How to Combine ChatGPT and GA4 for Content Analysis

hi this is Andy from orbit media and
this video is going to show you how to
combine
AI chat gpt4's code interpreter with
Google analytics ga4 to find insights
that will help you come up with ideas
for better content by analyzing your
existing content and showing its
Performance Based on things like the
words used in the titles and based on
like its performance in Search and
social uh
in just a few steps I'm going to walk
through it all now this is kind of fun
because it combines tools that maybe a
lot of people haven't combined yet using
some new technology that is actually
very accessible let's jump in
okay I'm here in Google analytics this
is the just real-time report okay the
first thing I want to do is to find the
uh the pages and see their relative
performance so I'm going to come in here
to the engagement section and go to
pages and screens so it's going to start
with just like the page location maybe
as the default and there's like here are
your metrics the first thing I'm going
to do is customize this to pick the
metrics that I like these are the
metrics that I like best so I click the
pencil to customize and then I click the
metrics download I've turned off those
charts because they're not super useful
to us and um I got users sessions
engagement rate average engagement time
views don't really care uh session
conversion rate I do like conversions I
like bounce rate great so if there's
anything here maybe like session
conversion rate that you don't see just
come down here and search for it and
then you can click to add it so this is
how to get the data that we want to
measure the performance of content which
generally we're talking about the kind
of top of funnel metrics middle of
funnel metrics and bottom of funnel
metrics I guess this would be kind of a
middle of funnel thing okay okay so then
I click apply and I click save three
times because that's how it works save
save save and then I'd wait for it to
say change is saved and then I come back
okay I want to see I want to give
AI the data about the titles so I'm
going to right away switch this to the
page title and screen name and I also
want to see the performance of these
things in terms of their traffic sources
which varies so I'm going to come in
here and just choose the traffic The
Source medium for example or let's just
take medium session medium and edit as a
secondary dimension
what you decide to do is to click a plus
to add data from another report right
and add another dimension to this
primary Dimension is going to be my page
title and my secondary Dimension is
going to be just medium
just big picture medium is the broad
order to traffic source is the specific
origin of traffic but this has got all
kinds of stuff and not just blog posts
I'm really trying to do this for my
content marketing so I'm going to do one
more thing to this report which is to
add a filter the filter is going to show
me just the pages for my content
marketing content so I'm going to
include only when the page location
contains the word blog so instead of
choosing the checking a box I'm just
going to type in the blog and then
choose all values containing blog apply
okay now I am looking at the traffic
perform at the performance of each
article
it's basically listed by Page title
against its in different mediums and
against those metrics that we selected
which were
um you know engagement rate average
engagement time session conversion rate
stuff like that
this is the data that I'm looking for
and I'm going to now export this data
I've got my date range set to January 1
and so I've got uh six months of data
here I think you'll need to especially
for low traffic sites you want to choose
a longer date range now I'm going to
click to share this report download the
file and I'm going to grab it just as a
CSV
it's now downloading the file that file
is sitting on my desktop and for the
next step I'm going to import this into
GPT
the the AI that I'm using is gpt4 and
gpt4 is the one you need if you want to
use code interpreter which is the the
version of of uh jgbt where you can
upload files and have a data analysis
this is for paid users you got to spend
the 20 bucks a month if you want to do
this so um if you have done if you are
paying for it you still need to turn it
on currently at the time of this
recording it's uh in the beta features
so down here by your name click on the
three dots
then we get the settings and from the
settings go to Beta features and then
toggle that on code interpreter
currently all paid users have access to
code interpreter they roll it out slowly
I had to wait I'm glad I have it now so
that now we can choose gp4 as our large
language model but not just that but
code interpreter as our model
so the next thing I'm going to do here
is to upload a upload the file which is
sitting on my desktop I think this is
called Data export click upload and it
just put that whole file quick caveat
I'm uploading my own Google analytics
data right into an AI I can never take
that back is that against my company's
policies am I doing this with my client
data is my client okay with that is this
is uh am I breaking someone's you know
the terms of an agreement that I have
with somebody am I training the next
version of GPT is this going to is going
to be in gpt5 now so be thoughtful be
deliberate uh and and uh be be careful
when you upload things to to AI because
uh this is like a you could never take
that back out of the database
okay so I've uploaded the file and I'm
going to give it the prompt my prompt
sounds like this I'm giving you a lot of
Google analytics data about traffic to a
Blog it shows the performance of all the
content marketing since January 1st can
you answer questions about this data set
so I do this first really just creating
a prompt to see if it's able to analyze
the data so it's actually going to start
working on that crunching the numbers
looking at the data and tell me if it
can understand what it finds and it
comes back and says basically yes like
this data set includes the following
columns it's interesting code
interpreter will actually show you its
work it's using I think python code to
try to understand this thing and then
here's the data it's good now it's
asking me now that we understand the
structure of the data set what specific
questions would you like to answer great
okay my next prompt is going to be a
little bit more detailed I like to begin
Promise by telling in by telling AI who
it is first you are a data marketing
analyst and what skills it has you are
an expert at making data-driven
recommendations now here's the here's my
specific task combine the data from the
columns to recommend new digital
marketing topics that would likely have
good traffic and engagement organize the
topics into two groups those who perform
well in search versus those who perform
well on social so you get the idea I
gave it remember that secondary
Dimension I gave it data about search
versus social performance so it should
be able to report that back to me
okay here's the answer which is pretty
fun it says to make data driven
recommendations it's going to do it
separately it shows its own method we're
going to do this separately on Organics
uh uh in Social traffic so for for these
traffic sources they're going to
consider the following actions then it
decides to filter the data set based on
the session medium to see those
separately and it gives me the highest
performing topics in both organic search
and social media this is looking at the
data set and showing me what worked best
in each Channel based on its own
weighting of all the metrics fast
analysis nothing Earth shattering yet
but that's interesting to me one thing
that really surprised me here is that it
didn't just choose the titles but it
made them into links I did not give it
the URLs it says the actual URLs are
inferred based on the title and not
provided in the data set
it's showing things that I didn't give
it I'm surprised by that uh and
generally pleased by that um amazed I
guess
now my next prompt is going to give me
even deeper insights because unlike
search which is only going to retrieve
information that exists I'm going to ask
it to create brand new information to
come up with brand new topics so my next
prompt sounds like this you can see it
here suggest 10 new topics that are not
in this data set divide them into two
groups good for search and good for
social
cheerful little guy uh sure I'd be happy
to suggest new topics for your content
marketing good for search these are
things that I had not written about
um these are some topics I hadn't
covered so this is uh not bad ideas here
already good for search and good for
social uh generally pleased with what
I'm seeing here
so if that didn't impress you and you
didn't think that that was very useful
keep in mind we only gave it a couple of
prompts the most insights that I get out
of AI are when I go much deeper into the
conversation and prompted you know five
or ten times uh what I did in this
conversation was actually go much deeper
I asked it to make me a scatter plot
chart which I thought was interesting
but everything was crowded here around
that that one axis so I asked it to make
the chart a log scale chart which it
redrew showing content performance
engagement versus traffic I was not able
to make an interactive chart where I
could click on the dots and I could see
the article so that's not really that
useful to me it it explained why then I
asked it to find articles that did well
in one channel but not another it
inferred what doing doing well and doing
terribly was and then gave me a
um and then gave me an answer these are
the ones that did great insert in Social
but not as well in search showing me the
data behind them
then I asked it for the opposite
um or
then I asked it for more details to do
the same analysis but based on
engagement and not traffic which it did
and did a pretty good job of
again showing me the regular the
relative engagement score for each of
these articles
I also asked it which articles had uh
above average engagement but below
average traffic which it did in a
uh which it did very quickly that's a
useful report because it shows you what
you can keep promoting these are things
that are very engaging but haven't been
seen by enough people
uh and then finally it just kept going
deeper I asked it to create a brand new
metric called awesomeness and to blend
traffic and engagement into one number
and then show me the top five articles
in terms of Awesomeness which it did but
I felt that these were
um a little bit heavily weighted uh in a
direction that I didn't like so I asked
it to change the awesomeness metric to
reduce the weight on traffic and then do
the analysis again which it did you can
see it's showing its work adjusting
awesomeness is uh and it gives me back a
list of um uh the adjusted awesomeness
score for these for a brand new metric
that it just created and that I asked to
tweak
um this is then I start to go deeper I'm
asking it questions about what are these
have what are these articles that did
well these out performers uh have in
common in their titles and it's telling
me what common phrases appeared in there
it's telling it's analyzing the length
of these titles which ones um that the
there's a correlation between use of
numbers and performance in a title it
also shows the correlation between use
of questions uh it's loading python tool
kits to be able to come back and show me
what the words that I use most often and
the use of numbers is in their
correlation and then finally a fun
prompt uh listen to this one I
understand the differences then titles
between articles that worked well in
Search and social what prompts should I
ask you that will be most likely to help
me find those insights and it gives me
some prompts that I could ask to go
deeper so kind of fun hopefully it's
helpful you can go into ga4 and Export
this data and then import into code
interpreter and really just start to
talk to your analytics data uh give it
give it both the page titles and the
traffic source and then you can start to
try to make correlations and find
insights that way ask it for things that
don't exist in the data set like to
suggest new topics
um really fun to combine these new tools
be the first on your team to bring these
insights to a meeting and let that start
a conversation and see if you can let if
AI can help you sharpen your focus on
certain topics and iterate on your
content strategy for better results
Forever After again Andy from orbit
media if you found this useful feel free
to pass it along to a friend we'd be
grateful and we'll see you next time