ChatGPT has created a new generation of transformers that have the ability to review, assist and transform your page exponentially. With Google’s review AI update, there have been vast improvements.
In today’s episode, we will gain insight into how ChatGPT utilizes transformers to classify a page based on its quality. We’ll also learn how to reverse engineer it and use ChatGPT to get a better understanding of review pages.
Here’s What We Will Cover Today:
Reverse Engineering via ChatGPT
Review AI Update
Check Pages For Maximum Quality
Links:
Find this podcast here:https://www.youtube.com/live/LiQOScT7QiE?feature=share
Newsletter:
Where to find Josh: joshbachynski@gmail.com
Where to find Greg: Support@handsoffpublishing.org
Referenced:
KeywordSpy: www.trykeywordspy.com
(0.00- 0.35)
Josh: Welcome to the SEO + AI show. My name is Josh Bachynski, and with me, I have my co-host, Greg.
Greg: Hey, Josh.
Josh: If you've never seen the SEO + AI show before, we talk about all the advancements in SEO and AI. All right, so today we want to talk about the review update.
(0.38- 0.50)
Google has recently put out many updates–all of them artificial intelligence based. All of them are unparalleled, none the least of which is the review update that Google has just released.
Now, Greg will attest to this in Underground SEO University, where we reverse engineer Google's algorithms and we test their AIs.
We figure out exactly what's going on. When they put out the product review AI a few years ago, I said, “This is not just going to be for products.” Greg can attest if this is a truth or a lie.
(0.53- 1.30)
Greg could either throw me on the bus here. Years ago, I said, this is not just going to be for products.
They're going to roll this out for services, for affiliate pages, for any kind of page where you're reviewing something. That could be very broadly applied. It's not just affiliate pages, any kind of page where you're reviewing something.
Did I not say that, Greg?
Greg: You did say it. I remember when you said it, because I'm in a nonproduct niche, and I thought that this is going to pertain to me so maybe I should listen to Josh here.
(1.31- 2.10)
So, I made the adjustments back then and did not get hit by review updates. That’s one of the benefits of being in the mentorship.
Josh: He even gave me a little plug for the group. Nice, Greg. But yes, I did say this, and it is good to be at Underground SEO University. If you want to know more about Underground SEO University, email me at joshbachynski@gmail.com.
(2.11- 2.34)
But how did I know that? The question is how, Josh? How did you know this? Is it because you've been in SEO for 25 years before Google was in it? Not really.
Is it because you've been reverse engineering their AIs for decades and they didn't have AIs two decades ago? Well, they did machine learning, but you've been reverse engineering their algorithms for over two decades, and you've been reverse engineering their AIs for the last ten to five years? No, not that either. It's because I've been using OpenAI's GPT.
(2.35- 3.09)
GPT is a transformer, and the phrase transformer was first made by Google. Google first made the transformer when they made Bert. They're bi-directional encoder representations from the transformer.
A transformer basically takes all the grammar of a language, encodes it into math, and then, because they've now got it into math, they can decode it into new sentences. They take sentences, they get a mathematical representation of it, and then they can decode it, and then they can encode it into new sentences. That's all a transformer does.
And that's what Chat GPT does. Chat GPT is based on GPT. I was on OpenAI's GPT way back when they opened the demo.
(3.10- 3.39)
I helped build chatGPT. They had machine-learned it off of some of my hundreds and hundreds of prompts, plus many other users, not just myself, of course. I knew exactly what a transformer can do.
It's easy. You just ask the transformer, “Hey, can you look at this content and tell me if it's a good product review? Does it have this? Does it have that? Does it have this? Does it have that?” That is something that a transformer is very good at doing. It is the perfect measuring stick.
(3.40- 4.19)
It is the perfect amalgamation of everybody's opinion. If you’re trying to get GPT or a transformer to give you specific knowledge or specialist knowledge, it can’t do that.
It can't tell you what stocks are going to be good next year. It can't tell you how to do SEO. It can't tell you how to do surgery.
It can't tell you how to do high-level programming. They've trained it on quite a bit of a corpus there. But if anyone has tried to program something in ChatGPT, they know that it's not perfect. You can’t get it to do high-level expertise work because there’s just not enough information in its corpus to do it.
(4.20- 4.50)
There are different kinds of reasoning and thinking it needs to do that ChatGPT just can’t do out of the box, right? If you need GPT or any transformer to give you a consolidation of everyone's opinion as to what may be spammy or not spammy, what is quality or not quality, what is good or what is bad, it can do that in spades all day long. I knew when I saw the product review update that, obviously they have a transformer doing this. It was probably Burt at the time, or Palm T5 is running the page through the transformer.
(4.51- 5.30)
The transformer is reading the whole page, and it’s giving a grade, idea, or classification. It's giving a classification of whether this page is of high quality or low quality, whether it has helpfulness or not, whether it has the experience, expertise, authority, and trust, and whether or not it's a good review based on certain criteria.
Now that we know how they’re doing it and they are definitely doing it with a transformer. I’m going to show you today how you can reverse engineer it and use ChatGPT to look at the huge list of helpfulness, experience, expertise, authority, and trust that Google gives us and the new review that Google gives us.
(5.31- 6.11)
I’ll modify it on the fly because I had already built it for something else. I’ll show you how in ChatGPT and you can take a look at a page or document and it will give you back this information. So let's go here to chat GPT.
Now, which model would be best to do this? I think the way I built this was by using the Browsing Alpha model. Now, not all of you might have access to the Browsing Alpha model and not all of you may have access to GPT4. You probably have access to the default 3.5 Turbo. I can modify it, later on, to show you how it would work with this, but just to make it easy, I'll probably do it here.
(6.12- 6.40)
So here's what I wrote, and this is what I wrote for helpfulness. Then we're going to add the review to this. This is how you would want to build a prompt like this.
Before I get into this, I would like to point out that this is all built into our tool, KeywordSpy. Keyword Spy asks all these questions in various different ways and builds all the content based on these questions. And if you're like, what are these questions, Josh? These are all the questions that Google asks you.
(6.43- 7.18)
If you don't know what KeywordSpy is, literally a ten-second plug. Keyword Spy is our on-page tool that Greg and I have engineered. It is completely a next-generation AI-based tool.
It is as smart as an SEO consultant. It will give you SEO consultant decisions based on my 25 years of experience and five years of reverse engineering AI in general, and 20 years of reverse engineering Google's algorithms in general. It is the best keyword-on-page tool on the planet.
You could try it free for two weeks, go to trykewordspy.com.
(7.19- 7.54)
It has a lot of things built into it, not the least of which is all of these questions from Google in various ways. We use this in various ways. These questions are all directly from Google.
Is it self-evident to your visitors who authored your content? Are you using extensive automation to produce it? The high-quality product review questions demonstrate your knowledge role in the products. This is what changed here.
It's the product section that changed recently with the review update. This will be the section that I’ll be changing. We’re going to plug it into ChatGPT and give it to a web page and see how ChatGPT thinks–that’s how high quality that page is.
(7.55- 8.26)
As you can see here, I've made it for the browse model. Now you might be like,” Josh, I didn't know you could go with Chat GPT Turbo and go to a web page and do that.”
You can’t. You have to have the browsing alpha to do it. So there's the top of the page to cover section and there’s a web page that you have to give to access it. You could rewrite this prompt to not do this.
And you just have to break it into smaller chunks. I want to do it all at once. This is what I want to do.
(8.27- 9.33)
I’m going to go to the product and find where it has high-quality review page suggestions.
It says, “Can you say yes to suggestions right on the 110 scale? Did the page do this? Evaluate the product or service and provide the review from a user's perspective. Demonstrate your knowledge about the products or services the subject reviewed.
Show you an expert. Provide evidence such as visuals, audio, or other lengths of your own experience with the subject to support your expertise and reinforce the authenticity of your review. Share quantitative measurements about how the subject measures up in various categories of performance, product, or service if applicable.”
(9.34- 10.33)
“Explain what sets the product or service or subject reviewed apart from its competitors. Cover comparable products or services to consider or explain which product or service might be best for certain users or circumstances.
Discuss the benefits and drawbacks of a particular of particular subject reviewed based on your own original research. Useful pros and cons described how the subject reviewed has evolved from previous iterations to provide improvements, address issues, and otherwise help users make a purchase decision. Identify key decision-making factors for the subject's category and how the subject performs in those areas.”
(10.34- 11.26)
“For example, a car review might determine that fuel economy, safety, and handling are key decision-making factors and rate performance in those areas. Describe key choices in how the subject reviewed has been designed and their effect on the users beyond what the manufacturer says include links to other useful resources beyond what the manufacturer or original sources or other experts say. Include links to other useful resources your own or from other sites to help a reader make a decision.
Consider including links to multiple sellers. If applicable, consider including links to multiple sellers who give the reader the option to purchase from their merchant of choice where applicable, or to other potential pages on your site. This is from the new review thing, which we're going to go over in a second.”
(11.27- 12.07)
“When recommending the subject that is reviewed on this page as the best overall or the best for a certain purpose, include why you consider this the best. I'll say product, service which means the best with firsthand supporting evidence.”
“Ensure there's enough useful content in your ranked list for them to stand on their own, even if you choose to write separate, in-depth single reviews for each recommended product or service.”
(12.08- 12.38)
So now we've changed that over. Now what I want to do is I want to quickly dip into the review I did.
It says you want to have an expert staff member or a merchant, right? So write high-quality reviews. So I'm going to go here and find the original Google. “Write high-quality reviews, Google.”
Now, here's how they changed it. That was from the old one, which I'm going to maintain. I'm going to retain it because none of it can hurt asking ChatGPT to rate content in those various different ways.
(12.39- 13.27)
Now what we're going to do is we're going to cross all of our T's and dot all of our I's, and we're going to make sure it has this information in it as well, which is considerably smaller and interesting. Greg, they've shortened it from the last one. Some of them are still there, but some of them are not.
It says, “Publishing high-quality reviews can help people learn about considerations product services, destinations, games, movies, and other topics.” I’m going to go onto ChatGPT. “Is this review from a reputable person who talks a lot and is an expert in this topic?”
(13.28- 14.02)
That's what I get here when I read this. They want an editorial staff member, an expert staff member, or a blogger. They say you can be a merchant but they don’t say an affiliate site.
“Does this seem like a spammy affiliate site? Is it from a merchant or a reputable merchant?” We asked it to evaluate from the user perspective already. We told it to demonstrate that you are knowledgeable about this. We said to provide evidence. We said to share quantitative measurements. We explained what sets it apart from the competitors.
(14.03- 14.36)
We told it to cover comparable things. We told it to discuss the benefits and drawbacks. We told it to do this. We told it to describe key choices or to rate whether our page is doing that. We told it to include links.
“Does this page include links to other resources, their own, or from other sites? Make a decision.”
We did links to multiple sellers. We told it to generate useful content in your right list. Okay, we've done all that.
(14.37- 15.12)
So the only thing we really needed to add was this. That was the only thing that was missing. All right, so now we have the prompt built.
It is 1400 words. That is just under the 2000-word limit that ChatGPT will allow. And I could do it by giving it the web page.
So if you want me to look at one of your pages live, put it in the chat right now, and I will tell you whether or not ChatGPT can rate it as to whether or not it has helpfulness expertise, authority, experience, trust, whether it is helpful, whether it has the proper review statistics.
(15.13- 15.50)
That’s because there's going to be an AI. Here's the point.
There's going to be an AI that looks for all of this. There is an AI that is looking for these specific words that you need to say. It could be on any page. “I have a free philosophy course from myself that I offer.” So I'm going to take this page and I'm going to ask it how well this page adds up. It's very short.
It's to the point. I'm going to ask it what it thinks about this page, whether my page has quality, expertise, authority, and trust. So we're going to go boom here.
(15.51- 16.32)
“Topic of the page: to cover free philosophy courses.” Now we're going to go here and see if is this going to work
Is it too long? All right, it's browsing the web. It's going to the page. Let's see if this is going to work.
If you have a page you want me to do, guys put it in the chat and we will take a look at it. Now, sometimes this browsing gets caught up. Oh, it's read my page and it says, based on the content provided on the web page, free philosophy courses study.
(16.33- 17.02)
Here's a rating for each of the question suggestions on a scale of one to ten, followed by a total percentage score.
“Content Quality Questions: Does the content provide original information, reporting, research, or analysis? Score six out of ten. Does the content provide a substantial, complete, or comprehensive description of the topic? Eight out of ten. Does the content provide insightful, analysis or interesting information that is beyond the obvious? Score five out of ten.”
“If the content draws on other sources, simply avoid copying and rewriting them. Six out of ten.” It thinks my page is a little light, which is not surprising.
(17.03- 17.44)
It is a little light. “Does the main heading or page title provide a descriptive, helpful summary of the content?” Score ten out of ten. Yes, it does.
Look at my H1 on the website. So you see, this is not just making things up.
It is giving you an actual, real rating of what it thinks is going on based on the opinions of all the SEOs out there. “Does the main heading or page title avoid exaggerating or being shocking? Nine out of ten. Is this the sort of page you'd want to bookmark, share a friend with, or recommend? Eight out of ten. Would you expect to see this content in or referenced by a printed magazine? Not really. Seven out of ten.” See, that's a bit of a lower score.
Does the content provide substantial value when compared to other pages and results? Seven out of ten.
(17.45- 18.14)
“Expertise Questions: Does the content present information in a way that you want to trust it? It says nine out of ten based on my authorship because it knows who I am. If you ask Chat GPT who I am, it knows who I am.
“If someone researched the site producing the content, would they come away with the impression that is well-trusted or widely recognized as an authority? Eight out of ten.” Yes a little bit higher. “Is the content written by an expert or enthusiast who knows the topic well? Yes, nine out of ten. It knows who I am.
(18.15- 18.33)
“Does the content have any easily verified factual errors? Nine out of ten.” No factual errors are evident, but because I digress and I change how I talk about philosophy in the way the rest of the industry does, you notice it didn’t give me a ten out of ten. It knows the general opinion of general information.
“Presentation of Production Questions: Does the content have any spelling or stylistic issues? No. Is the content produced well or does it appear sloppy or hastily produced? Eight out of ten.” It is a little short.
(18.34- 19.02)
“Is the content mass-produced by or outsourced to a large number of creators? Nine out of ten. It appears to be unique and not mass-produced. Does the content have an excessive amount of ads that distract or interfere to make content? No, ten out of ten. No ads present”
It knows what's going on here. “Does the content display well for mobile devices when viewed on them? Score eight out of ten. Focus on people first content it decided to collapse at 19 to 37. An average score is eight out of ten assuming the page aligns with most of these guidelines.”
(19.03- 19.36)
It can just summarize it for you high-quality review page suggestions 38 to 53. Multiple suggestions average a score of seven out of ten assuming the page meets some of these suggestions but not all you could also tell it to drill down a bit more.
“Total Percentage Score: The average score across all question suggestions is eight out of ten. Therefore, the total percentage is 80%. Please note that some of the assumptions were made in scoring the question suggestions. Due to limited information available on the web page. Additionally, some questions were grouped together in an average score, as they were repetitive or similar in nature.”
(19.37- 20.04)
As with all assessments, the evaluation is subjective, maybe based on the perspective of the reviewer. And it says, “It's important to note that certain questions were not fully applicable to the specific type of content on the page.
For example, it wasn't review-related suggestions for a non-review page. Therefore, this assessment serves as a general guideline rather than a definitive evaluation.” I think they're being weak on themselves.
I think it's a little bit better than a general guideline. And I would agree entirely that I shouldn't get any more than an eight out of ten. Yes, Google knows I'm all about ethics.
(20.05- 20.35)
I've been talking about ethics for almost as long as I've been talking about SEO. But is my page thin? For sure. So what do you think about that, Greg? I think it is pretty impressive.
Greg: Yeah, I think it is. The challenge to do this is obviously most of us don't have access to this browser feature, and so we have to copy and paste these articles in. And I've tried it myself, and it's a pain.
There are a lot of questions you're asking it, plus the article itself. It would be nice if there is a way to do this, and I think KeywordSpy is addressing that.
(20.36- 21:23)
I've run it on a few of my pages, and it's given me insights on what to key in on to improve. I’ve seen a boost and I really like it.
Josh: I would take it seriously. When it gives me a seven out of ten, that's a low review. And some of the ones up here were even worse.
It was giving me occasional ratings like six out of ten here in some places, or five out of ten.
Greg: I'm aiming for eight. If it's below eight, basically, I'm looking at that and trying to improve it. Yes. Seven, six, five. All those.
(21.24- 22.01)
Occasionally, you'll see, like, zero out of ten, four out of ten, but not too often. If it’s five, six, or seven, definitely key in those and try to improve that.
Josh: Yeah, I would agree. I do not trust ChatGPT when I tell it to just write content out of the box. I do not trust ChatGPT when I tell it to tell me how to do SEO, because those are specialist things that you need a specialist to do.
But if you're trying to get a litmus test as to the general opinion of the herd. The amalgamation of everyone's common opinion as to what is quality or not, or spammy or not, this is the perfect tool to do with it. It is absolutely engineered for this. It is absolutely perfect and truthful for this.
(22.05- 22.34)
Regarding what you said about browsing, you're absolutely right. I had to use Browse to do it. But the alpha will be released to everybody eventually. This is their killer app.
This is what they're going to compete with Google with. This is going to be their search agent.
This is going to be their autoGPT. This is going to be their AGI eventually. So this will be released to everyone eventually. You're right, you'd have to do very complex token management.
(22.35- 23.01)
You’d have to put in the page, limit the page, or analyze the page in certain ways with sections of questions. As many questions as you can fill in, and you'd have to record a score and have it programmatically working. This is what we're going to be building in KeywordSpy.
Greg: Yeah, I'd be curious what folks think of another dial or meter. Whatever you want to call it. A grading system on top of the keyword grading that we have currently. We’re adding another one plus this HEAT, we’ll call it metric and love to hear what people think. It’s something I haven’t seen anyone addressing on the market.
(23.02- 23.51)
Josh: Yeah, I agree completely. I think this is awesome. And you know why it's awesome? Let me tell you, as human beings, we are usually terrible at marketing ourselves, looking at our own pages.
As John Mueller and Google have said, we're typically terrible at looking at our own works and being honest and fair with ourselves about how high-quality our pages are or how high-quality our pages are not. We look at our page every day and we're satisfied with it. It doesn't need to change.
(23.52- 24.19)
That's not the mentality you need to take. You need to take its never-good-enough mentality and always be making it better.
This is a way to get concrete examples and concrete specific things that you can work on that come directly from Google. You know, with the highest degree of certainty, without a single variable testing it, which we do in Underground SEO University, this is exactly what Google is looking for and this is exactly what I should be doing. It's very important to get that right.
(24.20- 25.01)
Greg: I think one question that most people have, and you brought it up in the beginning, is how is it similar to how Google is looking at the page? People are like this is ChatGPT. That's a Microsoft or an OpenAI or a Bing thing. The technologies are completely different. It’s not Google.
As you explained, it is because it’s a transformer base. I think that’s a key thing that you explain things very high level that sophisticated people need to understand. You should tone it down for people a little bit like me in a way that we all can just say, these are apples-to-apples we’re comparing here. This isn't a waste of time.
(25.03- 25.49)
This actually is how Google looks at this page. Yeah, I'm not putting you on the spot, you can do it at some other show or something, but I think it's definitely for guys like me who just are not as advanced in this AI world yet. We need a sort of ‘dumbed-down’ version if you will.
Josh: I appreciate you saying that, and I can address that right now. No, you're totally correct. A Transformer is a Transformer.
All Transformers operate the same. A Ford and a GMC drive exactly the same. They're manufactured, they're totally different technologies by different companies, but they drive exactly on the same roads.
(24.50- 26.37)
You can't say that for everything in SEO. That's why in KeywordSpy, we don't just use an off-the-shelf, open-source NLP to choose our keywords.
We actually build you your own model for Rank brain, Neural Matching, BERT, helpful content, and review content. Because those are the six seven, six AIS that actually do all of the on page, and we build you your own models for them. That's why the testing takes so long in keywords.
So in those cases, you need to have those proper entities, those proper keywords from the properly used models of the used AIS that Google is using for on-page.
(26.38- 27.04)
Not to also mention the Dom vector AI, whatever they call their machine learning on the Dom, on HTML, the knowledge graph that ties into the KGMID mediums that locals are using. All those are built into on-page because that also has to do with the HTML in the background as well.
However, in this case, we are apples to apples because we took the actual questions that Google wrote themselves. The way they are worded, I can tell this is exactly how they prompted the transformer.
(27.05- 27.44)
They're using a transformer to see if it's helpful or has a good review, to see if it has the expertise, experience, authority, and trust, the way the other consensus of experts do.
If you read the white paper, the patent on the Mountaintop Bilslowski is called the Mountaintop algorithm. They’re using Transformers and NLPS to do this. That’s exactly how you would prompt the transformer.
Does the content provide original information, reporting, research, or analysis? That's what they're asking their transformer. This is very close to what Google is doing. There is no other way to do it. This is how the transformer works.
(27.48- 28.04)
This is how you would get it to classify. It would be doing this or something very similar to it. That is because this is apples to apples, these are the actual questions that come from Google because it is a transformer and a transformer is a transformer.
A truck is a truck. Whether it's made by GMC or Ford, it doesn't matter. It still drives the same.
(28.05- 28.35)
Despite people saying a Ford is better or not, they're the same thing. A transformer is a transformer. They're going to classify it the same, and they're going to prompt it the same.
So that's why this is super important to use and super a good suggestion for your blind spots. I thought I was providing insightful analysis and interesting information that was beyond the obvious. I thought I was doing that, but it says I'm not. Why not? You can drill it down.
(28.36- 29.14)
You could say, “Why did you give me a five out of ten for number three?” I could say that. For example, “Why did you give me a five out of ten for number three? What exactly can I do to get better there?”
It will think about it for a second. It says, “For question three, does the content provide insightful analysis or interesting information that is beyond the obvious? I gave this score because while the web page does provide information about free philosophy courses and the instructors’ qualifications, the content appears to be more focused on providing an overview of promotional material for the free philosophy courses rather than offering in depth analysis or unique insights into the subject of philosophy.”
(29.15- 29.40)
Dead on correct. That's exactly what I'm doing on that page.
It says, “The information presented seems relatively basic and does not delve into specific philosophical concepts, ideas, or debates that would offer readers more insight into the field of philosophy. Totally right.
I made that as a conscious choice because it was a sales page to get people into the free philosophy course.
Greg: You're thinking, why would I water it down with all this info that people probably don't care about on a page like that?
(29.45- 30.16)
The reason why I would do it is because Google does. Google cares about it. So this is how you would structure it.
This is a great question. How would you structure a sales page in that regard? What I would do is that I would leave I know this page converts really well because I get a lot of people signing up for my free philosophy course.
So I wouldn't want to make this any longer. What I could do is I could make a floating call to action button to join the discord for free. What I would do is underneath this content right here before these links, I would put more content underneath.
(30.17- 30.45)
So you have the sign-up higher to the top and you have your SEO content underneath. That's the standard way you would make a sales page like this. As to the suggestions it gives me, these are not bad either.
It says, “Include in-depth analysis, share your unique perspective, and provide excerpts or samples.” What a great idea that I never thought of. “Highlight real-life applications where you can use philosophy.” You can use philosophy for different things. If it comes to improving content quality, ChatGPT is the bomb.
(30.46- 31.26)
ChatGPT is really good. It does an exceptional job at telling you to improve quality and what you could possibly do to make it better. That's the only thing in SEO it does really well out of the box.
For everything else, you have to have an expert such as myself or Greg to prompt it in a certain way to get a certain answer and give it the expertise. Do you guys have any questions about that? Let’s see what questions we got going on here.
(31.27- 31.59)
Carol says, “ Absolutely incredible. You can literally create a map to improve improvement without blind guessing or a zillion hours of research to figure it out.”
Yes, exactly. Carol, this is extraordinary. This is built into KeywordSpy. You can also use this at Underground SEO University. All you need to do to remake this prompt is just go and find all the questions from Google and put them in here.
If you want to do this with 3.5, though, you can't use the browsing alpha. It's a bit of a pain.
(32.01- 32.33)
As Greg said, you're going to need to toggle between them.
Greg: Yeah. I was going to bring up another point of how what this could possibly just lead into.
Number one, looking at a page like this is just not being done. People look at title tags, meta descriptions, headings the basics of a page from an SEO perspective. This is brand new in terms of how to evaluate pages. How do you stack up against your competition?
Is there some correlation between pages that are in the top ten on your page based on this actual score? That's something we'll be building in with the HEAT bar.
Josh: The HEAT dial, yes.
(32.34- 33.30)
Greg: You'll have a comparison of your competitors and what you have, and then you can really start benchmarking things and seeing what you need to do. That's the key to this. To do that manually would be a pain.
Doing your own pages is difficult. To bring it even further, can we even suggest how to fix it? I know Chat GPT can do it, but can we in the tool say, you have a five out of ten. Well, this is what you need to do.
Josh: Yeah, completely. The three lowest scored once. Then it will follow up like I'm doing down here. The prompt will follow up and explain why they got a five out of ten for number three and exactly what can they do to get better? I’ll do the same question for this one and it will drill down for all of them.
(33.31- 34.29)
It says, “For this question scored out of ten, six out of ten was given because Whipey's permanently serves as a promotional page for free philosophy courses offered by the instructor, Josh Bachynski.
While it does include information about the courses and the instructor's background and qualifications, the content appears to be more of an overview and invitation to join the course rather than providing original research or analysis on philosophical topics.
So in order to make it better, showcase original research. If the instructor or the course creators have conducted original research in the field of philosophy, which I have, consider highlighting key findings or contributions to the field.”
(34.30- 35.00)
“This can include summaries of research papers, studies, or unique philosophical theories developed by the instructor. Consider providing exclusive content such as original essays or articles written by the instructor.” Another great idea.
“Share course development insights. Provide commentary on current events” Oh, I would have never even thought of that. This is so good because the AI doesn’t lie.
The AI looks at it impartially and rates it against the opinion of everyone for which it was trained. It keeps you honest and makes you realise that you could do better. What you thought was a ten out of ten just doesn’t have the substantial value you thought it did.
(35.01- 35.43)
I guess that this is not good enough to appear in a magazine or a book. I'm not doing good enough quality content because read between the lines for this. This is right from Google.
This is Google saying, “This is the barrier of entry.” If you want to rank number one, you better be good enough to be in a printed magazine, encyclopedia, or book. If you’re not, the first competitor that does this is going to outrank you. It’s a race between you and your competitors.
(35.44- 36.35)
Steven says, “We're all just dust in the wind,” Okay, so let's try another page and see how well it goes.
Let me edit this any particular page you want to do, Greg?
Greg: Yeah, I'm going to send one over.
So, guys, whatever SEO questions you have, please put them in the chat. I'd be happy to answer them.
If you're watching live, and if you're not watching live, please subscribe and click the bell so that you know when we're going live. Or follow me on Twitter. I also shared on Twitter usually first, that we're going live at a certain time at JoshBachynski.twitter.com
(36.36- 37.02)
Here we got restaurants for LaGuardia. All right, here's the page, there's the topic, and away we go. So now it's browsing the web. Now it's going to go look at your page. And this is the future of ChatGPT. If you’re thinking this is a feature that you will never release or have access to.
No, you will definitely have access to this. This is their killer app. This is their autoGPT.
This is their AGI. This is their take on Google. They're not going to have a search engine.
(37.20- 37.46)
They're going to have a search agent that does searches for you and does all the research for you, like a personal secretary, and comes back and could think about it on their own a little bit. If you watched our autoGPT broadcast from last time, you're going to see kind of how it thinks about it.
So here we go. “Do you provide original information, reporting, and analysis? Seven out of ten. Does the content provide a substantial, complete, or comprehensive description of the topic? Eight out of ten. Six out of ten on insightful analysis or interesting information that is beyond the obvious.
(37.47- 38.13)
“If it draws on other sources, does it avoid simply copying or rewriting those sources? It says the content is based on the author's first-hand knowledge and experience. Does the main heading or page title provide a descriptive, helpful summary of the content? Nine out of ten.” Good.
That means your SEO titles are done. Well, of course, they are. “Does the main heading or page title avoid exaggerating or being shocking in nature?” Yes.
You notice this you don't want to be shocking in nature in the title that the AI is looking for, that’s something I am a little guilty of. I've gone and taught you to be very sensational in your meta description, in your title, and in your H1. That's something where I have to remember the AI is looking for this.
(38.14- 39.08)
They do think it is not book worthy. Seven out of ten for bookmarking it and seven out of ten for substantial value. There's a lack of information about the author. Six out of ten.
Greg: Yeah, I don't have an author. Yeah, I didn't add it yet for some context.I literally built this page with our methods using Generate article, with some prompt rewriting and an expertise prompt that I came up with just to try to let Google know that I'm giving a first-person real-world experience of each page. It's indexed and ranked within a week.
(39.09- 39.45)
So I was very impressed. In a brand new domain, it wasn't anything that had backlinks or anything like that.
Josh: Yeah, so that's a good point. The content improver in KeywordSpy I think is a very underutilized tool for people using. If you’re wondering what that is, let me show you here. In Content Editor, there's a Generate Article button that generates the article in one click.
This is based on 3.5 Turbo. Right now we're going to move it over to GPT4.
(39.46- 40.16)
It does a really good job of automatically generating everything for you. It'll do the URL, your title, and your meta description, you give it an angle. If you have a sales angle, you give perks, like we're doing 30% off right now.
You tell it where in the sales cycle you want to be, top of the funnel, middle, or bottom, and you say, give it the heading and it'll do all that for you. But the Content Improver button, if you highlight some content and improve it, it does really well in avoiding AI detection. Probably the best content writer on the planet for avoiding AI detection.
(40.17- 40.58)
As Greg said, he used the Improver on this and it indexed and ranked within a week. As you said, for a brand-new site, the cross-off is the Improver making it less SEO predictable. It's going to give you some great scores on this, like we saw here firsthand and knowledgeable, the main headings are done but it's not necessarily going to be the best in terms of hitting your content scores.
So it's an interesting discussion that we're going to continue to have on the show, but let's see what else you got here. Oh, it says there's a lack of information about the author. No information was provided about the site's reputation.
(40.59- 41.46)
The author claims they have first-hand knowledge and experience. It has no obvious factual errors, but it's starting to think it's kind of spammy. Eight out of ten.
Seven out of ten for amounts of ads that distract it says there's a presence of ads at the top. Is this correct? Do you have ads on that page?
Greg: Not yet. And it shouldn't? No. I don't think so.
Josh: So what does it think is an ad? It could be Cloudflare, I don't know.
And there could be some of the ways I built these sites.
Josh: Do you mind if I look at it?
Greg: They're like template based. Yeah.
(41.47- 42.23)
Josh: It thinks this is an ad. ChatGPT thinks this is an ad, Google probably doesn't, but it might think this is an ad.
This exit is an ad, one or the other. Next, I will say “Please continue”. Hopefully, it will continue. Thinking about it, this is a big long prompt. Yep, it got it right exactly. Just kept on going from where it was. We're getting some lower scores of sevens, but you generated this all auto magically with AI content, so that's not bad.
(42.24- 43.19)
Once you see what's ranking, then you could improve it. You could run this and prove it but plus this is going to be built into the content generation process in KeywordSpy soon so it'll generate both the depth, the spread, the URL, the title, the H1, the meta description and it'll content improve it to make it less predictable and it'll make it high HEAT score as well.
No other tool on the planet comes even close to doing even one of those things very well. We’re going to do them all in one fell swoop. So that's going to be awesome.
Greg: Yeah. I wouldn't suspect this page to be getting eight out of ten. There are some, but not overall. I think it's what to be expected the first time out. I'm just happy that it was indexed and ranked.
(43:20- 44.21)
It's every day gaining keywords. I don't know how many pages there are, but it's probably close to 50 or 60, maybe more. Every day there are more keywords getting added to the Google Search Council.
It's not here to sell or do anything like that. I'm just here to explain my experience with it, with the tool and the methods and it's looking really good. This is awesome from an ROI perspective, because when you're maybe traditionally doing this years ago, it's taking you weeks to write the content, maybe more researching keywords, what have you, when you can pump these things out at this quality, at this rate, the ROIs become really attractive. I see a lot of settings, so there's work to be done.
(44.22- 45.06)
Yeah, that's great. I'll just say.Please continue. So, one thing I was going to ask you about is for it not to repeat the question is an idea of just saying, here's a table, give me the number on the left one through 50 or whatever it is, and then the score on the second column and that's it. I think that may save a lot of tokens and probably get through this prompt even quicker.
Josh: Yeah, one could do that. The benefit of repeating all the questions really reinforces the gestalt that Google is looking for. The drawback is it's a lot of tokens. It takes a lot of time.
(45.07- 45.52)
If you guys have any SEO questions, please put them in the chat here. Steven says, “ChatGPT creates an AI named Skynet that is self-aware.”
Yeah, it's harder to do than just that. Trust me, I tried. Stone DAPE says, “Is the prompt same for product pages?” Yep, it's the same for product pages.
Inquirer says, “How good will it work for a very high competition keyword like web design service?” This will work really well for that. It's not the only thing you need to do, but it will work really well for that.
Steven says, “The Improver sounds impressive.” Yes, it is quite impressive. We kind of lucked on a secret hack of using Chat GPT to get it to generate human content in a way that really no AI content detector out there can detect. So it's quite interesting how we're doing that.
(45.53- 46.30)
All right, so you ended up with 71%, which is not surprising because I saw a lot of seven out of ten. So an insightful analysis, it's not book quality. It doesn't seem trustworthy.
There's no clear sourcing, there's no evidence of the expertise involved. Background about the author of the site that publishes it. There's no author link there.
If someone researched the site producing the content, would they come away with an impression? Nothing like that. There are no visuals, audio, or other links.
(46.31- 46.54)
You're an experience with the subject. There are no pictures from that place. There are no quantitative measurements about which restaurant is better and where you should go.
No table of that includes link to other useful resources and there are no links to other resources. Again, there's no expert information. This page does not include links.
Again, as I said before. So that would be the lowest things to work on.
(47.01-
Greg: I know this niche and I know what the top pages have and they definitely have that a lot better than I do. I built the page on purpose to be like this, so I'm not expecting this, but it's so correct. It's kind of scary.
Josh: Yeah, it's awesome. It's super awesome. All right, folks.
Well, I hope you enjoyed that. I hope you see how useful those things are. Uranium says, “Such a good strategy. Can't wait to try it.” Yeah. So if you want to try this out, definitely try KeywordSpy
It's going to be built in. And also when you get browse alpha, you could try it there as well. When Browse alpha comes out, email me.
(47.44- 48.32)
I could try sending you the prompt, or you can join Underground SEOUniversity and get the prompt there and far, far more far other prompts that we use and other underground techniques that we use for all kinds of SEO to really boost your SEO. Do you want to be ranking in a week or two? You can too, as Greg said, he just did it as kind of a lark as a test. You can get a ranking there as well.
Join Underground SEO University and or try keywords buy for free for two weeks. So thanks for listening, guys. Greg, do you have any final things you want to say?
Greg: No. Thanks, Josh. Appreciate it. This was helpful. And see you guys all next week.
Josh: All right, guys, see you next week. And I don't need to say good luck in the SERPs because with our knowledge, you can rank and you will rank.
Share this post