any comp, especially competitors with text rise or everything, even links, whatever you want that detects entities is a great idea because I found that something in, even if I am super knowledgeable about the topic, I miss some details, some attributes or maybe some sub facts that I simply didn’t know that maybe are important for Google but not for you as a human.
Dixon Jones
Brilliant. Genie, you got anything to add onto there?
Genie Jones
Well I guess mine could build on that in a way of, to put a name on it, like flooding the niche in with everything that you possibly can is a really great way to demystify content for Google. So if you want people to understand that you are talking about cats or cat food or something like that, you need to talk about everything related to it. Well you don’t need to, but it’s a good idea with supporting content and that can come from figuring out which entities are related and how relevant they are in different contexts.
Dixon Jones
You say that by default helps to show that you are an authority in that subject matter.
Genie Jones
Yeah, absolutely.
Dixon Jones
Myriam, do you wanna add anything there?
Myriam Jessier
Yes. So something we talk about quite often with Marco is how exhausting it can all be because it’s an economy of scale. Like you just, Genie just said it, you have to flood the market. Marco was explaining you have to scrape at scale some things. So one of the shortcuts that we can take as eos build a brand. If you’re not a massive company, if you don’t have an unlimited budget, what can you do to send these signals, build a brand, this genuinely helps
Dixon Jones
And make your brand associated with a topic cluster
Myriam Jessier
Otherwise. Oh absolutely don’t, nobody builds a brand just to build a brand, I hope, except the Kardashians
Jess Peck
To make a lot of money.
Dixon Jones
Yeah, you do. All right. Yeah. Jess, what about you? Any last thoughts?
Jess Peck
Like obvious manipulation with AI and algorithms, places which they are, like the engineers miss other kinds of things that get like, well that kind of manipulation dies out quickly if you’re trying to manipulate things that have obvious fixes and rely on them. Like that isn’t the kind of thing that is sustainable. If you want to manipulate ai, do it in a way that builds and manipulates the stuff that it’s trying to get to if that, yeah.
Dixon Jones
Okay guys, I really do appreciate you guys coming on. I really would like to talk more. So before I hand back to David to talk us about the next session, if people want to ask you more, how do they get hold of you? Jess, why don’t start with you. Where do they get hold of you?
Jess Peck
I’m on,
Dixon Jones
You want them to?
Jess Peck
I’m on Twitter too much. I’m JesstheBP on Twitter. Talk me.
Dixon Jones
It’s fine. Okay, that’s brilliant. Myriam, where do they find how you
Myriam Jessier
Exactly? Like Jess said, I’m on Twitter way too much and I should stop, but you can find me at @MyriamJessier on LinkedIn – Myriam Jessier.
Dixon Jones
Okay. And Jessier is spelled, so Myriam is spelled with a y m y r i a m and jessier, j e double S I E r. That’s for people that can’t see the CS screen. Yes, Genie. How about you?
Genie Jones
Well, I’m not sure how long I’ll be on Twitter anymore now Elon has taken over, but you can find me there at GenieRJones, that’s g e n i e. And also probably the best one is LinkedIn though.
Dixon Jones
Okay, so Genie with
Genie Jones
A G like that.
Dixon Jones
Okay. And Marco, how can they find you and his microphone? Other?
Marco Giordano
Yeah, link in for sure. On Twitter – GiordMarco96. My handle is in Italian, so I don’t know if it is very clear, but link it in first. Just don’t sell me bad back links, please, Please. Nothing good at
Dixon Jones
Least. So Marco is m a r c o and then Giordano is G I O R D A N O. Okay, brilliant guys, thanks very much David. Brilliant conversation today and I’m sure it could have gone on for twice as long, but thank you very much for producing it. What have we got next time round?
David Bain
Yeah, absolutely. You can tell it was a wonderful conversation simply by the fact that the listener, the viewer numbers, the live viewer numbers went up and up during the conversations so everyone stuck around and more people arrived as well. So that’s a wonderful sign to see. So next month, or should I say actually the very last day of this month, on November the 30th, I have a special webinar that actually previews the launch of the SEO in 2023 book. So you may remember that last year I interviewed 66 leading SEOs on their thoughts on SEO in 2022. Done the same thing again. In fact, interviewed 101 SEOs on their thoughts on SEO in 2023. That book is being launched at the beginning of December and we’re gonna have a webinar as part of the ogar new blood series to discuss the launch of that new book and get a few of the contributors on as well. So how be on November the 30th
Dixon Jones
You’re gonna give me a pre pre-launch copy, aren’t you? So I could, I’ve got I’m organized before that one, aren’t you?
David Bain
Absolutely, absolutely. The initial version of the book has already been produced. All I need to do is write the opening thoughts, closing thoughts, maybe a little bit of editing, but we’re nearly there in terms of the book itself. So it will be ready for the beginning of December. It will be getting you a copy of course sticks and getting all the contributors a copy as well. So I just wanna say that that particular episode is gonna be on November the 30th. It’s a 12 time, 1200 PM Eastern standard time, 500 PM gmt. So if you can go to majestic.com/webinars and sign up and hopefully we’ll see you there.
Dixon Jones
Brilliant guys, thanks very much for coming on. Absolutely delightful to have you all on and thanks for the really interesting chat conversation. Bye. Thank you, bye.
Previous Webinars
Follow our Twitter account @Majestic to hear about more upcoming webinars!
Or if you want to catch up with all of our webinars, you can find them on our Digital Marketing Webinars page.
- Setting an SEO Strategy for 2025 - December 18, 2024
- What Changed in SEO in 2024? - November 27, 2024
- Get onto Bluesky. A Guide for SEOs - November 18, 2024