April 17, 2023
ChatGPT offers a seemingly neat solution to the pain points that come with SEO content creation. But what does this groundbreaking technology mean for the future of the web? It’s the question on every SEO’s mind—and what our hosts will be getting to the bottom of in this upcoming discussion with Michael King and Ross Hudgens.
Check out the webinar's decks:
In this webinar, we'll cover:
When to use AI writers (and when not to)
What AI-generated content means for SEO
The implications for the state of the web overall
Meet your hosts:
Michael King
Founder and CEO, iPullRank
An artist and a technologist all rolled into one, Mike is the Founder and CEO of digital marketing agency, iPullRank. Mike consults with companies all over the world, including brands ranging from SAP, American Express, HSBC, SanDisk, General Mills, and FTD, to a laundry list of promising eCommerce, publisher, and financial services organizations. Twitter | LinkedIn
Ross Hudgens
Founder and CEO, Siege Media
Founded and led by Ross, content marketing agency Siege Media has made the Inc. 5000 List for the last five years running. Before this, Ross built websites from launch to #1 rankings for extremely competitive queries. He’s been featured on the likes of Moz, Search Engine Land and Forbes and is also a frequent speaker at conferences such as MozCon and LearnInbound. Twitter | LinkedIn
Mordy Oberstein Head of SEO Branding, Wix
In addition to leading SEO Branding at Wix, Mordy also serves as a communications advisor for Semrush. Dedicated to SEO education, Mordy is an organizer of SEOchat and a popular industry author and speaker. Tune in to hear him on Wix’s SEO podcast SERP’s Up, as well as Edge of the Web.
Crystal Carter Head of SEO Communications, Wix
Crystal is an SEO and Digital Marketing professional with over 15 years of experience. Her global business clients have included Disney, McDonalds and Tomy. An avid SEO Communicator, her work has been featured at Google Search Central, brightonSEO, Moz, DeepCrawl (Lumar), Semrush and more.
Transcript: ChatGPT and AI writers in SEO content
Crystal Carter 0:00
Today we have myself and Mordy Oberstein who are permanent fixtures here on the Wix webinar scene. And then we also are joined today by Mike King, who is the founder and CEO of iPullRank. He is an artist and technologist all rolled into one. He runs a fantastic digital marketing agency, he consults with companies all over the world, and he's been working in AI and LLM models for years. We're so pleased to have you here today, Michael, thank you very much.
Mike King 0:32
Yeah, of course.
Crystal Carter 0:35
We are also joined today by Ross Hudgens. Ross is the founder and CEO of Siege media andhas made the Inc 5000 list for the last five years running. He's done some fantastic work with Moz and Search Engine Land and forums. He's a frequent speaker at LearnInBound, he is also someone who works in the AI space quite significantly, and has some fantastic 49ers memorabilia, we're so so pleased to have Ross here. Thank you for joining us Ross.
Ross Hudgens 1:12
Yeah, thanks for having me excited to be here.
Crystal Carter 1:13
So we're gonna get into it in just a moment. We are working on some of those admin things, we'll sort those out shortly. But just to let you know the webinar is being recorded. Absolutely, it's being recorded, it will not only be shared on the page that you signed up for the webinar on, but it will also be shared on YouTube. So find it in all of the places where you've enjoyed regular content. And you'll get a link to the YouTube page on email as well. And you are all very familiar with the Q&A panel, as we're working on sorting out the chat panel. So ask your questions in the Q&A panel, as we go along, and we will curate those and answer them at the end. If you want to know more about our future Wix webinars, then visit Wix.com/SEO/ learn/ webinars and you'll find all of those there. And moving on to our agenda for the day, I'm gonna hand it over to Mordy Oberstein.
Mordy Oberstein 2:02
So before we get started, you know, I come from the from the content side of SEO. So AI writers were an interesting avenue, AI generated content is fascinating, and it's novel, and it solves a lot of pain points. For those who are not on the writing side of the world, writing is difficult, and it's time consuming, and AI writers came along, and AI generated content seemingly has solved that problem. Writing is no longer hard, and it's no longer extremely time consuming, it would seem. So we thought it'd be really, really important to have a webinar to really dive into just when do you use AI written content? When is it not appropriate? What are the best ways to use it? How does it impact SEO going forward? Because it does seem like a panacea. But generally, things that are too good to be true or sound too good to be true, are generally too good to be true. And while AI writers are an amazing tool, I think it's worthwhile to have a look at how to use them responsibly because, I try to think about the web as a child, it goes through stages. It could be a child, it could be a toddler, it grows up, it becomes a preteen, it becomes a teenager. I would say the web is somewhere around like a teenager right now, if I had to like put my finger on where's the web, and teenagers are generally, just thinking about my own my own life, not the most responsible. And I think that there's a tremendous amount of opportunity with AI generated content. But I also think there's a tremendous amount of potential problems that come with it. So we thought it'd be really, really, really important to dive into how AI writers will impact the web and SEO. And for that we're gonna turn it over to Ross Hudgens who's going to handle AI for content creation, then dive into the impact of AI on search with Mike. And then of course, we'll have your questions at the end of that, if you see me, by the way, and I'm I'm scrambling around it's because I'm looking through the Q&A, as the webinar goes along, to try to take down the questions that are most pertinent. And with that, let's hand it over to Ross.
Crystal Carter 4:18
Fantastic. I'm just gonna stop sharing my screen.
Ross Hudgens 4:22
Thank you Mordy. Great. I will screen my share on my side. I'm excited to be here to talk about AI generated content, as Marty was kind of setting up there. I mean, as he nicely described, I think some of the initial reaction a lot of people can have is, you can press a button and create content with AI--why don't I do this with every keyword I'm trying to rank for? And can I go about doing that using these tools, that seems so possible and feasible to do that? So that's going to be essentially what I'm going to get into, with that in mind, maybe that's not really the case, unfortunately. But maybe there's a middle ground where we can still use this very powerful tool to help our businesses be successful through search, and what are some of the ways we can do that. So before I get into it, I've got a good introduction already, but just to kind of restate it, we've been around for 10 years, we're a content marketing agency with SEO specialization, so creating content online to rank on Google is pretty much all we do. So this is our sweet spot. And I founded the company as well. So I've been thinking about strategy on this side for a long, long time, as I was teed up previously, worked with a lot of great ecommerce brands across the web. And that's sort of some of the context for some of the companies we work with in terms of the advice I might give today. So first, essentially I want to start with the idea that the goal of SEO is not simply to publish content. If the goal of SEO was just to publish content, then creating content with AI would be optimal. But unfortunately, that is not the case. Instead, the goal of SEO is not to publish, it's to win is to build the absolute best thing or have the goal of doing so in order to achieve that goal. So really, to achieve the number one ranking, you need to have that argument that it's the best content, not the same content as everyone else, because if you're simply generating the exact same thing everyone else has generated, there's no way you can truly make the argument that you're the best thing for that result. So that is part of the conundrum here with AI content.
The issue is that essentially, content is a commodity. If all of us go to ChatGPT or your tool of choice, and you ask it's similar things in terms of creating a blog post, you're gonna get a very similar result, it's at that point, very hard to argue that you're truly making the best thing in that case. So to kind of give some examples and in flavor there, if we're searching for something like, "what is SEO", and we input it into ChatGPT, you end up with very similar feeling content, not that I'm saying these four on the left did that. But you can sense how sort of it ends up repurposing a lot of what exists, and you end up with very similar things. But there is a solution here. What you do want to do with these tools is to not use them as output engines in isolation. Don't just go to these tools and say, give me a blog post meant to rank for what is SEO, without providing anything, if we provide it something, or rather something great, that's when magic is possible. But before we get there, just want to show an example of this in practice, of just how sameness can occur with very simplistic prompts and inputs for the tool. So if we go there and say, hey, I'd love a guide to rank for Wix SEO, please give me some recommended URLs and maybe metadata and maybe some supporting images, I might add to that result as well, we get a rough looking blog post like this, we won't read through the whole thing. But for the most part, it's a relatively simple article that if you go search these things, Wix is ranking, good job Wix, so that makes sense. But the other articles as well feel very similar to this, but I'd say a little bit better on average. And some of the issues beyond just some SEO practices being signed off, at least from our definition, putting the keyword far to the right, or putting the keyword at the end of URL. One of the things that occurs on this suggestion from them, is that really the only Wix specific thing and the entire blog post is this first section of the Wix SEO Wiz. So that's not very Wix SEO specific, it's not additive, necessarily. It's not finding anything novel to add to this, it's essentially repurposing what already is there. And that, for that reason, is going to not allow an article like this to rank, even if it's a decent article, all things considered. So what you need to rank though, is the best content that does exist for that topic. So if we take this further, what do we need to get there? Essentially, the magic is when you can give this tool or these tools unique inputs, you can create content faster and better and get new things that would allow you to get to content that actually can win. So in this left chart, let's say we have a unique data set of 50 or whatever number of populated cities in the United States with unique data points around how many five star restaurants and bars they have, how many popular events are there are to maybe create a study on like the best cities to visit or best cities for things to do per capita, hypothetically, if we had and curated this dataset, brought it together and then put provided it to the tool, we then could ask it to create a blog post, or even a summary on this concept and get something that no one else ever had. Because we are bringing it something that nobody else ever has asked. So in this case, we do that. We ask it ChatGPT, can you please summarize my data, and you can see now it's actually immediately coming up with insights based on the data that's provided in that chart, we don't even need to think about it, it creates copying in a pretty well done way. And this is just a summary. It's definitely capable of taking that same unique data set, and if I go to it and say, hey, I'd like to create a study of the 10 best cities for things to do in the United States, please write in the brand voice of Trulia, 400 500 words that could be pretty solid, you'd probably likely need to improve it. But because you brought it unique insights from the start, rather than just asking for a blog post of things to do, you're more likely to actually create something that's great because you brought it something great up front.
Another example of this is if you bring it unique things about your company, and then ask it to describe that so ChatGPT4 can now interpret and view images when you share a URL with the interface. So if I share this shoulder bag, and then ask it to write 150 words of branded sales copy in the brand voice and tone of this brand, it does a very strong job of that. So what we're doing here is we're giving it something unique. This is a unique photo from our unique product line that we spent a tonne of time and effort to curate and get right and also define our own brand. And we're asking it something specific that only we as a business could ask. This is not something competitor Y is going to ask in the same way ever again. So this is going to give us unique, novel, well done content that otherwise if we had not given a unique, well done input, we would have never gotten out as an output. So we can see here where we're getting to, if we give it the right things, which can be data which can be designed, which can be our products, it can then get to a place where it can actually create high quality content that might be able to support ranking in many instances. Another example is simply editing. So it's a very good line editor on average. So if you input copy that has issues or just barely any copy, it's pretty good at editing that at a high level. So if we think about the elements of content, that's essentially the same idea, we are given an input to get an output that's unique, no one else will only give ChatGPT, this exact same set of copy. So it's now going to take what I had and further improve it, rather than simply asking it to create a blog post on post structure for SEO, I'm now saying edit my blog post. So you can see that difference in mindset means I get something that Google's happy with, and users are happy with, and hopefully you can rank because of that. So I have a few little different typos in here, it might be possible to quickly pick up on that. But that's part of the potential value add of a tool like this for sure. So sort of recapping that I think you can use it as an output engine using information you feed it. If you simply ask it to be an output engine without feeding information, that's where we get stuck. We create content that's like everyone else is that has risk for all of our businesses. Another way of saying that is you can use it as an editor summarizer. Rather than simply creating from scratch. If you're using it for that kind of thought process, you're more likely to again, get to things that Google and users like, all things considered equal. Another tool we like for line editing and sheets and the like, and I'm sure Mike will get into a lot of this good stuff, I would guess. But this is good on the content creation side, plugs into Google Docs pretty easily and allows you to input your ChatGP, the AI is really well done. I know Wix has versions of this too, that sort of definitely ties into some of this concept where if you have a well done landing page, it essentially has the AI then create content based on what is already there, you've essentially applied this concept that I previously described in a well done way. So use AI to create content that supports SEO. Also, this is another way of doing it using input: so you could technically create content from scratch for things that aren't trying to rank directly for SEO without many inputs and be relatively okay as long as it's not trying to rank itself. So some examples of that would be things like a privacy policy or return policy, terms and conditions, you probably would still want to customize those for your business, but you could reasonably ask these tools to generate those for you. But these other pages are all good examples of areas where you could feed it inputs, depending on the kind of content, to get pretty good content back. So we saw the example of data for the data studies, what if you gave it 30 different images for your category, and then said create 100, 200 words of copy that describe this category for Stella store, which is another brand that uses Wix, and then create that copy, that's gonna be well done in terms of output that can definitely support SEO. So we talked about product pages as well, it could reasonably create and help you develop a career page. It's not going to rank directly for SEO, but it's going to be a valuable place to build a page like that. Social media copy is another great place and generally, more and more thinking about visual content, a lot of the brands on Wix, you have great visuals, from your products, photography, video, etc. So if you share that URL with the tool, and then have it interpret that design in your brand voice, you can get pretty well done work where really the main weight is being carried by your images rather than the text. And that's where this can have a lot of value. So we see in this left example, for Stella Store's top left image, I fed that into GPT4 and asked for Instagram social media copy. And comparing to the existing Instagram, it's relatively solid, using similar use of emojis, it seems to be relatively on point with the hashtags used as well. And as we hopefully we would generally all agree most of the weight in social media, Instagram type content, is the image itself. So you can get to high quality content a lot faster when you think about what is the most important part of this content and the least important areas, those are very clearly opportunities today to use tools like this to kind of fill out that experience while maintaining your brand advantages. So to kind of restate that there are many places where high quality content needs to be some more supported by text content. But that text content itself is relatively a commodity, or just not adding that much value. That's where AI really has a great place to then round out those non commoditized areas. So we're referencing back to this handbag, or shoulder bag. It's the product here, it's the site design here that carries, I would argue the majority of the weight and importance, they do a great job with copy in the bottom left, they even have a little poem that occurs down there. But if you fed that to AI, I would pretty confidently say it can output a very similar framing or theme around that while maintaining the core most important elements, which of course, are those product photos, which you still need to do by hand, at least for now. And the overall site design and aesthetic that is maintained there. So with that thought process in mind, there's a lot of great things that can occur. So to kind of sum it up, if you stick to this idea that to be productive with AI, give it best in class inputs, that's the goal of SEO is you need to have the best in class things. So if you believe you have the best in class handbags, and you give it the best in class photos of those handbags, and ask it for supportive copy, you realistically can rank with that page. But if you simply ask it to create content from scratch without much guidance, that's not best in class, you can't realistically do that in a way that's going to reliably help you rank consistently in search. So use it as an output engine using your best in class inputs. And that's how you're going to consistently win with SEO using this great new technology. That's it for me. Thank you for the time.
Crystal Carter 19:10
Thank you so much, Ross. That was brilliant, some really great insights there. And I saw a lot of people commenting in the chat that was really great to see some practical applications there. So thank you so much for that. I'm sure we'll get to the questions afterwards but I guess we can jump straight in for Mike's section for the wider scope. But thank you so much for that, really good, you've got lots of big thumbs up in the chat. So thank you. Michael, let's get you up next.
Mike King 19:46
Confirming that you can see my screen? Good? Cool. All right. So let's talk about what generative AI means for SEO. So real quick. I'm Mike King. And you know, despite popular expectation, I am not the real estate agent for that house back there. But you can follow me on all the things for iPullRank. I'm from an agency called iPullRank. You know, we do all the SEO and content strategy work and so on. And we've actually been using GPT technology since 2020. Here's a quick example, where we drove $300 million in incremental revenue for our client just by, you know, getting more generative copy on to product listing pages. So let's talk about how Google is under a lot of threats right now. One of which is that TikTok supplanted Google as being the site that gets the most traffic on the web. ChatGPT, in general, was a code red for Google, they said, like, hey, we gotta figure this thing out, which is so remarkable because the T in GPT is from a technology that Google invented. We've also got users believing that Google search quality is on a steep decline, you know, people are adding Reddit to the end of queries, because they feel like without it, Google isn't giving you good results. And so all of these are threats that are going to impact your content marketing and SEO. So the TikTok threat means that Google is going to rank more visual content, you're seeing more examples where people are looking up queries, like how to tie a tie, and it's not your blog post that turns it into 30 steps of how you do a Windsor knot, it's three videos that are ranking at the top. And so if you're not, you know, creating content that's going to be there, you're just not going to get the clicks. And so short form video is gonna get a lot more competitive, we're seeing that, you know, more and more people are about to ramp that effort up, because they feel like that's how they reach Gen Z. And so if you're going to be doing short form video, don't forget to put it on your website, most people just put it in the channel, Instagram, YouTube, or whatever, you really want to have it on your website, and there's a great guide on the Wix website, it talks about how to how to do video optimization that I think you should check out. But the other thing is that ad sales are actually down for Google as well as a function of the economy, and also as a function of people going more towards other platforms like TikTok to spend their money. And so what does that mean for organic search, it means that you're going to be in situations like this, where Mesothelioma, one of the most expensive if not the most expensive keyword, at least in the States has, you know, more whitespace around these ads, you're seeing the feature snippet take up more space. And so the user is going to most likely click on those ads. And so what that means is that the real estate is going to get smaller, and you're gonna have to be more effective as early as your metadata in order to be, you know, driving those clicks. But the last time that people were saying that Google search quality wasn't so good, we got two things that changed SEO forever, which were the Panda and Penguin updates. And so really, what this is telling us is that Google is coming at us with this or not us coming at people who are making bad content with this helpful content update, that we are likely to see something like we saw before, where they're just demoting content that's not valuable.
So if you leave with one thing today, leave with this idea on here that you have on the screen, right? You should only be creating content that's at the intersection of your audience persona expectations, your buyer persona expectations, and then search engine expectations. So I kind of split the audience persona from the buyer persona, because there are people that will, you know, consume your content and may influence your buyer personas, but will never become buyer personas. I think we all know what buyer personas are. And beyond that, you know, search engines are another persona that you need to account for, because they have a lot of expectations of your content in order to rank for your target queries. And so that's what I want to talk to you about today, pretty much, but through the lens of the threat of generative AI. So if you're a local business, small business, or what have you, you probably shouldn't know about what's going on in the local search space. And what we were finding is that content is still a lot of the top factors when it comes to ranking for a local business. And so this is from the latest local ranking factors from Whitespark. Obviously, this is also true in standard organic search as well. But the whole point here is that a lot more people are going to be making content this year. And this stat right here 54% of businesses, comes from Ross's team. So thanks for putting that together for me, Ross. We're also seeing that 47% of people are going to be increasing their blog, blog content. And of course, SEO is more than just blogs. But typically, when we're thinking about SEO, we're like, Okay, well, how do we write more stuff? Right? Well, here's the other part, marketers have the highest adoption of generative AI and it makes sense because we're the people that have to create the most content right? And so there's this growing list of generative AI tools, there's a tonne of them out there. They're all effectively just using, you know, the API's for ChatGPT or GPT4 or whatever, and then putting their spin on it with their own sorts of prompts. If you're using chat GPT directly, which I would actually recommend over any of those tools, I recommend you check out a tool called AI PRM, which is effectively like a prompt management tool. And it's also a community of prompt engineers. So you can pull from prompts that they've already written, and use that to generate your own content. And so one of the things that I've noticed by playing with this is that some of these prompts are like a paragraph long, and they generate really good copy. Whereas in most cases, people are just like, hey, give me a blog post about x. And they kind of leave it at that. And so if your prompt is one sentence, and you get back garbage, it's what you pretty much should expect. And so I would say just play with that tool, and also look at other people's prompts and see what you can learn from it and creating your content.
But the reality of it is that every tool on the planet is integrating ChatGPT in some form or fashion, including Wix. And I actually spent five minutes last night, building a website using Wix AI. And in fact, like, literally, it took five minutes only because I had to figure out which pictures I wanted to use, because the process is so easy. And so here's the example, right, like, I'm setting up my site, I'm generating my About copy, I'm giving the things that I want to include in that. So enterprise experience, full-stack developer, over $4 billion in incremental revenue driven, and then it gives me three different options to choose from, didn't love the first one, second one a lot better, selected it and then I've got my website. And so you know, I think that this is the sort of thing we're going to see a lot more of where people can really get to the point of generating content so that they can create the presence that they want. And in fact, things are already moving well beyond that, in that we have this new technology called AutoGPT, which is still kind of nascent at this point. But it can, you can just say like, Hey, I want you to act as let's say, a marketer, and figure out how to build me a presence that is going to make me money. And what it does is that just continues to make new prompts and do new things, to figure out what you want it to do without you having to prompt that every time. And so that just gives us a glimpse of where the future is going. But a big result of that is we're gonna get a whole lot more crap too. And so there's a gentleman named Doug Kessler, he gave a really compelling talk about this back in 2013. This idea that more and more brands, we're going to get into content marketing, is going to really be like a race to the bottom for constant quality. But another thing you can get is people giving you the answers of how to fix your chat while you're on a webinar. Who knew? Thanks, Steve. But Google actually loosened their stance on generated content as well, a few months ago, actually about a month or two ago. And I think this is one of the things that was holding up this big deluge of content before, because so many people were like, oh, you might get penalized by Google. Well, they're saying like, you know, as long as it's made for humans, it doesn't matter how was created. And if you look at the guidelines and more depth, they're just like, hey, we're actually really good at detecting spam anyway. So make it however you want, you can't really beat us, which I don't know is true. In fact, I don't think Google can reliably detect LLM continent or generative AI content. Because, you know, I've played with all the tools, and they all give a lot of false positives and false negatives. And so it's not really a strong enough signal to rely on by itself. In fact, open AI can't even reliably detect it, their tool only detects correctly 26% of the time. And so what Google has to do is combine any detection like that with the other signals that they have. So you may or may not be familiar with E-A-T, which I refused to say as EAT, I call it E-T instead. But anyway, if they layer those signals with, you know, whether or not they believe something is generated or not, then they can say, okay, well, it's generated. And plus, it's off the baseline for this given author for this given website. And so they're seeing a lot of, or we're seeing a lot of reports in the SEO community where people are saying, like, hey, I've been trying this out, my site got crushed. I guarantee you that those are also sites that just like, you know, spun up the content, didn't edit it, and just straight up published it. If you do that, you deserve what you get.
Now to the point that Ross was making, a lot of us are doing copycat content. You know, we're using tools like Surfer and Phrase and it's saying, like, hey, anything on this keyword features these keywords, has these headings and so on. And it's always been weird to me that because marketing is very much about differentiating yourself and standing out, but instead, people just copy each other endlessly. I don't get it as an artist, but you know, it is what it is. So if you want to survive, what you also really need to do is make better content than what's out there. And so, also, I feel like we need to update our understanding of search because it's very much out of date. And yes, this is going to be one of those like, oh, this guy's talking about patents but really, it's all about content and links.
So how Google works and all sorts of things actually work is what's called the vector space model. So you take a query, and you plot it in a multi dimensional space, you also take the document or webpages, and you plot those in multi dimensional space, and whichever ones are closer to the query are considered more relevant. So relevance is not like a qualitative idea. It's very much a quantitative idea. And so Google has a whole pipeline when they do they're crawling, they're processing, rendering, indexing, and ultimately ranking. That's not a shock to you, if you've been following SEO. But if you haven't, or even if you have been following SEO, the one key change that Google made, starting in about 2013 that really was like a quantum leap in better performance. And that was their switch from what's called lexical search to semantic search. So lexical search is really about counting words. Are these words on the page? How often are they on the page? How do they relate to other words that should be on the page, whereas semantic search is very much about meaning. And the way that that works is through that idea that I just talked to you about of converting words into multi dimensional coordinates in vector space. So as an example, here, you take the phrase, "how old are you?" It's converted into a series of decimal numbers, and then you can compare against other words in that same way. So here's an example from what's called the words of vector methodology that Google built back in 2013. So in the example, you take the vector for the word King, you subtract the vector for the word man, and then you add the vector for the word woman, and you get the vector for the word queen. So effectively, you can do mathematical operations, to determine the relevance and the relatedness of subjects, topics, keywords, and so on. So relevance is a function of a mathematical operation called cosine similarity. And so when the cosine similarity between these two vectors is close to one, that means that they're very close together or similar. When it's close to zero, that means they're orthogonal or not related. And if it's close to negative one, then it means they're opposites. And so Google came up with something called Bert, which allows them to have better vectors basically, where those relationships and the context was really captured. And what do I mean by that? Well, previously, the way that that was built, the word bank in both of these sentences would be considered the same. But with Bert, the word bank is understanding that river bank means something different than making a deposit in the bank. And so now that they can do that, they have a better understanding of the meaning and your content. And so you get these, what we call higher dimensionality vectors that allows you to really capture the information. And so that's how all of Google search works across YouTube images, everything, they're just looking at these different numbers that represent content, and then determining which is closest to the actual query. And so where Google has improved here, so this idea was called dense retrieval, where they can better understand the different aspects within your copy, not just looking at the copy in aggregate, but also, what does this paragraph mean versus what this other paragraph might mean. And so when you're seeing the feature snippets where they're highlighting, specifically in the paragraph where it is, that's this work, that's this methodology at work.
So the whole thing to take away here is these vectors, these embeddings, is how Google really understands content relevance in a way that they never did before. And so your website has a vector, your author, or you as an author has one. And so Google can associate things with your representation in this mathematical way. So another thing to know about this is that they're able to understand the relationships between different pages. So you're gonna want to build any links to your site, from pages that are actually relevant to what you're discussing, that's something that we've said in SEO for forever. But it's really why you're seeing that link building is working differently than it did before. Other things that if you're doing any sort of content marketing, your byline is an asset. So don't just be writing on random topics, because Google is associating your expertise based on this author vector that they're creating about you.
So the last thing I want to really talk about here is that relevance is not a qualitative measure. I've built a tool called Orbitwise, you can measure this and compare yourself to other, you know, websites. And so this is a free tool that you can try out. So here's my friend Vanessa. She's actually a brand strategist and a Wix user. I was looking at her website because she hit me up, she saw that I was doing So I was like, oh, let me see if I can bring her site into this. And so she, there's a keyword that she ranks for, which is how to build brand awareness. And I look to see who ranked number one for that. It's a site called Wordstream. And I wanted to see, is her page less relevant than Wordstreams because it doesn't rank as well, or is it just a function of authority. And so you put it in, you put the keyword into the tool, "how to build brand awareness", you put her website in, you put your email address, because you know, lead generation, and then you put in your URL, if it doesn't make it a top 10. And so what it's doing is it's scoring everything in that same idea that I just talked about that vector space model. And each of these dots represents one of those websites that's ranking for that keyword, but you're also getting these relevance scores. What you can see here is that the average score for was ranking in the top 10 is 72.42, or 7.43 and Vanessa's page is actually a 73, whereas Wordstream's pages a 72. And all these other ones are, you know, varying in that range. And so in this case, it isn't a function of her page being less relevant, it's more a function of her needing more authority. And so this is a good thing for you to know. Because you never know, do I need to, you know, optimize my content, or do I need to build more links? Now you can know. And so you can just check out the tool here. Alright, last point is the future of content and links. So we moved here, this is still a question that people are asking, like, how long should my content be for SEO? That actually should have never been the question because we've always been evolved beyond the word count idea.
In fact, the guy that used to run search a gentleman I met named Tsingtao. Back in the early 90s, he wrote a paper about why you shouldn't just look at the length of content to determine whether or not it should rank, what they do is what's called document length normalization, so everything is effectively compared, as though it's the same length. And, you know, again, marketers are still copying, we're all doing the Skyscraper Technique, except we're skipping the part where you make better content, because we're just going in and looking at what these tools tell us to do, and then inserting more keywords. And in fact, ChastGPT too can do that, or ChatGPT can do that for you. This is an example where I had written something about generative AI and I said, hey, put the keyword in here more in places where it makes sense and it does it quite well. So there's no reason that you should be toiling over there. Just give it after you've written your content over and then it can do it for you. But the reality is that everyone is going to be creating what we call like perfectly optimized content. That's an example of the integration between Jasper and Surfer. And you can literally drag and drop different headings and say, write me something about this. So really, we need to evolve beyond this complex version of keyword density, because Google has to sort all this out, you know, if everybody's writing the same stuff, who do they rank, right? Obviously, authority comes into play. But nevertheless, if everything else is equal, who do they rank? And so that's where this idea of information gain comes into play. And basically, what that saying is, if I've got 100 pages to choose from, which of these pages is saying something new that the other ones don't, and then based on that, I can give it a booster, so it ranks better. Google has a whole patent around this, they have a whole score for it. And I think as part of the helpful content update, they are kind of giving more weight to the score. So it's definitely something that you need to account for. So really, you want to cover all the bases for whatever expectations people have. But you also want to focus on saying something new. So we've got a guide on AI and content and SEO. So check it out. Just want to wrap it up real quick. Your relevance and authority needs to be your primary focuses so you can survive and always create your content at this intersection of audience personas, buyer personas and search engine expectations. I'm from iPullrank, we got a show called The SEO Weekly, check it out. And I got a book, you should buy it. That's all I got.
Crystal Carter 39:04
Thank you so much, Mike. That was fantastic. That was a whirlwind tour for anyone for anyone who was thinking that was that's a lot of information, and it was, this is all being recorded. It will be on YouTube. So you can go to YouTube, you can slow it down because Mike was moving quickly through a lot of a lot of some great concepts there and if you need to look down that fine, and the links will be shared. So we will share the decks with you as well, if you can see the link there. And a lot of people were really interested in Orbitwise. So that's great to see as well. And we've had a lot of questions. So I'm going to hand it over to Mordy to wrangle the questions from our very active audience. We always have a very engaged audience here for webinars. And so I'll hand it over to Mordy and we can have a little bit of a discussion.
Mordy Oberstein 39:51
So before I get into the questions plus one for The SEO Weekly with Garrett Sussman. Garrett is amazing, a fan favorite, so definitely check it out as a great way to keep up with what's going on in the wide world of SEO. There are a gazillion questions. I don't know where to start. Let me piggyback on one of the later points that Mike was actually talking about information gain. What about a scenario where and this is a common question around just creating content with AI, where it's already been done before a million times over? Can I use AI for that? And where's the added benefit that I'm going to be able to add? If the topic's already been covered before a 100 times over, I'm only covering it because I need to cover as part of my corpus of content. How does that work?
Mike King 40:41
Yeah, I actually kind of disagree with that idea that just because it's been, you know, kind of comprehensively covered that there's nothing new to say, as an example, me and Ross just talked about the same thing. We both had different takes on it, right? Like, I think it really comes down to the level of expertise. But what I generally will say to people, is that you can take a look at the entity graph and figure out what's related there. And then based on that, talk about something new, right? Like, you can use a tool like it's a tree, or even looking at Wiki data, or whatever, and then put in your topic and then see what else is relevant to it, and then figure out how you can work in something related to that, to whatever is related to it. So as an example, let's say I'm talking about vinyl records, right? You know, there are different types of vinyl records. If I was just going to talk about vinyl records in general, you know, you're just saying like, a record player and listening to music and so on. But then you can see, okay, well, the 78s, there's 33s and 45s. Like, there are different elements to that and then when you keep traversing the entity graph, there are more things to talk about that you may not be aware of. That's generally where I tell people to start if they just can't find an expert to talk about the subject.
Crystal Carter 41:56
And I think if we have a few beginners who are with us who are unaware of what an entity is, it's essentially, the way I normally describe it is it's pretty much a noun, or something that Google knows exists in real life online, but it's essentially essentially a noun. So something that has a Wikipedia page Disneyland, or wherever, these are things that are considered entities. And when you get knowledge graph that says that this is Disneyland and it's here and it's founde there, blah blah blah.
Ross Hudgens 42:31
Yeah, love that. We have a concept we tell our team to use called Vines, it's sort of like how to think outside the box different areas to look to, news experts, social, images, video, did that backwards, but I dropped the link in the chat. That's some different areas to kind of like think through and yeah, potentially find some additional entities as well. Right.
Mordy Oberstein 42:50
And that's one of the things a little bit dangerous about one method that people do to analyze the results pages. Let's see, okay, what's ranking for the keyword? And you can only tell what Google's already ranking, but you can't tell what doesn't exist yet that Google would rank should it exist. Let's look at a more general question. How is AI going to improve? And do you think it will ever get to a point where it'll be consistently updated? I think what that question meant was ChatGPT, the data set has ended in 2021. I think it may have updated it. But will it ever be to a point where the data set that the AI is being trained on will be in real time? And if so what does that mean for the future of the web?
Mike King 43:34
Yeah, I definitely think it's going to improve. I mean, that's just like the nature of technology, right? Like, and so many people are focused on this, like, every day, there's some new tool that people are building, like, I can't even keep up. But as far as like, you know, will it ever be trained in real time? I don't know how realistic that is, as far as like, you know, in the near future, because the way training works is like, distributed across a tonne of different websites, and obviously, not websites, computers, and then you've got to, like, you know, curate the data to some degree. But at the same time, I don't think it matters as long as the base layer of the language model is good, which it very much is and you're seeing that they're doing what's called retrieval augmented generation where they're pulling in results from search and then using that to inform what the answer is. So you don't necessarily need it to have real time data. You just need it to be able to pull in stuff when it doesn't know if it has something accurate.
Ross Hudgens 44:31
I know new plugins are available that you can add crawl ChatGPT that allow you to crawl the web to that direct URL. I believe I don't have access to that yet but.
Mike King 44:43
Yeah, AIPRM does that.
Crystal Carter 44:45
And then Bing's new thing or whatever they ground ChatGPT in their chatbot with Bing result results. I recently tested it to ask it if Zedaya had gone to the Oscars. And there was a photo that was circulating of Zendaya at the Oscars and she was not, she was in London with her boyfriend. And yeah, Bing backed with ChatGPT said no, she was not in at the Oscars. So that's something that does seem to be working so far.
Mordy Oberstein 45:18
Sort of, I've had a lot of experiences where it's if the news is very, very, very current, it's still a little bit stale. So be careful. Check the citations. This brings us to the next question. What will be the future of organic traffic? Should AI become a prominent feature in search engines? Such as it already is in Bing? Yeah, I think that landmine.
Mike King 45:44
I mean, we don't really know. Like, I think the reality is that people prefer answers. But at the same time, I think what we saw with the feature snippets is that it displaces the search volume. And what I mean by that is like, let's say you used to just search for, you know, Eiffel Tower, right? And you would go to some random page and read facts about the Eiffel Tower. But now you're getting those facts right there in the SERP. So it's not that you like, close Google, you're feeling like okay, well, what's my next question? And so they get sent somewhere else? And so I think the issue is that there may be situations where the Chatbot doesn't give a good enough answer. And then in that situation, people are going to go back to search. But also I don't know that, you know, this, this mode is going to be what everyone prefers all the time. Because you don't always want to have a conversation. Sometimes you just want an answer.
Mordy Oberstein 46:39
Yeah, I always thought that was interesting. Because to me, the novel of the AI experience is the chat itself, not the answer. I can get a featured snippet. I have a direct answer. I do think that it's interesting; I have a controversial take on this, I guess. I think that things like Peters did with direct answers, I guess now the AI chat experience, they incentivize now the secondary layer of knowledge. So the first the first layer of knowledge is basically handled by whether it's a feature snippet, whether it's an AI chat experience, whatever it is, and it's really the secondary layer of content that needs to be handled on an actual website, which means that the incentive to create content should be a little bit more specific. But that's just my personal controversial take on the lack of incentive to write top level content in 2023. Okay, who was the author? If AI writes content? Who is the author?
I think that's a good question for Ross.
Ross Hudgens 47:39
I mean some of my presentation was actually arguing, don't have it write that much content, you're sort of the base layer, in some ways it might be rounding you out. And if someone's contributing 20% 30%, I wouldn't probably call them an author in that case. I think that's probably the most powerful use. Yeah.
Mordy Oberstein 47:59
That was the beauty of your presentation, I think. Like if you if you look at AI as being open, and you just give it a prompt and go and you have no borders around it and you don't give it a controlled environment: disaster. And impersonal. But if you give it constraints, and you give it delineation, you give it a context to work out of then it's an amazing tool. Okay. Do you think there are going to be people who are going to specifically seek out brands who write human content? I think that question was more about do you think that there will be cases where brands will up their level of content to distinguish themselves from other brands that are predominantly using AI? And will people be able to recognise that and seek that out?
Mike King 48:47
I think that's definitely going to happen as far as people can. As far as people being able to recognise it, I don't know. Because the way that these things work, is all in like mimicking how we write. And even if even once people start creating other new, interesting, original stuff, then it just gets fed back in and it learns again, and then copies how we write. So I don't know that it's going to be the sort of thing where people can naturally detect it and I think if we think about Cambridge Analytica and how all that went down, and how people couldn't discern the difference between like real social posts and fake ones, it tells you everything you need to know as far as that, but I do think that people are going to start saying things like, yeah, this was generated with no AI as an example on 60 minutes last night, at the end of the segment about AI with Google, they were like, yeah, just so you know, this wasn't made with AI and I think we're gonna see a lot more of that and people are going to embrace that.
Crystal Carter 49:45
I think the tricky thing with it is like, where do you draw the line because I use Grammarly, for instance, and they're adding a lot of this stuff into into Google Docs, they're adding it into into Microsoft Word for instance. So it if that helped you, does that mean that you had some AI? Similarly, I very often will dictate, and that is using voice chat or voice recognition to create the text. So I think that can be a bit tricky. I'm not sure if Ross has an opinion.
Ross Hudgens 50:16
Yeah, I mean, I generally agree with all that it's sort of up to us to guide it. I think today, you could go to content sites and see very bad SEO content. And you just feel that and it's sort of getting to that best result thing. If that's done with a lot of AI content in there, I don't think people are gonna know the difference. But if you go to someone who's misapplied using the tool, it's a lot of junk it'll just read just like that same kind of bad SEO content today. So as long as they're achieving that goal, I don't think it should matter too much.
Mordy Oberstein 50:51
Yeah, let's point out the state of content now isn't great already.
Crystal Carter 50:57
I think also what you were saying about the different prompts and how you have people who are prompt engineers, like there is definitely a skill in writing a prompt. If you ever, if you log into MIT journey, you can see like, if I log into MIT journey and say, oh, I want to see a dog riding a skateboard, and the picture comes that awful, terrible, and I see other people who are making you know, Rembrandt, you know, the Sistine Chapel over here. And I'm like, how did you do that? And they've got these really complicated, really complex prompts. So I think you're right that you know, you both talked about the value of the input. There's the traditional adage for AI garbage in, garbage out. And it's very important that you that you finetune that
Mordy Oberstein 51:36
Ross, this one's right up your alley, would you repurpose content with ChatGPT?
Ross Hudgens 51:44
I think so I mean, that social media example I shared is a relatively reasonable example of that, where you're using a product photo that lives on your website, you're now asking it to create social media copy, I think that's a reasonable way to do that, where essentially, you're still using that base layer of best in class thing, which is the image and then supporting it in some way where I know I haven't tested it. I didn't do with my own presentation, but apparently you could have it take a blog post and make a presentation out of it. I think that'd be a powerful thing to do.
Mordy Oberstein 52:13
Someone pointed out in the chat that in your example, the AI even added in the emojis.
Yeah, exactly. They got it.
Nailed it. What's going to happen with Google, if there's going to be a plethora of mediocre content coming into into the ecosystem? A deluge of it, an enormous amount of, maybe not spam content, but what's basically out there right now, but a lot more of it. How will Google respond?
Mike King 52:46
Yeah, I mean, I think it's what I was saying before, you know, that's where the E-A-T thing comes into play. Andyou can spin up all the content you want but if you don't have the authority, it kind of stops it right there. And then you can go into the longtail of course, but you know, less people see that. So they probably don't care as much. But at the same time, there's gonna be a lot more bad content and a lot more good content as well. And so as a result of that, all of those signals that Google already uses is going to make the good content continue to perform better. So I don't think that Google is going to be under attack, because they've getting millions more pages. But I think that they are already mitigating that with E-A-T, the helpful content update, and so on.
Mordy Oberstein 53:34
Yeah that makes a lot of sense. If you look at the site, overall, the authority of the website overall, they can help you with the various pages. So again, if you're competing with two pages that are competing against each other that are relatively similar, if one website is far more authoritative, and far more topical, that's another bit of a boost for that particular page, and also, what Google can do. In terms of his language profiling, I personally believe that it's able to profile language in order to understand whether or not you have a first hand experience with the product that you're reviewing. It's not far-fetched to realize that Google can profile language or to say, hey, this is closer to what language would look like if you actually use the product versus if you're just randomly spun it up with AI. I feel like I have to ask this question. And I'm almost dreading it. But I'm gonna ask it anyway. Because someone asked it, and we're just gonna do it. Will AI gain consciousness? I have strong feelings about this.
Ross Hudgens 54:37
I'm not the expert for that question. Will it achieve AGI I think is the right word? Probably. Mike's way more technical than me.
Mike King 54:47
I definitely think AGI is realistic. You know, probably not as soon as everyone's predicting like, you know, there are tool or libraries out right now called like, Baby AGI like I think we're getting ahead of ourselves. But, you know, I think it's really a function more of computing power, which we have a tonne of, and we're just getting more as we go. So I think that it's possible. But I think that at some point, our regulators are going to step in and slow things down. Like the EU is already trying to ban ChatGPT. So you know, I think it's possible, it's just will we see it anytime soon is a different story.
Mordy Oberstein 55:25
Fun Mordy fact is that I'm a little bit of a philosophy nerd. And I think we're talking about human consciousness. I don't personally, and this is just my honest opinion, I don't think that's possible. I think the human human mind that persona is ineffable and you can't qualify it, and therefore you can't pass it on. And generally speaking, that which is the created thing can never surpass the creator of it. So personally, I don't think it's possible.
Mike King 55:59
I think there's flaws in your argument
Mordy Oberstein 56:02
We should have a conversation about this at another time. I'll pull out my Aristotle and my Kierkegaard and we'll have a long conversation about this. But I saved it for last, so we wouldn't be able to have a philosophical conversation on our webinar. So I think we're out of time. Sorry, Mike.
Crystal Carter 56:27
Alright. Well, thank you to everyone. Thank you Mordy for bringing the philosophy to the discussion. Thank you, to Mike for bringing in the algebra. I wasn't expecting to have to remember my vectors. I was like, oh, snap, we're getting mathematical over here. Okay. And thank you to Ross for bringing so many practical examples. It's been a fantastic session. Thank you to everyone for being really involved in the chat and for sharing your information. Again, it's going to be online, it will be on YouTube, it will be on our website. And we'll be sharing it on social media as well. So if you missed anything, please go back. Rewatch slow it down if you need to pause it, rewind all of that stuff. Thank you very much for joining us. Our next webinar will be next month. We'll be talking about Google Search Console with Daniel Waisberg from Google and it will be fantastic. So I hope to see you there. Thank you both. Thank you all and good evening. Good afternoon. Goodnight.
Mordy Oberstein 57:25
Oh wait, don't forget to sign up for our newsletters Searchlight over on the Wix SEO hub. Hey, marketers gonna market. Thanks, everyone.
Ross Hudgens 57:33
Thanks, everybody.