ChatGPT + GPT-3.5: Google, SEO, Content. Das war’s!
You'll learn if ChatGPT is a threat
to Google and SEO or if it's just hype.
We look at what ChatGPT can, what it can't and how it can enrich your life. Have fun! My name is Alexander Rus
and SEO is my bread and butter. On this channel we chat
about SEO and content marketing. If you want to learn how you can sustainably acquire
customers through your website, then subscribe to this channel right now. ChatGPT is absolutely insane and
simply gets things moving again. Nevertheless: In this episode we will
separate tough facts from hype. I know you should never
say that at the beginning of an episode, but this isn't a thin,
easily digestible episode.
We won't discuss here 5 magic tips to use ChatGPT to automate
your life. We take a critical look at what ChatGPT can do? Very much. And what not. Towards the end I will show you an example of whether AI content generated by
GPT can rank on Google or not. But let's get started. We at Evergreen Media have been experimenting with GPT-3, i.e. what
ChatGPT is based on, since the end of 2020.
That's why it wasn't such an extreme mindblow for us now as to where the direction is going,
but so that we have a good foundation: What is ChatGPT? What is GPT-3
and what is the difference? Both are once great language models of OpenAI, but they have extremely important differences,
which we'll now take a quick look at in this source. The base layer is this language model GPT-3, and ChatGPT is also based on it with additional
layers that have different purposes.
One, of course, that it's more
about the conversation and one about the security of the information that
comes out. So. To understand how it all
works and what it can do: What is GPT-3 anyway?
What does this abbreviation even mean? The first word is "Generative",
which means the goal of the model is to generate text.
In a way, it always wants to predict what the next word after a word is –
put simply – and that shows you what will eventually be possible
and what will never be possible as a result.
Then:
"Pre-Trained" This simply means that the algorithm has already been trained on a huge data set
and simply has already been configured and
that it no longer needs to be trained. And "Transformer" is simply the
algorithm that the model uses. He just specializes
, I have to read that now, processing long sections of text, paying attention to the
important information they contain and understanding how words are used
in natural language. The first release of this third
version of GPT was in May 2020. It is one of the most powerful
language models on the market.
Of course there is a lot that is
not on the market where you don't know how good it actually is. And that can be used for
just about anything, like language translations, use with
chatbots, or text summaries. And GPT-3.5 or ChatGPT are based on GPT-3
, but work within certain guidelines. What exactly is ChatGPT now? It was specially developed for
chatbot applications and was accordingly trained on suitable data – i.e.
Conversational texts.
And the most important question should be: Is ChatGPT more powerful than GPT-3? No, or not really.
It's up to how you want to use it. After all, the hype came from the cool, free demo, the nice
user interface and the fact that it is "more secure", but basically this raw GPT-3
from back then is still more powerful. So. But still it has to be said that ChatGPT gained
a million users within 5 days in December.
That has never happened before and
that's a sign. But just for understanding:
Both are designed for different purposes and have different strengths and
weaknesses and different areas of application. What is perhaps also generally exciting: why is
artificial intelligence making so much progress right now? Now independent of ChatGPT and GPT-3. This is explained insanely well in this The
Economist article on Foundation Models. So I don't want to explain what
foundation models are, but there's a reason why so much is happening right now
and all the tools are coming out and now the game changers are really coming.
And it will
simply continue like this for the next few years, because a
breakthrough has simply been made. So. Now the big question:
What can GPT-3 or ChatGPT do? First the short answer and then
a super detailed answer. Basically, very, very
much and at the same time much, much less than the hype is currently promising. But I encourage everyone to
use it and make up your own mind. And don't just play around with the
demo for a moment and then be ecstatic that the answer makes sense,
but actually try to work with it, because that's
a huge difference. Now how does GPT work again in
simple terms so you understand what you can and can't use it for
and where the journey is going. So for GPT, it's all about finding the next word and that word
matching everything that came before it. This generates probabilities for different words with a large number of
parameters and then
selects the most probable word.
That is, to use ChatGPT as an example: You type in some
prompt and then this model tries to predict what the answer is based on what you read and
the rules you learned . And that's
very different from how people write and think. And thinking is a keyword, very important: GPT can't think and does
n't understand what it's writing or doing. It's not conscious. It just says based on
a huge amount of data and what it's learned
and within certain rules: Hey, this word should come after this
word if these words were there before. So how does a human work now in comparison? Very, very much more complex and different. We build some
abstract mental models. Link different
topics and experiences. Now, when we write a text, we have a
construct or concept of the entire article in mind, and when we start writing,
we manifest this construct word for word.
Which is just the big difference:
GPT or ChatGPT is mainly a memory, and then it's about probabilities, and
a human is so much more than a memory. We can think, plan, be cognitively
sensitive, problem solve, and so on. And then I haven't even
talked about emotions, which allow us to understand others, have opinions and care more about an idea
than others, and so on. ChatGPT, GPT and all that stuff
is always based on existing data, so you can't invent anything new. And with that comes the following: You can't really have
unique perspectives on a topic. Now that we understand how a human works and how ChatGPT
works, what can ChatGPT do very well? For one
thing, it's incredibly easy to use if you're using ChatGPT now.
But also if you are using GPT-3
because the API is mostly used via some language generation tool and is also
extremely easy to use and extremely good at getting something down on paper quickly.
That means generating text. But it's very important: The difference between "it's something
good, meaningful" and "just a text". Most people are just so intrigued when they get
a reasonably good answer to something, but that's not game changer content. Now, if we really look at content. Even today, most people
confuse word count with a quality metric.
For example, Google itself says that content
with too many words is also thin content. Or simply, if the content of the page has no informational content, then it is thin
content, no matter how many words there are. So. You can have a build-up conversation with
ChatGPT because ChatGPT is always looking at what came before, meaning you
can talk in context too. We will then discuss what kind of errors occur
in the next point. It's an incredibly good
content sparring partner.
You can easily generate ideas, you can
generate outlines, you can generate formulations or ideas for what is
semantically relevant in this context. That means you can do a lot and that
's a cool help. For example, if you generate many short paragraphs and then,
let's say that everything has a common thread, then sometimes
something really useful can come out of it. You can quickly generate a lot of ideas,
some of which are really good. You can ask simple, non-layered
questions and get easy answers. And what I haven't even mentioned yet,
because we've been talking more in the context of content: ChatGPT can program,
because ultimately it's also a language. That means you can use it
to generate any codes, for example. So code is exactly like language, where things repeat themselves and so on and
where you can generate something in principle. That doesn't mean it can
program perfectly, because there are already a lot of counterexamples, but
it can help you with programming.
What Can't ChatGPT Do? I'll just use the
ChatGPT example now because that's what people are familiar with. In principle, GPT-3 is the interesting thing,
but we've already discussed that , you've known that for a long time. So
what can't it do? If you now generate a text
with ChatGPT, and we'll look at an
example: It doesn't get to the point. That means you ask him a question
and it doesn't get to the point.
You're like,
oh my god, what a load of nonsense. The red thread is simply missing when writing. It can't write a coherent contribution to something
, but if you – let's say you have a longer text, then there will
be repetitions or things that don't fit together and so on,
because that doesn't work like a brain, but word for Word by word by word
will calculate probabilities. So it can't make any connections. Also, and I think that's what
most don't understand and why ChatGPT isn't revolutionizing everything now:
It can't tell the difference between facts and [ __ ]. That is, it will even invent quotes,
it will invent numbers, and so on. It's like "oh my god, if
ChatGPT says that then it must be true" but no.
That means the application in the real
world is still very limited. Of course, it has an incredible number of sources,
but what it cannot do, for example: It has no capacity
to sort these sources by authority and to provide any information from the most reliable source
. Then, in principle,
if you see this as a chatbot, it gives you an answer to your question.
Not a choice. That's more in context now because some
people are like, oh my god, ChatGPT is like a search engine.” We'll talk about that – no, just no. It cannot provide up-to-date information,
it was trained with 753 GB of data from books, web texts, Wikipedia articles and
other texts on the Internet.
So this is from this source. Let's look at that very briefly. It explains where
the information comes from anyway. And that went sometime until 2021,
which means it doesn't have any more recent information. If you write with it now,
you'll find it's extremely good at making claims, but
extremely bad at backing them up with data and arguments. And that is the most important
thing for a text from my point of view. It simply cannot bring in new ideas
– keyword information gain. We'll talk about that later. It has no empathy and it's full
of prejudices because of course it's trained based on data and they
're getting better at establishing guardrails, but of course it's fundamentally full of prejudices.
You see that in jokes and stuff like that. This is now a problem
for a company like Google, for example. Google can't
bring out such a chatbot because it would just not only be reputation-damaging, but destructive. That means, very important, what you
should understand at this point: The data basis for this
AI, but also for any other AI, is always human information
or human content.
Otherwise there would never be anything new again. That means people still have to
find information, process it, etc. But why should you
still deal with ChatGPT and GPT-3 and use it? So I think it's super cool. It's just like a tool. An SEO has an SEO tool. Anyone who publishes content should use this
tool because it simply helps you become more productive and also more creative. I know that's a contradiction with: It can't come up with ideas,
but it can say very quickly: You ask:
"Hey, give me ideas for so-and-so" Then it gives you 10 points.
9 out of 10 is total bogus, but then one idea is pretty cool. Just a new approach. It cannot now write text on a new approach
, but it can generate ideas. However, it does not replace a journalist
or a content creator. I'm just going to assume that you
're a good journalist or content creator. If you're still in elementary school
and you're a content creator working at a company, chances are
the AI can produce something more meaningful
than you probably can.
But now when we talk about adults who
have an education in this field , we are light years – and really
please use it, "light years" – away from being competitive with the journalists
or content creators. I know other people see it differently, but
please send me texts where this is the case. I've been using this for a relatively
long time and no, just no. What's cool is that it's incredibly well-read and you can
easily use this information via ChatGPT. It can generate ideas for you very quickly , it can help you with formulations, it
can sometimes generate ideas for paragraphs. When you generate content, it simply saves you an
incredible amount of brain resources that you can better invest – namely
that the content becomes even more legendary, that you stand out even more, that your
content becomes even more outstanding. And now, for example, if you're an SEO,
then it's handy because it can help you with programming stuff for Excel or
Google Sheets or whatever you use.
So,
now of course we come to the topic: What does ChatGPT mean for Google? And I find it really funny
how others see it and also write "Oh my God, Google
is dead!" and blah blah blah in the comments. I will now put forward my arguments, which I think are relatively difficult to
refute, but we shall see. For one thing, Google Search
is so much more than just a chatbot. Google is a multi-layered construct
that combines text, video, businesses, maps, images, news, hotels, jobs, advertising, flights, etc. and effectively delivers based on very
complex algorithms. It gives good answers. This source explains in extreme detail where is ChatGPT cool
now and what's so great about it, but why it's just wrong
to see ChatGPT as competition for Google on the one hand : One is a chatbot.
Google even has its own chatbot, which means these would now be products that
could theoretically compete with each other, but seeing ChatGPT as an alternative to Google
Search is completely preposterous at this point.
If we think ahead now and let's say
that this is evolving, then of course you have to look at it, but compared to OpenAI, Google now has
an infinite amount of data and various platforms to collect data that others never or
only in the very distant future will have. That means, imagine, Google already has
a product, Google is releasing a chatbot that merges with Google Search. That is, for simple questions—well, I ask: how many calories are in a pumpkin? A chatbot is much better there than
what Google has there, because after all Google now has featured snippets
– a chatbot would be much better here, and so on.
I believe you have to see a chatbot as a personal assistant
and not a search engine. Chatbots will
complement or enhance search engines, but will not replace them.
So, now that you understand how these language models work,
you should understand that the answers can never become unimaginably reliable
and a search engine is just more than that. There. Now some will say: Yes, that's quite nice what you say .
Why is a chatbot not a search engine? A chatbot now provides an answer to
a question and that definitely makes sense when I ask "Why is the sky blue?" "Will it rain in Innsbruck tomorrow?" "How many calories does an apple have?" and so on.
And Google is trying to do that in
Featured Snippets, ultimately with a language model. That means the difference is that
ChatGPT generates the text and Google's language model tells you:
On a website, what is the text that is relevant here and delivers you
featured snippets here – also for understanding. But in many cases, and especially in interesting
ones, we humans want to research information on a wide range of topics
or simply research, read and understand in depth . And we read different
sources and then form an opinion. Then, we are unimaginably
sensitive to misinformation. That means, that's why Google is
so important with the authority, so that information from this source … So I want the medical
information I'm researching to be not gibberish from ChatGPT and
driving me to my death, but me wants this to be written by a doctor who
is an expert in the field.
Can ChatGPT do this? no So what is ChatGPT? As already said, an excellent
addition to a search engine
to be able to answer simple questions faster. Because I absolutely see the problem that simple questions are poorly answered in Google search
. I would never doubt that, that's just the way it is. But we have now with… ChatGPT is
ultimately just a demo, but now there will finally be functioning digital assistance
for everyday use through something like ChatGPT. Now it's still important, many don't know either, but that's just unimaginably crucial,
I think, to form an opinion here: What about Google's chatbot? So Google has its own product called
LaMDA which according to the people who have seen this is very good and far better than GPT
because it is also trained on a lot more data. I think the big question for Google is:
How do I use a chatbot effectively? How do I establish this in search without having my
existing business model destroyed. There's a cool
article about it at Seeking Alpha anyway. Oh wait, that was the wrong guy.
It explains why Google
hasn't released LaMDA yet. The bottom line is this: Chatbots
are just insanely vulnerable to misinformation and that's just a
disaster for an established company, and especially a company
like Google that's under so much pressure for misinformation,
and the tech giants in general. That's why OpenAI is… A demo is cool, but let's see how it
will actually look in practice. What is exciting to say about this: Hey, but Microsoft is planning
to integrate ChatGPT into Bing. Microsoft was originally an investor and is now investing
even more and wants to incorporate that more. We will then see
what false information is like, because Microsoft will also come under extreme
pressure. And as we've learned now, chatbots are good for the easy questions,
but the easy questions aren't… So, "how many calories are in an apple?" or
those kinds of questions that are so easy to answer…
So let's say times, all the keywords
where featured snippets show up are often not the keywords that are really
interesting where the money is being made. So, now I've lost the thread for a
moment, but I'm back now. The thing why
everyone is hyping this right now and saying, "Google is dead now" and blah blah, is because there's
just bad anti-Google sentiment. But it's always really important to me, I also think
that Google doesn't do a lot of things right, but the dangerous thing is what's
perhaps the disadvantage for us humans: we let our emotions
influence the facts about a thing.
And this article describes it perfectly:
"The danger of disruption is real." No it is not. "Google and YouTube aren't innovating?" If that's how you feel, then… In the top 3 largest
websites in the world, that's two. If you don't find that innovative,
I'm surprised you're not a billionaire. And one must
not forget the company sizes. I'll summarize that again at the end
, but it's just funny to see ChatGPT as a competitor for Google
Search, but also for Bing Search. So I think we have
a better picture of it now. What does ChatGPT mean for SEO now? So, for one thing,
we as SEOs have a policy of "AI content is against the Google Guidelines". John Müller
recently said something up-to-date about the fact that this is actually against the Google
Guidelines, but that's really important to know.
The detection of AI content created by
language models like GPT-3 is not (far from) trivial. This article explains very well
what the difficulties are and that you can't just do it on a large
scale now. In any case, it must be said that there is a Red Queen effect in everything – I hope
that most of you are familiar with the Red Queen effect. Ultimately, this means that it is
always like this in evolution: Whenever a species makes a big leap, all the others who are
in contact with this species also follow suit. That means, let's say you have a
new virus and it affects a certain species, then they will
follow suit relatively quickly. This means that if a
lot is happening now in generative AI, then a lot will also happen in generative AI recognition. Very important to understand. We can see that in the end, because of course
one language model, if trained on the other language model, can probably recognize the other
language model perfectly well and so on.
And of course GPT and ChatGPT have
an impact on our whole society and that's why watermarking is already in the air
, so that you can see whether anything was generated by AI, and that would then
completely destroy the topic once and for all anyway. I agree.
So. But I formed this opinion through actual experiments.
Then there's a video where I talk more about the
experiments, but I'll show you one. Yes, thin [ __ ] AI content
can definitely rank. Today is Monday, on Friday
before the end of my working day I quickly generated this page here.
Including maintenance, the effort was 10 minutes. The site is via LaMDA,
i.e. Google's ChatBot. I thought it would be pretty funny if
I did an article about it. It's very well written semantically, but it's just gibberish,
you read this and you're like: Ok, I dunno… No informational content, it's
just classic AI gibberish. But, this page, i.e. this
URL, has been completely republished. We are now ranking for these two keywords,
so "google lambda" we are now in 9th place with 1,300 search volumes and for "lamda" itself
with 9,900 search volumes we are now in 14th place.
Time required: 10 minutes. That is, can thin AI content rank? Yes. What I'm still finding out: I'm very curious whether you can really
rank in the top 5 for something halfway competitive, where you really get traffic and then Google
really sees user signals, because I think then the whole thing just kind of drops Stein, but
up to this point it's all about: domain authority, topic relevance of the domain and, let's say, on-page factors, and that
's excellent – and boom, it ranks.
This means that it
is absolutely possible to rank thin AI content, but whether this is sustainable and
continues to rank remains to be seen. But what is important now, because now
everyone thinks "oh my god, AI content can rank!" – Yes, but we don't know yet whether it
can rank at the top in the long term, etc. But this content offers no added value
and no information content and makes the Evergreen Media brand
(in that case) look bad. So someone reading this article thinks:
what kind of idiots are they. They are never experts
, they have no idea.
And that's why it's very important, yes, SEO has the
goal of ranking things, but what's ranked, the content, is supposed to do something.
It should convey expert status, it should convince the user to take an
action and so on. It's not just about visibility for us. Above all, visibility, which we
then cannot monetize. And this is very important to understand: ChatGPT and GPT-3 cannot give you a good article that
convinces anyone of anything. So this article can only convince users
to never visit your website again. But in the next few weeks
we will look at many examples, also mixed solutions, where we worked
with GPT, but also with human editing, and what
came out of it, and so on.
We'll talk about that in detail, but now how do I use ChatGP and GPT-3
in the first place? There will be separate episodes, but
just a few specific tips and thoughts. I would definitely try it
, experiment with it once. So on the one hand with the ChatGPT demo, but also with tools that
access GPT-3, the API, and then just see how I can create AI-supported content with
it and then simply find a process. Then, it's super cool right now to generate ideas
for outlines and topics. It's unbelievably good at content repurposing. That means you have an article
and want to summarize it or you have a transcript and want to turn it into a
blog post – whatever. It's quite good at these things – not without a human,
but it can help you go faster. You're making a guide and you want to
turn it into a transcript for a video, and you want to turn it into a social
media post, and you want to turn it into a synopsis and a newsletter
– GPT can be incredibly helpful with that.
Then, as SEO, use ChatGPT to generate formulas and
scripts in Google Sheets and Excel. And another word of warning!
We're going to talk a lot more about that: If everyone
uses the same language model, everything is on one level. In this case, that means anyone can now
generate endless [ __ ] articles. Because for me, let's say I
would classify all of this as weak content, not even mediocre
and certainly not high quality. That means anyone can now
generate infinite amounts of content for any keyword, and of course that means
search engines will respond. Google, Bing and whatever. Authority
signals become more important, E-E-A-T signals become more important, user experience becomes more important,
information gain becomes more important. What Information Gain is is
wonderfully explained in this animalz blog post. It's simply about you providing really
unique knowledge or covering unique intents
that nobody else covers in their articles. That means you can cover the basics,
but also more. And maybe finally: In
my opinion, the application in the SEO area is very limited.
I know YouTubers like
Mat Diggity have been churning out videos now claiming you can do keyword research with
them, cobble together lyrics, and so on. I looked at all of that. What he calls keyword research,
I call a medical condition. That said, this isn't keyword research,
this is gibberish. And the lyrics that
come out of it are also gibberish. That means it's basically true:
Yes, you can use it for SEO purposes. The question is at what level. If you can't do anything right now
, have no idea, can't afford any tools, then ChatGPT can
put something on paper for you. And the level of putting it on paper is… I'm sure you can use this to convince your 75 year old
boss/CEO that this is going to be a big hit, the
SEO strategy that ChatGPT generated for you. You're less likely to convince me or other SEOs or our clients
because that's just thin [ __ ],
that's exactly the same as saying: I read a backlinko article and now
I can develop SEO and an SEO strategy.
Because at the end of
the day, that's what ChatGPT did. So. In summary. I know, a long video or
episode for those who are listening to the podcast now. The bottom line is, GPT-3 is a fantastic
sparring partner for your content, it generates you ideas, outlines, definitions (that's where
it's incredibly powerful), FAQs (incredibly powerful), it can help you find wording,
it can help you when you have a blockade, that you get ideas on how to
proceed, because it's really good at continuing to write. Generative AI tools will generally make
people very, much more productive over the next few years . But, if you now
really understand what generative means after this episode, then you also know that it's called "artificial
intelligence", but actually the component intelligence or what constitutes intelligence
and how we measure intelligence is missing. OpenAI – awesome company.
As I said, I'm an Elon Musk fan anyway.
That really gets things moving again
, but they're not competition for Google, chatbots are a separate product group that will also
establish themselves in search engines as supplements and are more likely
to be seen as digital assistants . ChatGPT + Bing is
also not a thing from my point of view. I think I've
explained the reasons in detail, especially given that – I think that
's an extremely good final word – that Google has the better search engine,
probably the better language model and chatbot too, and infinitely
more data available . That is, understand such a chatbot or
such a language model, it's all about how much data you have. And who has a
larger data width and data depth than Google? So I look forward to the comments if you can think of any companies. I ca
n't think of any companies at the moment. And now it's your turn:
how do you view CHatGPT? Do you use the language model? And if so, how do you use ChatGPT? I really look forward to comments,
I also like critical comments because I know that this is a very heated
topic and some are of the opinion that there will be no more jobs in the next 5 days
because ChatGPT does everything.
Use ChatGPT and I'll be happy to see
if your opinion has changed. With that, thank you so much for
watching and see you next time. Bye!.
