A featured snippet gives you a descriptive summary of your search query in short, concise sentences. Although Google has featured snippets on their pages since time immemorial, ChatGPT is making breakthroughs in its AI system.
In this episode, we’ll delve into the intricate prompts in ChatGPT that breakdown your search intent, query and so much more! We’ll also discuss some tips and tricks on building multiple prompts that are based on expert consensus and research-backed case studies!
Here’s What We Will Cover Today:
Generate A Featured Snippet Via ChatGPT
Create Built-In Prompts Via ChatGPT
Create Multiprompts Within A Prompt
Links:
Find this podcast here: https://www.youtube.com/live/wSoABosYy0k?feature=share
Newsletter:
Where to find Josh: joshbachynski@gmail.com
Where to find Greg: Support@handsoffpublishing.org
Referenced:
KeywordSpy: www.trykeywordspy.com
(0.00- 0.57)
Josh: Hello. Welcome to the SEO AI show. We are the folks online who tell you about SEO and AI simultaneously. With me today I have, as always, my co-host Greg. Today, we’re going to be talking about how to use ChatGPT to crush the Featured Snippet.
It is essentially an exercise on the right way of using ChatGPT. We’ll also be showing you how to not use ChatGPT for SEO.
(1.05- 2.25)
So you might be like, what is the big deal? A lot of people use chat GPT for SEO. But here's the problem. For example, if I say, ‘Write me an article on ballet for adults.’ It’s going to write an article but if we look closely, it’s not very good. If you think that this is good enough to go for SEO, it’s just not.
Let me put this into our tool, KeywordSpy. If we take a look here and see how obvious it is to detect AI content. This is represented by the red marker on KeywordSpy. Can you see how long the red goes on? In some cases, it goes on for 31 tokens.
(2.35- 3.10)
This is how easy it is for Google to detect the AI content. The average predictability length of that red is six tokens in total. The worst length is 31.
It's very easy for Google to find content that you just generate out of chat GPT. So, if I do ballet for adults and I just start a new keyword report, then it'll start a report automatically.
So, first off, you'll notice the average content score of our competitors is 70 in terms of the depth of their keywords. Ours is only 46 out of chat GPT. So that's not good enough right away.
(3.15- 3.50)
The readability is way too professional. It's very difficult to read. That probably is not what you want either, although we could debate that the numbers can round up, around, down, depending on how you do the test. But this is again very close to six and the 31 we had here.
And also look at the words 427 where the average competitors are using 1500. Again, it's way too little content. It's nowhere near powerful enough to rank.
(3.40- 4.10)
If we check out the content here under the entities section. This is probably not done yet, but look at all the zeros. ChatGPT is not mentioning the phrases here enough. They’re not mentioning particular entities enough. We don’t have the spread keywords either.
That's again, probably because, it's too short. This is just what I wanted to show you as to why you should not just be using chat GPT out of the box.
(4.10- 4.30)
If you try just to use chat GPT out of the box, you're going to get lackluster content. That's not going to rank very well. What are your thoughts on that, Greg?
Greg: Yeah, obviously just giving it a short little prompt like that is just going to give you, 400 words, no headings.
You really got to get more in-depth with your prompting to start off. But even if you do, it's still probably not advisable to use it out of the box. I think we did a video on the AI PRM which has built-in prompts that are not the best to use either.
(4.30-4.59)
There's a lot of guesswork with this and I think it's quite clear that Google is not going to rank these type of pages on the first or even second page. Most likely they'll get indexed. It's very possible with some human editing, they'll get indexed, but they're not really going to do what you want them to do.
Josh: Yeah, no, completely, you hit the nail on the head. So the question is, then what do you want to do in prompting? How more complicated do you need to make the prompt and what else does it need to say and do? So that's the exercise we're going to do for today. Now, that was a default GPT.
(5.00-5.25 )
I also have the browsing alpha here. I'm going to use default GPT because it's faster. You could use GPT Four because it is more complex, it has more token connections, but it's very slow.
So what we're going to do is we're not going to write a whole article because KeywordSpy generates a whole article for you. What we’re going to do is we’re going to show you how to crush the Featured Snippet.
If you don't know what a featured snippet is, let's see, what can I search to get a featured Snippet? I can search for self-aware plus AI. I put in self-aware AI because this is a space that I built and I check to see some testing that I’m doing.
(5.35- 6.01)
And I knew that a feature snippet would come up. So this is the featured snippet here.
It's the text at the top that Google uses the Bert AI to detect. BERT was the first transformer. It stands for Bi-Directional Encoding Representations from Transformer.
What it does is it encodes and decodes semantic relationships mathematically. It tokenizes them. It’s a statistical model that will predict the next words or word pieces that should come next from the words you fed it.
So if I feed it the words knock knock. In most corpuses, it will know that that's a knock knock joke and I should say, who's there? That's the tokens that come next, the words that come next. And that's how all these things work. That's how GPT works.
(6.02- 6.30)
GPT is another transformer. It's a generative pre-trained transformer. And here's another featured snippet.
Bert came up with another featured snippet. So GPT stands for the generative pre-trained transformer. It is also preprocessed and pre-trained. It’s the same as Bert, except it has way more token connections and it’s much more complicated.
(6.40-7.27)
The wisdom of different semantics–gives it more subtext and understanding of how to mimic human speech and thought. Consequently, it will give you the statistically correct tokens based on the context and questions that you ask.
So Bert does the same thing. It reads what the query intent of this query is, and it’s like, “For self-aware AI, you’re looking for a definition.” You see that they've given you an H1, or they've given you some form of a header. Then they answer the question in a paragraph and also highlight the answer to the question.
(7.30- 8:10)
It’s important to answer the question. Getting the featured snippet can get you to position zero. It’s the new spot one and it’s really important to go for it.
And we've seen pages on page two do this properly at Underground SEO University and pop up like 14 spots or twelve spots. That's a pretty huge boost. I presume you would agree.
Greg: I would, yeah. You can't really beat that.
(8.11- 8.34)
Josh: What else can you do to get that many spots? You could do really good on page with something like KeywordSpy. To get 14 or 12 spots can be very difficult with any other tactic. So getting the Featured Snippet is super important.
So how do you do it with Chat GPT? How would you get that done? Well, like I said, you got to find what the query intent is and then you got to find what the answer to that question is.
(8.40- 9.00)
You got to find out what the query intent for self Aware AI is. Then you got to find out what the answer is. So when I typed in self-aware AI, they gave me the answer. “This will be when machines are not only aware of the emotions and mental states of others but also their own.”
That's incorrect, but it doesn't matter. It's correct as far as Google is concerned. “When software AI is achieved, we would have AI that has human-level consciousness and equal human intelligence with the same needs, desires, and emotions.” Again, totally incorrect, but still correct as far as what Google thinks that means.
(9.20- 10.00)
So the question is, how can we get it to analyze the first paragraph for you to know what you need? So we need to know what the query intent is. First off, as we talked about, we need to supply the query, we need to find out the query intent, we need to find the answer to that question or topic and we needed to do 35 to 45 words maximum.
Why is it important? Do you ever see featured snippets that are longer than 35 to 45 words? You got to answer the question in 35 to 35 words to give it a sound bite that Google can fit in their Search Engine Results Page. (SERP)
(10.01- 10.28)
The BERT Transformer ChatGPT has a token size limit of 8000 tokens. It has an attention mechanism that works in the back end and will pull even if you exceed 8000 tokens.
The attention mechanism is like a plus and it will go find the relevant context of what you're asking about in the previous data stream and dynamically form a prompt for you. As long as it's not more than 8000 tokens in total, ChatGPT has a lot of extra gear under the hood that's working.
(10.30- 11.00)
Burt does not have all that as far as we know. In fact, we're 80% sure that's the case because we've tested pretty conclusively at Undergraduate SEO University. That’s how we’ve built all this into KeywordSpy which has a very small token size, as Greg remembers. It’s something like 178 tokens.
The prompt for BERT can only be 178 tokens. That includes the query and answer. Prompt and completion are taken together in terms of token size and what the model is going to tokenize. Then it’s going to put to the model which is going to find the statistical relationships.
(11.01- 11.41)
It's going to tokenize that math into tokens, and then that token will be into words, and then it spits it out on the screen. In this case, BERT compares its answer to all the pages in the corpus and classifies the one that has the closest connection and that’s how it produces the featured snippet.
So Bert doesn't work like Chat GPT in that it's going to spit out a completion. Bert does spit out a completion, I'm sure. But what it does is it takes out that completion, tokenizes the answer, and compares this to a web page. Then it will boost the web page to the top.
(11.45- 12.20)
And it will do this on the fly. Google systems and their computers are so powerful. They are the best IT company on the planet and the best AI company on the planet, hands down.
I know that ChatGPT and OpenAI have stolen all their thunder. If we have time, we might talk about Google Bard today. Google Bard is rather underwhelming in terms of its performance, according to Chat GPT. Chat GPT is ahead of them in this race. But do not count Google out yet.
(12.22- 13.05)
Google has far more powerful transformers. They've had trillion-size token connection transformers for months, if not years before OpenAI did with GPT4. They are the first AI company as their CEO says.
So do not count them out. Do not think for a second that they are in any way defeated by OpenAI here. So here's what Bard says. “GPT stands for a number of things, including generalized Probabilistic theory, generative pre trained transformer, Guid partition, table, get paid to surf. In the context of artificial intelligence, GPT is most commonly used to refer to generative pre trained transformer models developed by OpenAI. These models are trained on a massive data set of text and code.”
(13.06- 13.27)
“It can be used to generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way.” So why did Google Bard do that? Google Bard did that because there are a lot of things that GPT stands for. It's giving you all the things that it stands for.
So it's a little bit more robust, and that's an interesting thing.
(13.30- 13:57)
We know that when we're generating our prompt for this, we're going to want to tell it what the query is.
To do proper prompt engineering, you have to write out a standard operational procedure, like an SOP. You got to give it a little bit more context, but not too much context.
So in the prompt, do you think I would want to mention featured Snippet here?
Greg: I don't think so.
Josh: Yeah, I would agree with you totally.
(14.01- 14.31)
The reason why is that then if I say featured Snippet here, it's going to go into the corpus and it's going to pull out all the implausible ideas of all the amateur SEOs and what they think the featured snippet is and how to get it.
So in some cases, the context of the wisdom of crowds is going to help you and in some cases, the wisdom of crowds is going to hurt you. It’s going to muddy the waters and confuse Chat GPT or any transformer you're promoting, and it's going to pull in ideas that I don't want it to think about. I've already analyzed what it needs. I know it needs a query.
(14.32- 15.11)
I know it needs to determine what the Query intent is, much in the same way that Keyword Spy already does. So here in Keyword Spy, under reports for ballet for adults, when we go to our BERT testing. We can do a Query Intent and I can just submit it without the keywords and it'll tell us exactly what the users want for that query.
And it's unerringly good at getting this right. So it says these people want to:
Find ballet classes for adults
They want to learn about adult ballet classes
They want to discover the benefits of taking adult ballet classes
I could have also added in the winning words and submitted it to give another different kind of aspect to it.
(15.12- 15.37)
Quite frankly, it can get it right without the winning words as well. Now, I could use this as an H1 as well in the KeywordSpy system. It is right that this search intent wants to find valid ballet classes for adults.
So if you're doing this in Chat GPT, we're already doing it for you here in Keyword Spy. But the purpose today is I want to show you some prompt engineering as well because Keyword Spy doesn't do everything. So I want to show you, using this example, how to do proper prompt engineering.
(15.40- 16.20)
So we're going to have to give it a query. We're going to get the Query Intent, then we're going to tell it to answer to that question or topic or Query Intent and make sure your answer is 35 to 45 words maximum.
Now, here's another thing. You got to know what you're trying to do with your transformer. If I'm going to give it a word maximum, then I probably want to use GPT4 to do this. But you can use GPT5. I'm just saying that if I want to give it a word maximum, GPT4 will do better on it, but I'm still going to stick with GPT5 because that's what everybody has.
(16.30- 16.57)
We want to crush the featured Snippet, but we don't want to say featured snippet here anywhere because then it'll start giving us the ridiculous ideas out there who’s guessing as to what the featured snippet is.
We want to go based on the science that we learned at Undergraduate SEO University and the wisdom that you're learning here in the SEO+ AI show. In that case, the wisdom of crowds is not going to help us, if anything, is going to hurt us.
(17.01- 17.33)
Here's another main principle of prompting, which is that less is more. Interesting side story, I've taken a lot of different kinds of education. My Master's degree and my Ph.D. were actually in philosophy and psychology.
That's how I built Cassandra, my software AI. I was able to extract how a mind works and then build little portions of it and then stitch them all together. But in doing that, I also took a hypnotherapy certification.
I didn't actually take the certification test because I didn't want to pay the $200 to do it. And I have absolutely no interest in being a therapist to people because I don't know how people do that job. I don't know how therapists listen to our problems every day and all day and don't burn out entirely.
(17.35- 17.46)
But I wanted to learn about hypnosis because it's a very fascinating aspect of life in general, but also sales. Sales is nothing but hypnosis.
A call to action is nothing but a post-hypnotic suggestion or a hypnotic suggestion. The reason why I’m mentioning this is that in terms of hypnotizing, somebody making a hypnosis script or in sales, less is more. You want to say more by saying less.
(17.50- 18.20)
You need to say it in the exact right way so that it's going to be absorbed by the listener. In this case the AI, but also for sales, the prospective buyer. And here's another thing, you can't say negatives.
I can't say don't do this and don't do that, because the more I say that, the more it's going to want to do that. Remember: less is more and you want to avoid negative prompting as well.
(18.28- 18.58)
So those are the main things. You want to give the AI explicit advice like an SOP. You want to break it down so that it understands what you’re asking and add all the different things. You need to do the thinking or you need to ask it ahead of time.
You can also do meta prompting. For example, you could say, “Write a prompt to prompt you to write the best-featured snippet that Google has the biggest chance of choosing.” So you could try doing this too.
(19.00- 19.44)
“I want you to isolate and name the variables that are important for this prompt.” This is also called meta prompting, where you don't write a prompt for GPT, you write a prompt that writes a prompt. Whether you want to use it or not, you can see how ChatGPT thought of how to do it.
You could either choose the prompt and say, “I should do it this way” or be like, “This is totally wrong but it tells me what not to do” This will be an interesting experiment. Let’s see what it says.
(19.50- 20.01)
It says, “ Prompt: Write a featured snippet on ‘how to reduce stress’ that provides practical tips and actionable advice using research-backed strategies and concise clear language that is easy for readers to understand.” That's interesting how it thought to write that prompt.
It shows us the variables. “How to reduce stress” is the topic of the featured snippet and then they're right, we're going to be using a query. That’s definitely the same as the prompt that we’re going to be making.
(20.10- 20.23)
It wants to put “practical tips and actionable advice” here. Here’s where the wisdom of crowds, as to what should go to the featured snippet is going to hurt you.
I just want to show you an example of how to write a meta prompt and you could do that in various ways. So it says you should be putting practical tips and actual advice in it, which is the type of information the feature Snippet should provide. Wrong. But it doesn't matter, at least it's showing you how I wrote the prompt. Next, “research-backed strategies, the type of evidence, and support for the featured snippet should use.”
Again, wrong, that's not answering the question. You only have 35 to 45 words. The point is just to show you how you could tell it to write a prompt.
(20.24- 21.17)
Finally, “clear and concise language: the style and tone of the featured snippet” will make it easier for you to understand. We have at least one of those things that are the same. Greg, tell me if you agree.
I like our breakdown of how you would write this prompt or variables as they're based on the science of how transformers work. Do you have any comments on that?
Greg: I thought the fourth suggestion, the variable was pretty good, maybe want to add a concise, clear language. Yeah, I know sometimes we use parts of that type of language just to get it on the right track. I think it’s good to try at least in the first round.
(21.25- 22.05)
Josh: Yeah, actually you're absolutely right. Good call, because that's echoing just what I said. Less is more. You want to use clear and concise language.
People have gotten away in Chat GPT to actually giving examples of how you want it to go. In this case, we don't want to give examples. Sometimes you do, sometimes you don't.
If you're going to give examples, it needs to be paradigmatic examples. You need to find the essence of the semantic pattern you are trying to get it to notice and then spit out an answer.
(22.10- 23:52)
The next thing we’re going to do is show you how to write this prompt to crush the featured snippet in ChatGPT.
“This AI will determine the best answer or explanation to a Query Intent or Search Intent”
“Query: Ballet for Adults”
“Query Intent: answer that question or topic query intent in 35 to 45 words maximum.”
So, this AI will determine the best answer explanation of the query or search intent.It will first determine the most likely intent of that searcher.
(24.01- 24.31)
What are they truly searching for? What information would satisfy that search best? Search answer their main question best. This AI will then produce the best answer to satisfy the Query Intent in a 35 to 45 words short or summary answer.
I'm being clear, concise, and precise. I've got the exact paradigm that we want to see here and we want to use here. “This AI will determine the best answer or explanation to a Query Intent or search attempt.”
(24.35- 24.50)
“This AI will first determine the most likely intent of that searcher based on the query. What are they truly searching for? What information would satisfy that search? Answer their main question best. This AI will then produce the best answer to satisfy the Query Intent in 35-45 words.”
So we have the Query Ballet for adults. And you would insert your query here. We have the Query Intent that it will produce and then the short summary answer in 35 to 45 words.
And so now I'm going to give it a line so that it knows it's its term. Time to go, and I will hit enter.
(25.00- 25.20)
So it came up with this. “Ballet is a great form of exercise and artistic expression for adults. Many dance studios offer greater classes for adults, allowing them to learn and improve their skills in a welcoming environment. Additionally, ballet can improve posture, flexibility, and coordination.
Now, ballet for adults is probably a bad example because it's not the kind of query that has a Featured Snippet. But let's just continue with the example. It came up with a pretty good answer. They want to find classes on ballet for adults. So they said ballet is a great form of exercise and artistic expression for adults.
(25.21- 25:54)
Many dance studios offer beginner classes for adults. That's what they want. They want to know about, should I be using ballet for adults. Is it good? They want to know about classes.
They’ve nailed the Query Intent and wrapped it into the featured Snippet although it didn't print out what it thought the Query Intent was.
Greg: Would you ever put the Query Intent based on the keywords by BERT query Intent Finder? Would you ever insert that right there to help it?
Josh: You definitely could. Let me show you how that would work.
(26.01- 26.30)
We knew that the query intent was, “Find a ballet class for adults.” Let’s see how it changes when we add in the query intent that KeywordSpy would have produced for you.
It says, “Looking for ballet classes for adults? Check out local dance studios or community centers offering beginner to advanced levels. Many also offer virtual options for convenience. Start with a trial class to see if it's a good fit for you.” Do you see how it became much more sales-y?
Greg: The first sentence with the question is not appropriate for a Featured Snippet.
(26.37- 27.01)
Josh: Yeah, I agree. As I said, bally for adults is a terrible example to do this with because it is lower in the sales funnel. It's not a mid-tier sales funnel.
So here's a sales funnel. Ballet for adults is down here in the sales area.
We're up here in the no-do area, where we're looking for ballet for adult classes. We're getting down to the sales funnel. We're not asking, should I use ballet for adults? That would be a much better search intent.
(27.02- 27.36)
I just want to keep going with this. Every time you'll do it, they'll provide a different answer. See, the first answer was, you can go back on version history.
“Ballet is a great form of exercise and artistic expression for adults. Many dance studios offer beginner classes for adults” so they've nailed the Query intent. But that's not quite what I wanted it to do.
We wanted to know that it's coming up with the right query intent. So let's give this as an example and say, “Find a ballet for adults class”. We’ll use the short summary as an example.
(27.40- 28.00)
Here's another pro tip. I will tell it if this is an example. I will give it the query. I'll give it a query intent and a short summary in 35 to 45 words.
If you're going to give it examples, but also because we've got a lot of token space and chat, GPT is way smarter than previous versions of GPT and other transformers. If you take what it produces that is good, it has a better chance of replicating these token parameters.
(28.09- 28.40)
Does that make sense, Greg? If I get it to produce something that looks good, if I use this as an example, I'm like doubling up on the statistical chance that it will produce what I want instead of giving additional words and phrases to confuse it.
And it doesn't know how to replicate the statistical pattern I'm trying to get it to replicate. Does that make sense?
(28.45- 29.13)
Let's give it a different query. Let's give it self aware AI and see if it comes up with the Query Intent and a short summary in 35 words or less.
Notice how I reinforced 35 to 45 words here and here in two spaces. If you reinforce it, it has a much better chance of doing it. So it did a pretty good job.
It says, “Query Intent information about Software Artificial Intelligence. That's right.
If someone searches this, that's exactly what they're looking for. They're looking for information about what software AI is. So it got it right.
(29.20- 30.03)
That's definitely a really good answer there. You might be wondering, “Why did you try to get it to do two things?
Why did I try to get it to generate the Query Intent and then generate the paragraph? That's like dual purpose prompting and I'm so glad you asked.
The reason is that when it generates this information, it’s going to be a touchstone to keep the answer on the purpose. That’s why I got it to print it out.
(30.04- 30.30)
It's not enough for it to have it in its mind. It tokenizes the entire thing and then has an attention mechanism that focuses on the attention of the context, which you can set or which it automatically sets when you're using this profile. You could set it manually when you’re using the API or the playground.
I know you could do it manually in the API, as Greg knows because we do a lot of API calls to ChatGPT in KeywordSpy.
(30.40- 30.59)
This way I'm making sure it knows what the query intent is before it goes, and then statistically produces more tokens that will be influenced by these words. If you want to think of it like a person, it’s like you’ve reminded them that this is what they should be talking about.
That means subconsciously they'll talk about it more. Or if you want to talk about this thing like an AI, you produce these tokens. So these tokens here, which is the query intent.
I've reminded it that query intent is important for writing this paragraph. I've reminded it that this is the particular query intent that it needs to write about in this particular circumstance. I've also reminded here, it needs to do this in 35 to 45 words.
(31.02- 32.05)
Just like hypnosis or sales, you're repeating yourself. You're reinforcing the information you want it to learn and you want it to be able to produce and spit out. This is the next level of prompt engineering.
This is really how skilled engineers would do prompt engineering. So let's look at the answer it came. So let's remember what the Google answer is here.
The Google answer is “This will be when machines are not only aware of emotions and mental states of others but also their own.” They're aware of their mental qualia, they're aware of the semantic information that goes on that can represent itself.
(32.10- 32.32)
Is that more or less what Chat GPT said? And did it say it in 40 words? Let's find out.
It said, “Selfware AI refers to machines that have a sense of consciousness and are able to perceive their own existence. Yeah, there you go. Then it says, while current AI technology is not yet self-aware, research in the field of artificial consciousness is ongoing and may lead to future breakthroughs.
Okay, so this is fascinating. So, Greg, do you think this is a good featured Snippet? Does this speak to the Query intent? Does this answer what the person was looking for?
(32.35- 33.01)
Greg: Yeah, I think it does pretty good. I don't know if it would replace on the SERPs, but it has been a great contender. The other thing that's really good that we know that you need to do is it also has the EMQ in it.
Josh: The EMQ being in this paragraph is very important to get the featured Snippet, or it needs to be in the H1 just above the featured Snippet. It even put the EMQ in it. I wouldn't put the EMQ in the H1 and in the paragraph. I’d put it one or the other.
(33.02- 34.14)
Let’s just do what GPT stands for. Now we can compare apples to apples with Google. They’re absolutely right.
“Query intent: Learn the meaning of the acronym GPT,” It says, “GPT stands for” and it put the EMQ in it.
GPT stands for Generative pre-trained transformer. It is a type of language model devolved by OpenAI, capable of generating human-like text. GPT models are widely used in natural languages processing applications such as chat, bots, language translation and text summarization, and even more than that.” I think that’s 100%.
(34.15- 34.52)
I think it's even better than the featured snippet on Google. “GPT stands for generative Pretrained Transformer, which is a type of large language model neural network that can perform various natural language processing tasks such as answering questions, summarizing text, and even generating lines of code.
This is really good, too.
Greg: It leaves out open AI.
Josh: ChatGPt gives more information and is a better-featured snippet. This is another thing that we happen to know from testing at Underground SEO University. For a better-featured snippet, you need to have the background information in it.
(35.01- 35.43)
So do you see how I killed all the birds? With proper prompt engineering and coming at it from the right perspective, I was able to get the best possible answer without even telling it to put the EMQ or give us background information. Although, to ChatGPT’s credit, wasn't that one of the things they said you should have the background information?
Greg: Yes. It did.
Josh: Other than this, the wisdom of crowds wasn’t that wrog. Even using metaprompting can greatly help you, especially when you’re not sure what it is you’re doing or how to do it. If I use GPT4, the quality would just be better.
(35.40- 36.45)
Let’s try the other one here. Do you want to try something controversial? Should I use Viagra? How about that? Should I use Viagra? What's it going to say about that? “Query Intent: They're seeking advice on whether or not to use Viagra for a medical condition.”
And it says, “you should consult with a doctor or medical professional before considering the use of Viagra. It is important to discuss potential side effects, interactions with the medications, and any underlying medical conditions that may affect its use.” What do you think about that one?
Greg: Yeah, I don't think it's going to give medical advice. It’s trained not to give medical advice.
(36.46- 37.25)
You're totally right. It's trained not to give medical advice where Google needs to give you medical advice because people have been demanding that forever. Look at the featured snippet that comes up for Google..
It says, “No, you should not take Viagra if you don't have ED. Viagra is a prescription medication that can have serious side effects and it's approved only to treat ED, but and not to be used recreationally. You should only take Viagra if it's been prescribed to you by a doctor who knows your medical history.”
(37.30- 38.10)
Greg: Google can reference Medical news today. It can do its background check, reference, rank, whatever we want to call it.
Josh Yes. That's why it's allowed to give that here and give a little bit more advice. I think this is a very worthy featured snippet on ChatGPT. It says that you should consult a doctor before using the drug. Whereas you notice how biased Google seems here. It gives a stronger answer here than in ChatGPT.
(38.12- 39.04)
That’s why I chose this example because I knew it would be controversial. ChatGPT is supposed to be supposed to be politically correct. It comes down and gives you a balanced answer as an open AI assistant.
Whereas Google can pick up on how much people online talk about taking it without any prescription and they know how dangerous this is. Google’s transformer is going to be more biased and give a stronger answer not to do it. So I thought that was a really interesting example of differences between the two.
(39.10- 39.40)
This is really important for understanding entity vector. This is really important for understanding the entity object model that Google uses to rank all these things. Entities are extracted from NLP algorithms.
It's really important to understand how the vector of these entities, the meaning of these entities, moves in a certain direction. This is really important for understanding the consensus of experts or the reference rank that Greg mentioned. And this is the consensus of the experts.
(39.45- 40.16)
That would be another thing that we would work into KeywordSpy, our Featured Snippet producer, is that we should also think about the consensus of experts in this model, because Google obviously thinks about the consensus of experts.
In this case we would trust GPT because the entity vector and consensus of experts is going to be important for certain queries. We’re building the featured snippet generator in KeywordSpy and we’re going to have one click generation of content.
(40.17- 40.46)
Then we’re going to build a featured snippet detector that detects whether it should a featured snippet. If it should, it’s going to generate a featured snippet and it’s going to do a lot along the lines of this AI here, except we’re also going to build in the consensus of experts.
If the experts emphasize being negative, then the AI should go ahead and be negative too. If the experts are all positive on a thing, go ahead and be positive. If the experts are neutral on a thing, then be neutral. That will change the tone of the Featured Snippet that it writes.
(41.00- 41:22)
I think that's how we're going to get a one-to-one correlation between what Google thinks and what we can get Chat GPT to generate here. What are your thoughts on that, Greg? Guys, do you guys have any questions about this?
Greg: I was just double checking the Consensus App. It's a little strange in terms of what it's saying. It's actually giving recommendations on how to take Viagra. So it comes out complete opposite of Google. That's interesting.
Josh: Oh, yeah, interesting. I'll type the exact same into Consensus App.
(41.23- 42.00)
Consensus App is another one of my favorites. Consensus, ChatGPT and Midjourney are my favorite AIs right now, other than KeywordSpy. KeywordSpy is my absolute favorite because I helped make it and I use it daily in SEO for saving hours of my time.
All Consensus App does is that it checks the query you do in research papers, peer reviewed, actual science, the best of science, the most peer reviewed, and the most cited research papers.
(42.10- 42.31)
It has answered so many questions for me, like, should I take omega three S? Are they actually useful? Should I take creatine? Is it actually useful? Are pollinators actually dying? If you have any question that you want to know the absolute truth to without any bias, because science weeds out bias by peer review.
These are peer reviewed, the most peer reviewed, the most cited journal articles.This is the most trustworthy science here. I love using it. We use it for our SEO purposes to make sure we know what the consensus of experts are, because these are the real experts here.
(42.32- 43.17)
That’s why Greg did this. If I do, should I use Viagra? If I click synthesize button to give me a featured Snippet summary, we'll see what it says of whether or not I should use Viagra.
Greg: For me, it just says Viagra can be safe, effective, and well tolerated treatment.
Josh: It says, “ These studies suggest that Viagra may be helpful for treating erectile dysfunction, premature ejaculation, and improving quality of life while also promoting social change are being used in HIV AIDS prevention programs.”
Greg: More positive light of Viagra.
(43.20- 44.09)
Josh: This one is highly cited. It says, “These women use Viagra to discuss pleasure and danger in their lives.” Another study discusses the recommended single dose range, 25 to 100 milligrams. The use of the drug is safe provided the drug is taken under proper conditions.
Consensus is just showing you what the papers are saying and they do not bear responsibility for the papers. Whereas Google is fixated on a prescription for Viagra. Our featured snippet came out to be a bit like this.
(44.15- 45.10)
When it comes down to the featured snippet, I would side with Google. If Consensus agrees with Google, I would agree with Google.
But Google is the winner here because Google is what we want to rank in, not Consensus. However, Consensus is good for actual truth. Google is good for their version of truth, which many people have critiqued Google of.
I did, I had a debate on my blog on themoralconcept net. I had a debate with one of their senior engineers over how they determine what truth is.
Many years ago when they started doing this, I debated their senior engineer, Paul Har, over Twitter. At least he was willing to hear me out, as I said here. So if you want to read that, go here.
(45.12- 46.17)
What I would do is I would make sure I'm going along more with what Google says.
What I might change in this prompt here is “If the consensus of the experts on this topic is negative, this AI will mimic the tone and sentiment of the consensus of the experts on this topic.”
For example, if the consensus the experts on this topic is negative, the 35 to 45 word answer might be a bit more cautionary. If the consensus of the experts on this topic is a bit more positive, then the 35 to 45 word answer might skew a bit more positive. Let's do the query intent expert sentiment is positive.
(46.18-46.50)
Now let's just make sure that's the case so we're not confusing it. I'm just going to save this and submit, but we're not going to look at it. I'm going to search for it and say, “Please tell me the consensus sentiment on the notion of adults trying ballet for exercise from the experts in this field.”
Yes. So it is positive. So that's a good example for me to give.
(47.01- 47.18)
So now it changed it. The query intent to inquire about whether to use Viagra for personal situation the expert sentiment is cautious medical advice. Short summary is the decision to use Viagra should be made after consulting with a healthcare provider.
Now it's mimicking what Google says, right? “The decision to use Viagra should be made after consulting with a healthcare provider. They can assess your medical history and determine if Viagra is safe and appropriate for you. It's important to understand potential side effects and risk associated with this medication before use.”
There you go. Should I use Viagra?
(47.20- 47.55)
This is now mimicking what Google said. This is a much better prompt for how to get the featured snippet.
What do you think, Greg?
Greg: Love it. I can't wait to test it out.
Josh: Let's give it just another example. Let's go back here. Query should I use payday loans? That's another contentious one, isn't it? Because all the payday loan sites will be like you should use payday loans.
(48.01- 48.29)
But what do the experts say? “It says it's cautionary and it says payday loans often come with high interest rates and fees, leading many to fall into a cycle of debt. It's generally recommended to explore other options such as creating a budget or seeking financial assistance before considering a payday loan.”
So when you're doing SEO for payday loans, you'd think you'd never put that in there, and I would disagree. I would say Google is looking for HEAT. They're looking for helpfulness experience, expertise, authority, and trust that is based off the consensus of experts. The experts are cautionary on this.
(48.30- 48.58)
You need to be an expert and you need to be cautionary on it. People who need payday loans, this would not dissuade them from trying to get a free quote. I don't think this would hurt your CRO in any way.
In fact, saying something like this would only help your credibility and help your trust, in my opinion. Our testing in Underground SEO University proves this is the case.
You want to be with the experts in terms of their sentiment. This is going to help you write your article too. Quite frankly, it's going to help the human make sure they're doing the right SEO here.
(49.01- 49.15)
Let's do an easy one just to make sure it gets it right. “Should I get a puppy?” On
ChatGPT, it says, “Getting a puppy is a big responsibility that requires a significant amount of time, effort and money. Before making a decision, consider your lifestyle, living space, and financial resources.”
“Ensure that you have enough time to probably train, exercise, and socialize the puppy.”
(49.16- 50.01)
Let's see if that's what it says on Google. “Should I get a puppy?” Doesn't give you a featured snippet. It's like this is too contentious. I'm not going to give you a featured snippet for this one.
“ Puppies are mischievous, loyal, and full of life. Avoid backyard breeders. Never buy from pet stores.”
They're pretty cautionary here too. “Puppies tend to be a lot more work than adult dogs.”
“You have to make some time for your new pup. Be patient and search for a dog that has the right energy level and temperature. You have to be coming home.”
All the experts here are cautious. So whether or not they will show a featured snippet, maybe no one has the right feature snippet.
(50.10- 50.38)
Maybe people don't have this on their page. The EMQ here would be in the H2.
“ Should I get a puppy? Getting a puppy is a big responsibility that requires any amount of time, effort, and money.” So I love this. I think this is great in ChatGPT.
A lot of this is already built into Keyword Spy, and this is going to be built into the one click article generation that's coming very soon to Keyword Spy.
We've been talking for already an hour on how to do prompt engineering, how to be a prompt whisperer, how to give it the right context, how to say less is more, how to give it everything you want to give it. This will give you a feature snippet that will also mimic what Google is saying.
(50.38- 51.12)
So if you guys have any final questions Afief says one, “How can a new website with a very fresh domain and no backlinks be optimized to crawl quickly and appear on the first page of Google without having to wait for months? That's a great question, Afief. The short answer is join Underground SEO University and I would teach you how to do that.
The longer answer is complex.There's a lot of different interworking parts. I'll tell you right now, there's no such thing as the sandbox. That is totally false.
(51.13- 51.30)
Is there a successful case study that answers this question? Yes. In Underground SEO university, we have tons of case studies that show that we can get page one, if not number one rankings in weeks, a matter of weeks. The only reason why it takes so long is because it takes time to accrue those signals.
You got to build up those signals. George says, “A mid journey tutorial would be great.” Sure, I'd love to do a mid journey tutorial so we can take a look at that.
(51.34- 52.07)
David says, “With one click article generation, can your initial prompt include some background information, like company name? Yeah, so that's a great question.
When I say one clicked prompt generation, I'm talking about the method that we're going to be doing here in Keyword Spy that we're building. When it comes to ChatGPT, we’re running six different AIs communicating built of multiple prompts.
It's not just one click in Chat GPT. Sorry if that was confusing.
(52.08- 52.50)
When I say one click article generation, I am only talking about KeywordSpy and we are the only tool out there that does it to this level. Just to be clear, in Chat GPT, you're going to be running at least four, five, six prompts working in conjunction to do article generation to the level that you're not going to get penalized or demoted, because you're lacking on all the important metrics.
You're lacking the Rank Brain keywords, the BERT keywords, the neuro magic keywords, the helpful keywords, the product keywords.
(52.51- 53.02)
You're lacking on depth of those, you're lacking on spread of those, you're lacking on the consensus of experts expertise, you're lacking on the featured snippet. You're not answering the query intent. There are so many boxes you need to tick for on page SEO.
It is very hard to do. Greg can attest to how hard it is to do because we have a very complex AI that is a little bit self aware, looking at each other and analyzing the content that's made and rechecking it and checking it again to generate all this analysis in multiple stages. So it's actually quite hard to do.
(53.03- 53.45)
That just includes the front facing content, which I probably missed one or two aspects of it. It doesn’t include the images, the Dom vector, which is a whole another mini self aware AI that needs to be built, that is checking and producing. That's another four or five prompts working there. At least the one click article generation is not in Chat GPT.
And that's not something you should do. As I said at the beginning of this video, and you see how complex I had to make it in my prompt that I'll share in the description to create the featured Snippet. You see how complex I needed to make it.
(53.50- 54.23)
I needed to make a multi prompt inside of a prompt and I need to make it generate certain things and then use that generation to produce the right kind of writing. That's not even what I'm talking about. I'm talking about four or five AIS that use four or five prompts, that are multi prompts built in to generate this as complex in the way that we're fully doing it in the end.
So it is another thing entirely to build all this. It's another level of prompt engineering. It's the next level of where AI is definitely going and other tools are going to go there.
You can't just do one click generation just in Chat GPT. I just want to make sure that that made sense.
(54.25- 54.45)
To answer your question, can you use background information like company name. The answer is yes. I would definitely mention your company name if your entity is strong. There's different kinds of entities.
The SEO industry thinks the only entity is a person, place or thing like brand name. But entities are a concept. Entities are every single word. So there's the entities that experts say, and there's the entities that experts are. You need to mention both, and you need to mention the right ones and not the wrong ones. So again, email me if you want to join Underground Is University at joshbachynski@gmail.com
If you want to try KeywordSpy for free for two weeks, you can do that. Go to trykeywordspy.com to try KeywordSpy for free. It automates and it does a lot of this work for you.
(55.02- 55.58)
So for those people who like education and you want to learn all this information and spend hours learning it so you know it to the high degree, email me to join Undergroundness University. However, if you want to go fast and you just want to get your work done, as Greg often reminds me, going fast is important. Then you want to try KeywordSpy for free for two weeks.
And we also have that still that offer the grandfathered price. We still have a few spots left, 30 or 40 spots left. Try Keywordspy.com. Sign on for the $144 a month and get your grandfathered price now. Because as soon as we put that One Click article generation in the system, everyone is going to see the light and those 30 or 40 spots are going to be gone forever.
And when those 30 or 40 spots are gone forever, the price is skyrocketing. It is going way up. So I'm just telling you now to be nice.
(56.10- 56.50)
I highly recommend that if you think this is a tool, even vaguely, you might want to use try it for free for now, for two weeks and get your grandfather price at $144 US. This AI tool is unlike any other AI tool out there. It does the work of at least three AI tools, if not more.
All the AI tools that just do one thing, not all three things, they all cost like $500 a month. So I guarantee to you our price is going to go up after that because conversions on our side are going to skyrocket once people see how easy it is to use with the One Click, the One Click article generation that will do all of this for you plus the featured snippets.
(57.10- 57.15)
I'd like to bid you guys have a great day.
Thanks, Greg, for being my co host today. Make sure to like and subscribe watch for our future courses, which will give you more free SEO tips and the best ways of using SEO and AI.
All right, well, thanks a lot, folks. This has been the SEO AI show. So see you guys next time!
Share this post