SEO and AI Generated Content

hello and welcome back to the digital
marketing podcast my name is Kieran
Rogers I'm Louise Crossley and I'm
Daniel rolls and today we are discussing
SEO and AI generated content
[Music]
so this one comes from a really really
important blog that was published by
Google so let's go back
um last year so 2022 Google comes out
and says that you are not allowed AI
generated content on your website there
wasn't really clear definition of that
but obviously since then everyone
started using all these AI tools and
what we're seeing is that everyone's
generating this content but you're not
really supposed to be using it on your
website and everyone's kind of like well
does that mean we're going to get
punished in the search rankings could we
get the indexed what's what's the you
know what's the risk then lots of people
saying well you can't detect it anyway
um there are AI tools being created to
detect AI generated content which all
gets a bit meta but but the reality
being that actually can't even detect
anyway so how would Google go about
doing that so
we're all kind of arming an RSL we
shouldn't really be doing it maybe it's
good for assisting but it shouldn't
write the whole article and then two
three weeks ago now Google come out and
say actually
if you're using AI tools to manipulate
the search engines yes that's a bad
thing okay but good content is good
content and actually if you use AI tools
it's absolutely fine so they've done a
complete turnaround and that it kind of
makes sense because
if the AI tools have been trained to
write really great content and it is
genuinely great content and you're using
it to augment what you're doing then why
shouldn't use it from that point of view
it's almost like they can't stop people
from using it anyway so it's better to
allow it and just police it well yeah
absolutely but also let's bear in mind
they've got their own AI tools that do
writing and they're releasing Bard
really soon as well so that would be a
bit hypocritical so we've got this
really good writing tool but you mustn't
choose it okay so it wouldn't make any
sense at all
um so that's a big shift
which is a big I would say a bit of a
relief for a lot of people
I also think it's going to lead to a lot
of lazy content creation I I put this up
onto LinkedIn and it got loads because
oh great and loads of people tagging
people in because it's obviously been a
conversation and and this guy came out
and said yeah I did I built a new
website did 250 articles in a day and
they've all been indexed it was kind of
super pleased with himself I was like no
no no no this is not the point of this
there used to be tools that would do
this that would literally just fill a
WordPress blog by scraping other
people's website and you you do you are
going to have problems and the reason
being is as the algorithm is more about
more about user signals and you've
spoken about this before here and
actually if you create rubbish content
it's not going to get those user signals
people aren't going to dwell they're not
going to scroll on it they're going to
leave your website they're going to
bounce back to Google so actually
rubbish content is going to still work
the way rubbish content does well in a
way nothing's changed is it like they've
always claimed that what we aim for is
that you know really good high quality
original content yeah should rank well
now whether that's been been you know
written by AI or written by human or
written by human AR combo which in my
experience at the moment is the best way
of of doing this
um because there are still issues with
AI getting it wrong AI hallucinating you
know I've had AI inserting other people
and other companies names into articles
that I've been really like wait what
happened there because we'll talk about
where this data is coming yeah as well I
had just an example of this yeah I had a
brilliant one over there I got it to
write me some code so I said write me
the python code to be able to access the
API via my website I thought it's been
quite clever I'll access the API by
getting it to write the code for me and
it did it and I ran it and it didn't
work and I scanned the code and luckily
because I can code it went through and I
could see what the problem was and I
said no you've done this wrong I'm
really sorry for my mistake here's the
corrected code and it gave me the same
code and I went no you've made the same
mistake again anyway yes no I realize
it's this particular line and it should
be like this yeah here's the code it was
wrong again and I got into this I
probably lost half an hour of my life
yeah but it kept getting Well why'd you
keep getting it wrong and it was it got
more and more apologetic as it went
through and in the end I gave up and I
wrote the code myself because I knew the
solution so
um it's not infallible I think that's
actually that's absolutely right so let
let's let's come back to that that thing
about how this data where this data is
coming from how they're training the AI
in a moment
but there are some tools the um this
this is kind of being integrated into
everything at the moment so talk us
through where you think the good tools
are at the moment and what you're using
so did you know a lot of this is using
older technology older techniques but
incorporating it with you know the the
AI models
um for a long time I've been fascinated
by a thing called
tfidf yes which stands for term
frequency inverse document frequency and
it's uh yeah yeah so Snappy
um it's marketing but it's not a full
like three letter acronyms
too many letters can't get my head
around it
um but what I mean that's that's a very
old technique it was used by a lot of
light like quite old library systems
back in the 80s where you know you look
at how many times a particular phrase is
mentioned within a document and then
that gives it a score and that gives it
a relevancy for certain types of topic
you know fast forward like 40 50 years
is and these things are still being used
now Google have come out and said that
we don't use that
no but they use something similar like
an advanced version of that I'm sure
like because it just makes sense if
you're doing anything that's keyword
kind of driven and what a lot of the AI
um connected SEO tools now do is they
leverage on you know ai's ability to go
out and literally analyze okay give me
the subjects
um so we were writing about SEO okay and
so we want we want to go away go away
and find all the top 100 pages that do
rank on Google and do uh you know tfidf
um analysis of those pages and what you
get is you know and you've seen this in
lots of tools I know sem Rush
um
Surfer SEO I I really love it I really
really do love it if you haven't checked
it out we've definitely mentioned it
before on the podcast but I think it's
incredibly powerful plugs into Jasper it
because it plugs into Jasper but
actually they've just enabled you to
plug it into sem rush so like that
that's amazing I haven't played with
that integration but it'll be doing
similar things to what it's doing with
with Jasper and it enables you to
literally get AIS like look around the
keyword landscape to give you input on
well what should you have and it'll do
things like it'll recommend top sub
headlines that you should have in there
where is it getting that information
from when it's looking at common ones
that are appearing everywhere on the the
options that Rank and actually what it
does is it gives you a lovely simple
score out of 100.
for how hot to trot your content is
um no it's very much aiming at whatever
key phrase you've gone after see you
know you have to do your keyword
research but a lot of the tools are
bringing in keyword research tools so
you instantly know you know what phrases
have got the most traffic around them
and it just helps you get your head
around it and actually in quite a short
space of time optimizing the content so
I'm using the tool to augment my
knowledge of the keyword landscape
optimize stuff the Golden Rule when
you're doing this is not to just shove
random headlines in there sometimes
it'll highlight oh gosh there's a whole
subcategory we've missed out we need to
expand the copy and the AI tools can
help you draft a first draft with that
but again I would always get them
checked by a subject expert to make sure
that it's high yeah you can't be using
these if you don't know the subject let
me give you an example
what we've been doing is going through
and I've started to work through this
process of looking at what's ranking
really well already and say how am I
going to do it better and I will look at
it and I'll run the tour and go I
haven't talked about those three topics
right I could add something and I can
make this a bit more in-depth and
interesting and a bit more well-rounded
but I've been using this for offline
writing stuff so if I've been writing a
report for a couple of business
Publications
it helps me go through when I bullet
point things up the way I write and I've
written a lot in terms of I did that
project last year it was cool to a
million words and it's like the way I
always write things is you bullet point
it and then you sub bullet it and you
sub bullet it and you suddenly you're
just writing 200 words in each of the
subplot and it becomes so much more
achievable because it's looking at blank
page is overwhelming I mean you've
always talked to me about this yeah as
well and actually breaking it down and
what this helps you do is you start
doing it and it gives you all the
bullets and you're oh that would be a
good idea I could put that in there and
then I will rearrange it and then I'll
start to fill it out and then it'll give
me other sub ideas yeah and that works
quite nicely the key to this and Google
have stated this is they're after high
quality original content and that's its
weak point at the moment it's not great
at doing like you can read an article
written by Ai and you're never going to
go oh my goodness that's amazing that's
just blown my mind no it's not because
you've read it everywhere before and
that that Seth go derivative yeah we
should link to Seth godin's recent post
about this because I thought it was
genius and a really great way of
thinking about this I'm going to argue
about this oh yeah okay so what Seth
came out and said um Seth if you're
listening I apologize I'm going to
paraphrase horribly here but like all
credit to you because it really woke
awoke a whole new kind of angular debate
in my mind he said look you you would
never drive a car similarly by looking
through the rear view mirror and that in
a way when you think about you know
these language models and how these
systems have been trained they've been
trained on tons of data that exists
actually in the past yeah right it's not
looking at Future data it can't doesn't
exist yet so it's using that to to model
out all of this this knowledge and all
of this intelligence and and his angle
was very much like you still need your
strategists because they've got a far
greater capability than the AI does to
imagine future scenarios rather than
base everything on what's gone before or
rules and so there's certain areas where
it can Excel but go on Daniel come on
hit me with it right so hit me with it
um our brains are neural networks and
you can have neural network based AIS
okay and a neural network basically says
you have these kind of points of
knowledge yeah and then activation
levels are when two points trigger
together yeah okay
so the point you connect something so
the human brain is quite good at
connecting things that seem very
unconnected and we can make our brain
try and connect things that aren't
connected on purpose we have little kind
of brainstorming methodologies and
things like that for doing it but
actually if our brain works that way and
we can connect oh that thing over there
well that could be I could make a kind
of correlation causation connection
between those two things then an
artificial neural network could do
exactly the same thing and all you've
got to do is is turn up the activation
level sometimes to do that so actually
there is a very clear path to creating
creative AI because you just change
those activation levels between nodes
within the neural network
um so I I think that's really
interesting I think the current way it's
done I mean I I've got it before to do
prompts to try and make it creative
stories yeah and every creative story is
there's a oh that's Lord of the Rings oh
that's Harry Potter all that it's the
point where it had the same movie title
it basically it it wrote the follow-up
to Terminator rise of the Machines and
all this kind of just getting the same
title and things like that so it is very
derivative at the moment generally
speaking but we're in like months three
of this kind of thing in terms of from a
marketing perspective when you're
thinking two years time it'll be so
Advanced yeah like we were talking about
Auto GPT well I'll talk about that but
amazingly that only released on GitHub
on like the 30th of March like it's not
even a month old when we're recording
this yeah I just so let's talk about
that yeah okay and then we'll come back
to a few of these points
um these Auto AI agents uh Auto GTP
basically what it does it's using the
existing kind of Technology but it you
use the API and you do it via code
mostly so just to be clear on this we've
got chat GPT yeah which is a chat
interface chat interface and then we've
got Auto GPT so it sounds similar yeah
so different things so DBT is based on
okay so open AI built GPT 3.5 they've
released four that's the language
natural language processing engine that
chat GPT is based on yeah but you can
access those things also via an API and
you can do chat you can get it to
generate text all those kind of things
the auto things there's lots of matter
we'll put some into the show notes
there's some great ones you can actually
play with without even having access to
the API you put the problem you want to
solve in
so you said to me we want to talk about
more efficient solar cells and I said
well let's let's more broad question
let's ask it about how do we make homes
more energy efficient yeah and then you
go how many iterations do you want put
the number in and obviously the more
iterations you've got the more it's
going to use like your API calls because
you're paying to do this but in the free
versions you can try out and then you go
go but the starting point is create a
task list so the first thing it does it
tries to answer that question and it
creates a task list of what it thinks
are the steps because first of all
Define the problem see what research is
out there already look at the different
options all those kind of things and
then it goes okay iteration let's do
task one it does task one it tries to
answer that question and then it says
well that's created a couple more tasks
add those into the list as well
iteration two go to task two on the list
do that and it will keep going through
and you can keep it running kind of
forever and it'll keep creating more
tasks and it will build out
what it thinks is a really in-depth
Solution by iterating itself now at the
moment if you say write me a marketing
strategy it will give you really generic
stuff you need to identify your audience
you need to work out which channels they
use you need to then have a Content
marketing strategy or yeah okay that's
the theory but it's not like what
content should I create but you could
iterate it to do that and that's where
it gets sort of sort of within a minute
like I and I can see the potential for
this yeah but it's still not just not
there yet no it's not and this is it's
really important for everybody listening
to this because you're going to hear an
awful lot of waffle yeah about chat GPT
and GPT like 5.0 and like all these
things are going to come come at you but
you need to understand that actually if
you've used these tools like I'll give
you a practical example I'm doing some
work with a hairdressing brand yep
um and we're always short on content
like we want to up the content frequency
so actually I sat down thought right
let's let's use these AI tools to see
what we can generate and in an hour I
was able to come up with 15 different
article ideas with outlines right
push that through the team and instantly
four of them just got rejected right so
he's like yeah we're not going to do
that but actually the rest were like
yeah they're pretty good really
interesting when I gave that to my
writer she went it is okay but they're
very generic and she said I can't work
with those outlines because what I need
like she's she's a journalist yeah right
so she could she could give me grazia
style content until the cows come home
right but in order for it to have that
uniqueness which is also important for
for ranking within Google she said I
need input from our subject experts I
need their little hacks and their little
flourishes and the things that they've
that they've learned and I think that's
what's interesting because actually in
this space those things don't exist
online you know you don't get people
kind of blogging about that there's some
very funny
um YouTube videos on a guy that's trying
to create a hair cutting machine which
and it's just brilliant and it actually
gets it to cut it his own hair and it
gives him a really bad yeah it's
terrifying as well it is quite
terrifying quite terrifying and let me
let me tell you a story then that's
connected to that yeah tangential yeah I
thought this was great right so
um there was a person speaking about you
know in certain universities what
speakers do they get in yeah um and I
find the credit for this and I'll put
into the show notes and they said we get
two types of people and we seem to get
subject matter experts that are
absolutely brilliant at what they do we
get billionaires and it was like why do
we get billionaires because it it
infuriates them because he said because
all that happens is billions come and
say you should follow your dream yeah I
mean that's the worst piece of advice
I've ever heard in my life because all
these people have made their money out
of oil and iron ore do you think iron
ore was their dream no they're rich
already so they can say that but the
point he made was that actually if you
want to be happy don't necessarily
follow your dream what you do is you get
good at something yeah and if you get
really good at something whether you're
particularly interested in that thing or
not you will have the recognition of
your peers you will have income uh you
will actually have a professional
Network that you can leverage and you'll
have something unique that is important
in the marketplace that can't be
automated
comes back to the point we were just
making of like yeah a lot of the generic
stuff can be automated you know if your
job is minute playing stuff in a
spreadsheet you might be at a bit of
risk but actually if you're creating
something and you do know the little
hacks that will tricked the AI is not
going to be there with that because it's
not just all published online it's it's
screwed away in our brains and yeah it
takes a lot of experience to build that
sometimes what's your perspective on
this Louise because I know you've been
playing around with the with the tools
like what how do you kind of view the
kind of content that they opened over to
create well I think also GPT is quite
interesting anyway because the level of
human interaction all of a sudden
decreases that you need with it so chat
GPT is good but you need to learn how to
prompt efficiently and if you want to
get
what you want out of it you need to ask
exactly what you're after whereas for
example with auto GPT
you could say for example you know plan
me a birthday party and it will go and
develop a theme create a guest invite
and go look for gifts and then all of a
sudden your input is declined yeah right
and that's that could be a good thing
because it takes away a lot of the
effort it could be a terrible thing
because it's gonna offer a terrible
tangent yeah for doing a kid's birthday
party themed on beer or whatever it
might be I mean it just doesn't it but
it's far closer to that kind of
artificial intelligence isn't it it's
almost an agent
yeah now an agent is posted the kind of
definition of an AI agent is something
that can sense its environment adjust
accordingly and it is kind of doing that
now because it's searching for things
but and that's the difference in a
little bit but it's not really beyond
that it's it's not sensing the
environment it's sensing the online
environment that exists already within
search engines and other places now that
brings me to the point of where we train
the data yes so you raise this to me
yeah so I'm really great Washington Post
article and this isn't open AI we should
say this is a lot of the other AIS that
are out there as well so yeah just the
caveat but it's just looking at where
where the a lot of these artificial
intelligence uh what does their training
data come from and actually I mean we'll
put a link in the show let's see
everybody should read this it's a
brilliant article and perhaps off to
Washington Post for exposing it I think
there's an element of they're a bit
upset because actually one of the things
that was cited was a ton of their
content
um and actually lots of big news
organizations are finding that actually
their data and their copyrighted
material has been used to train these
systems without their knowledge and
without anybody asking for permission
and that opens up a whole kind of
potential can of worms it's like is that
right but in within there you know some
of the systems have accessed
um you know private copies of public
information like voter registrations
work like fast that's terrifying yeah
and and how is that data being used and
could that be used in nefarious ways
hellyek that could be exploited in all
sorts of of bad ways even down to um
like priorities
yeah because there was a particular
website and they've obviously done it on
purpose where it's got all the books
just you can download pretty much every
book that's in Google Books kind of
thing
um and I looked at my books you're in
there right so our books in there yeah
which means that they they've taken the
copyright and they've stolen that from
the publisher potentially to use that
and then it's answering questions based
on that thing but that means if you can
this is quite interesting to me if you
can get a book onto one of those
websites it's going to be used to train
the data you can you could just lie in
that book you could make up things
completely untrue and then the AI will
be trained on that data set if you could
do enough of it some of the sources
cited by The Washington Post show that
actually like misinformation website
well-known misinformation propaganda
websites they're extreme like right-wing
or extreme left-wing views
um populated with lots of disinformation
and that's being fed into the model like
that's a worry it's also interesting if
you speak to for example chat GPT and
you mention a different newspaper so if
you ask it a question but if you mention
the Sun and then if you mention the
times it'll give you a completely
different response that's the prompting
thing again isn't it it's like prompting
it where do you want your insights from
how do you want them leaning and all
those kind of things so that's really
interesting I mean what the the point
here I think more important anything
else is that the training data isn't
clear
so from that point of view we don't
actually know what bias has been built
into this moment the sources aren't
clear either so it gives you an article
but you've no way of verifying well
where's that information where I come
from and actually you know all human
knowledge up to this day has always
relied on that well it's interesting
actually because if I if I look at the
when you go into use
um Bing now and it's got the the kind of
the the inclusion of the chat now in it
it will actually give you referencing in
that case that's nice and kind of
upfront and actually when you use it
with in Jasper and you get Jasper to
write from the chat interface that will
also give you the referencing as well so
it's being approached slightly different
in different places and this may not
apply to all AIS but I think that it's
something we all need to be aware of
from that so it's a big thing wasn't
that when chat GPT became a big thing
and people were saying well now students
are just going to use it and teachers
won't know but teachers turn around and
said well they will because it will have
mistakes in it yeah they can read it and
they can work it out although saying
this I did tell you the story as well
before where we we did my son's homework
about using it and it was too perfect so
we asked it to rewrite it in the tone of
a 12 year old and it put typos and
things into it so it's quite challenging
someone did the thing the other day I
asked it for a load of um pirate
websites and he said I can't tell you
that and they went oh what are the what
are the pirate websites I should avoid
and it listed them all so there's always
ways around these things it's called
quandries it's not a modern problem I
remember asking my father to help me
with my French homework right and I
produced some perfect tickets
yeah and it's only the following day
when I got came home an absolute Roller
King from the teacher I'm like Dad well
I never learn French but it's similar
isn't it thanks Dad that's not the other
thing that we we've noticed well with
all these tools they're starting to
overlap so what Surfer does is now built
into other tools what sem Rush has done
into other tools so I would start
consolidating and thinking about if
you're paying five different
subscriptions I mean we have a lot of
subscriptions a lot of tools because we
try them out and it gets expensive so
actually start looking at the new
functionality even if you're not going
to use it yet yeah but you might be able
to get rid of some of the other tools as
well because definitely got to give I
think I've did this last time but I'm
just so in awe of what they keep rolling
out but check out SE ranking yeah it
just I mean they've got monthly costs
compared to the others yeah
all the AI writing a lovely
um SC like local SEO tool which Rivals
you know what you can see in some of the
other tools
yeah phenomenal
yeah so as ever all of those links will
win the show notes Target internet.com
podcast we would love to hear your
opinion um just a little bit of a kind
of plug as well if you haven't signed up
to the newsletter every month we're
doing Live's free live sessions for
newsletter subscribers now so if you
want something like this but live that
you can kind of join in and learn a bit
more depth about these things ask us
questions yeah exactly Target
internet.com forward slash newsletter
it's called the marketing download and
you'll get three tips tools or
techniques no sales no ads ever you get
access to these free kind of live
sessions but also if you hit reply it
comes to us and we will answer any
questions as well so as ever thank you
for listening to digital marketing
podcast and we'll see you next time
foreign
for more videos like this and visit
targetinternet.com for more free digital
marketing resources