Speaker 1: 0:52

Hi Ed Dawson here, and, as I'm a bit busy at the moment and need a break, welcome to another one of my best of SEO is not that hard podcasts. These are the episodes from the back catalog that I think have the greatest hits and ones that are still relevant and provide great value for you. So, without further ado, let's get into the episode. Hello and welcome to episode 62 of SEO is not that hard. I'm your host, ed Dawson, the founder of KeywordsPeopleUsecom, the solution to finding the questions people ask online. I'm an affiliate marketer, seo doing and monetizing websites for over 20 years. I've built sites from the ground up. I've bought sites and sold sites in large exits. I'm here to share with you the SEO knowledge, hints and tips I've built up over the years.

Speaker 1: 1:38

Today, I'm going to talk about ChatGPT versus Google as a source of SEO data. I'm seeing more and more people say that ChatGPT is a suitable resource for some SEO tasks like keyword research and keyword clustering and topical map creation. What I don't see so much is people suggesting that Google is the best source of data for these tasks, which, I have to be absolutely honest, baffles me. So let me explain why this messes with my head so much Now, don't get me wrong. Chatgpt can be an awesome tool in many cases. I use it regularly myself as a coding assistant when I'm prototyping new tools. It's really good for improving productivity in that respect. It's also great for content in many respects, especially outlines, if you prompt it with enough base input data.

Speaker 1: 2:20

But when it comes to SEO research, it has a few fundamental flaws built in now. Chat GPT is a large language model. It's trained on a huge data set of information from a number of sources. Some of these are books, some websites and other information sources. There's all textual information that it works from. Now, from this data set, it can make some very good output. Like I mentioned before, I'm especially impressed with how it works as a coding assistant.

Speaker 1: 2:49

But what it doesn't have is a model of the web and how websites relate to each other. It doesn't have any idea of how the content on different pages and different to each other. It doesn't have any idea of how the content on different pages and different sites work together. It also has no idea about how real people search online, nor how they interact with the content online in sites. So when you ask it for keywords for a topic or questions that people ask, or to cluster keywords to build topical Macs. It will give you an answer, you know, but that answer at first glance the results might look quite credible, but what you've got to appreciate is these answers are not based on any underlying data on how the web is built, how the web works, links together, how people use the web and what actually is happening online. It's only looking at content. It doesn't look at how that content is used, linked, all those things. So what it's giving you is just guesswork. And then there's also the limitation that ChatGPT's data sets are static. They're only updated periodically. So I think what's available at the moment is April 2023 is when it's based from, so you know, getting towards a year out of date. So the data it does have goes stale over time.

Speaker 1: 4:09

In contrast, google is actually based on a real-time model of the web, based on the content that it actually crawls and the link relationships between pages. As we know, google, when it crawls, it's not just looking at the content, it's also looking at the content in the context of how it relates to other pages and how other websites link to each other, and it's that model that adds a huge amount of value that ChatGPT doesn't have adds a huge amount of value that ChatGPT doesn't have. Now, google also knows exactly what people are searching for, with what keywords and questions, and they can see how this changes over time. Google also knows how people interact with web pages from user behavior data that they collect as people search and bounce back from websites to SERPs. They also have behavioral data from places like Google Chrome and Google also are constantly crawling the web, constantly being fed more search input from users and more behavioural data, and this is why, if you pull data from Google wherever possible, it's a much, much superior source of SEO data as compared to chat GPT. Now, a great example is the obvious upfront data that Google shows in its search results, like people also ask in Google auto suggest data. You know this is up to date, it's geographically and language relevant and it's based on real user input the actual questions and keywords. People are using the questions and keywords you want to be targeting. But there's also additional layers of data that we can pull out of Google if you have the right.

Speaker 1: 5:42

It's what it sounds simple. It's essentially the task of taking a whole bunch of keywords and then grouping them together into related keywords. Now we do this so that we can build content for each kind of cluster of keywords, so that we don't cannibalize keywords over multiple pages. We get everything clustered together nicely so we can create nice tight content around those clusters on single pages or groups of related pages. It's the kind of the basis of building topical maps which a lot of people are talking about at the moment. Now it sounds like a simple task to do, but as you scale it up over a large number of keywords it gets more and more difficult to do. But also to do it well, you need data on how well related different keywords are and by what degree.

Speaker 1: 6:32

Now this is a problem that we've been working on at keywords people use for probably the best part of a year now. We've been trying all sorts of different techniques and we've tried, using chat gpt, this and we've tried multiple different prompts, multiple layers of prompts, doing partial prompts. So we're putting data in, getting data out and then manipulating that data and putting it back into it, manipulating with other prompts and trying to build up these clusters using ChatGPT's OpenAI API and what we found is that at first glance on small, especially on small sort of you know, a few dozen keywords, it can give what looks like a good answer, but when you try and build it up to larger data sets or more complex topics, it really starts to fall away and the quality just isn't there. And we found that it it's not something that we'd be happy to use and happy to put our name to, and that's because of these fundamental problems with chat gpt and the way that it doesn't really know how content relates to each other and the way that people use content. So we got to the point a few months ago and we were like it's not going to work using chat gpt to try and do this clustering.

Speaker 1: 7:54

And this is where we went back and thought right, let's go back to first principles. Who's more likely to have a better source of data that we can interpret to build these keyword clusters? And we went back to Google and we had the hypothesis that if we looked at the live Google search results for a keyword and looked at what websites, what pages, were ranking for that keyword, and then we take other keywords and do the same, see what websites they're ranking for what URLs, and build this map of all the different keywords that we're looking to cluster, look at all the sites that are ranking and the URLs are ranking for each of those keywords and then look for intersects between those keywords. That might give us a clue as to which keywords should cluster together, because if similar pages are ranking for similar keywords, then that would mean that those keywords are very closely related, because Google's already done this big task of crawling the web, working out what content relates to each other, what keywords relate to what content, and then working out where those overlaps are. So if we can do that same process, kind of in reverse reverse engineer Google to find that data, then it might give us good keyword clusters. And we prototyped this to start with, just simply to see would it work. That data, then it might give us good keyword clusters.

Speaker 1: 9:20

And we prototype this to start with just, you know, simply to see would it would it work. And I'll be honest, we were absolutely astounded with how good the results were, how tight those keyword clusters are, how how much they make sense and how they build together and you don't get erroneous questions in there. It it nicely clusters things that are related. It keeps things that are unrelated separate so you don't end up with garbage data. Basically, it's not guesswork, it's actually based on real data, based on google's crawls, based on how users react to keywords, what they search for, what they click on, builds up that data set that we can then reverse engineer by making these queries to Google. Now, obviously it is more complex than just typing in a prompt chat GPT and getting a guess of chat GPT and it does take slightly longer to run.

Speaker 1: 10:17

There's a lot of queries going on. If you've got thousands of keywords, you're making thousands of queries to google and google doesn't like being queried. That's why you know you need tools that um, like keywords people use, that know how to get around the issue of crawling google and other sources at scale and get around the the um, the protections they put into trying to do this. You can't just run it from from a python script at home. Um, you know you'll soon find, after a few queries, you'll, you'll band, you'll be getting um captures coming up. It's just not possible to do. But we got around those issues. And, yeah, the data that you get back in the clusters you get back absolutely incredible.

Speaker 1: 10:59

Now, as I said, we built this originally. You know we brought a sort of a prototype that worked on our servers and you had to get these results now. What we're currently in the process of doing is just turning that into actual production ready code with production ready interfaces that we can put onto keywordspeopleusecom, and that is coming imminently hopefully, um, within the next couple of weeks, certainly sometime during february, I imagine um. So we're going to make this available for people to try and and and to to play with. So do look out for when it comes. But the thing that I want to sort of return to here is the fact that the thing that I want to sort of return to here is the fact that chat GPT is being given as a kind of like a magic black box that can do everything for you and come up with all the answers, and there's a lot I'm seeing people suggesting it's the real quick and easy way to get the results that you want.

Speaker 1: 11:56

Now it's not really um. It can be a good start. It can be a great tool for doing some research, basic research. It can be useful in certain respects. You know, and I said I do use it all the time for many, many tasks. But when it comes to finding out what people are actually searching online, the questions are actually asking and then taking keyword data and question data and trying to pull it together into topical maps and clusters for actually producing content and producing a plan with it really isn't great, and if that's the only thing you use, then you're not really going to, you're not going to get the full value. You're're asking the right questions, but you're asking the wrong place and, like I said, the reasons why ChatGPT doesn't have that data it doesn't know the relationships between content, it doesn't know how people are searching online. There's all those missing parts of the key to answering the question just aren't there with ChatGPT. So that's why you really do need to go to Google and mine Google to find those answers that you need, and you'll get much better answers that will lead you to create the content that will drive the traffic that will take you and your sites forward, and it's what you need Now.

Speaker 1: 13:18

I don't expect you to just take my word for it, but you can actually check for yourself Right now. I'd say go to keywordspeopleusecom, put in a topic, find the questions that we mine for Google for that topic, and you'll see we'll return a whole load of questions that people are actually using, based on People's First Data or Google Auto Suggest Data, whichever one you want to choose, which is both of them. Get that data. It's free to try it and then ask the same thing of chat gpt, ask, ask for it's got any questions that people ask on that topic? And then compare the two and you will see that the results that you get from mining google are better than chat gpt, and that it's even more obvious when it's a trending topic, because ChatGPT has out-of-date data. Google has up-to-date data. You can see that straight away as soon as we've got the clustering tool up and running.

Speaker 1: 14:11

I'd say again do the same thing. You can then go and give us, give us some list of keywords, see how our clusters can compare to the clusters that ChatGPT creates, and you will be astounded by the difference in quality. If you just sit there and look and analyze that data and think, have I got the right content on all these things, you will see the results are far better for mining Google as compared to ChatGPT. There's no reason then. Obviously, once you know what you're going to write, take it to ChatGPT when you know what you're going to write, take it to chat gpt when you've. When you know what you want to write about and you know the keywords and questions you need to target and you know how the clusters need to work. You can chat. Gpt can be excellent in terms of helping with content. Briefs can be excellent with writing content, but it's only as good as the questions you ask. Chat gpt and the best questions come from mining Google in the first place.

Speaker 1: 15:02

Thanks for listening to episode 62. 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 why thousands of people use us every day. Try it today for free at keywordspeopleusecom. If you want to get in touch questions I'd love to hear from you. I'm at channel 5 on twitter. You can email me at podcast at keywordspeopleusecom. Bye for now and see you in the next episode of seo is not that hard.