ChatGPT ads deliver you a lot of
decent clicks, good CPM, good CPC, good
CTRs, but the challenge is downstream.
The conversion rate into trial,
paying user, purchaser is very low.
One study I did across eight apps,
six of them just total awful.
I didn't get any conversions.
Hey, my name is Jacob Rushfinn,
and I'm the CEO of Botsie and the
host of the Price Power Podcast.
We're talking with Gabe Kwakyi today,
who was actually our first guest.
Gabe is a mobile app OG.
He was previously the CEO of Lingvano,
the number one sign language app, and he's
now doing a mix of consulting, advising,
and fractional head of growth for apps,
as well as building some stuff on his own.
Uh, this is a pretty cool episode
because Gabe has been- actually been
testing ChatGPT ads, and we get to learn
from his real experience in testing
about how they were, what works today,
and what the future could look like.
Hey, Gabe.
How's it going?
I'm so excited to have
you on the podcast again.
Thanks for joining.
Yeah, Jacob, thanks so much for having me.
I'm excited as well.
This is great.
Yeah, we've got a fun topic today.
ChatGPT ads are super new, but
people are starting to experiment
with them and test them out.
Yeah.
I recently saw that you'd been
doing that, kind of digging in and
learning what's working, what's
not, and you have a presentation
you recently gave on ChatGPT ads.
Yeah.
Can you tell me a little about what
you got started on a- and why you
were testing out, uh, ChatGPT ads?
Absolutely.
And I think, like, anyone listening
to this is probably gonna be like,
"Hmm, yeah, like I, I use ChatGPT.
I've heard they're running ads."
And for me, it was like maybe the second
part, the, the interesting part is there's
900 million weekly active users, probably
now a billion, on ChatGPT, and that's
a lot of people you can advertise to.
And moreover, you know, I think s- so many
of us, like mobile marketers or marketers
in general are like, we want alternatives.
We want an alternative to Meta.
Everyone says, you know, "I'm
spending 70%, 90%, 50% on Meta."
And then if you're iOS, you have
Apple search ads, and if you're
Android, you have, uh, Google ads.
But just having another platform
in the mix, uh, would be great.
So that's what really got me
excited about ChatGPT ads.
And I have been running, um--
So I, uh, have kind of a little
portfolio of my own, own apps.
I work with some other apps, and so, uh,
I've done a few tests, um, have explored a
few different directions, and have talked
to a lot of people who have also run
ChatGPT ads, so got some wisdom to share.
Cool.
I'm pumped.
Uh, I'm, I'm super curious myself.
I've clicked around.
Uh, I got connected in the--
set up in the platform, but I
haven't run the ads myself, so
I'm, I'm super curious to learn.
Yeah.
Everything is so new.
Maybe you can just start with
a quick overview of how things
work and what's possible today.
Yeah.
Um, it's kind of refreshingly minimal.
If you pop into the, the platform--
I mean, I guess also ChatGPT is,
is pretty minimal, so, uh, uh, the
design decisions carry over there.
You know, Meta, like there's stuff
everywhere, so many configurations.
Google has gone down this weird,
like modern, uh, view where
everything's just really huge and
it's hard to like click around.
In ChatGPT, you've got the campaigns,
ad sets, um, ads, you know, that's
quite similar anywhere you go.
And then you've got a few configurations.
So you have country, platform,
um, iOS, Android, desktop.
iOS and Android can be
split into web and app.
Uh, and one of the stats that was
interesting was that about 50% of, uh,
ChatGPT users are on mobile, I believe.
Um, we'll pull up the, the, the numbers
here and, and confirm, but yeah, it's 46%.
Um, and so yeah, you can target, you
know, uh, by, by user, uh, platform,
bid, conversions, um, that you
can optimize for if you integrate
through an MMP or have the pixel.
And then you can create your ad.
There's a image right now
only, headline, a description.
And the most interesting
part is context hints.
That is new, and it's ChatGPT native,
and it's different from keywords or
interests or audiences or behaviors.
Um, or Google UAC where you,
you know, don't have that.
But you do have this targeting
mechanism, but it's kind of a black
box right now, and that's what I
found the most interesting about
ChatGPT ads, these context hints.
What, yeah, what is a context hint?
Is thatâ¦
I, I know in the context of LLMs, there's
all this talk about, like, context.
Is- Yeah ⦠a context hint like a,
like a search term or like a question?
Like h- how do Iâ¦
I guess explain it to, like, what
it is, but then maybe also how I
should think about in, in, in context
of other, um, uh, ad platforms.
Yes.
Honestly, I think everyone is still
trying to figure out what is a good
context hint to use for your ChatGPT ads.
What it is, is ChatGPT says, like,
it's a way of describing your product
or the keywords or the kind of way
that users interact with your product.
But exactly how to structure
and format it is, is still kind
of, hey, it's a very open thing.
So in a way, it's nice that rather than
going, you know, the UAC route where
you don't have anything to target,
you have context hints which allow you
to target infinite different things.
You're going-- You can have
super long tail context hints.
Um, and so that's nice, uh, as a marketer.
But yeah, like, Iâ¦
You could, you could
put keywords in there.
I've tested that.
It's done okay.
You know, like a couples question app.
Um, you could put, uh, kind of a
description of, uh, of the conversation.
So conversations, that's probably the
one that I used the most at first.
Conversations where X, where
the user is discussing their
relationship, where the user wants
some ideas for dates, uh, coming up.
You could do a goal, you know, userâ¦
Conversations where the user's goal
is to Go on a, a, you know, to carry
forward this dating example, go
on an interesting date, you know,
not just another boring thing.
Or conversations where
the user wants something.
Um, but yeah, it's, uh,
it's, it's still pretty open.
Like ChatGPT's example is, okay,
don't just use running shoes, quote
unquote, use triathl- like triathlon
training running shoes for users
who are, you know, training for an
upcoming, uh, triathlon or Ironman.
Okay?
So then you're- Yeah ⦠ChatGPT is
saying like, get more into the specifics.
Go more long tail.
Um, what you end up finding is just
that it's, it's kinda hard to know, um,
before you test it, is there volume here?
What's the cost?
What's the CTR?
What's the conversion rate?
And it just goes back to this old like
test everything and segment things so
that you have one or related context
hints in each ad set, and you can
identify this one is performing in this
way and that one that way, is theâ¦
One of the challenging parts is
there's not a lot of suggestions
that ChatGPT gives you.
It doesn't tell you high
volume, low volume, no volume.
Um, it just says they're,
they're even optional.
You don't have to put a context hint in.
Um, but I would certainly, uh,
always put them, uh, in there.
And Yeah, it's, it's back to this trial
and error This is what you mentioned
of, like, keyword volume, of audience
volume or traffic for these terms,
or, you know, interests, kind of
where-- is where my mind went back to.
Did you do any pre-planning or research
of these types of phrases or terms on
other platforms, where maybe on Google
you can type in similar search terms and
go, "Okay," like, "They're-- this type
of phrase, there is people searching for
this, so can I-- I can assume that if
people are searching for this in search,
there's probably some volume or interest
in this type of thing, but I'm gonna
put it in a more conversational tone"?
Um, did you- Yeah ⦠did
you do that at all?
Yeah, that's, that's an interesting
approach, and I actually didn't do that
because I haven't put too much budget
into ChatGPT ads- Mm-hmm ⦠for each
app, just to make sure, like, what's
the performance like before I'm r- you
know, really ready to spend thousands
a day, um, or even hundreds a day.
So, I actually think volume is not
really the, the, the issue or the
thing to focus on with ChatGPT ads.
It's more so just, like, does it
work for your product, and can you
find the context hints that make
it work or make it work better?
Um, and so you, you really
don't want too much volume.
And what, kind of spoiler, what, what
I've found, what I've found ta-talking
with others, is that ChatGPT ads
deliver you a lot of decent clicks,
you know, good CPM, good CPC, good
CTRs, and you can get, you know, for
apps, like a, a decent install rate.
But the challenge is downstream.
The conversion rate into trial, paying
user, you know, purchaser, is very low.
And I've tried a d- bunch of different
things, and I've found, like for
instance, one study I did across eight
apps, six of them just total awful.
I didn't get any conversions.
Two of them got conversions.
They were just kind of expensive,
and they were volatile.
I've been running with, with a, a phase
two test with another app, and I'm
getting more conversions this time,
but still they are few and far between.
So, I think the main thing here
is, like, trying to find the most
relevant, um, situations- Context,
uh, to show your ad to users.
And yeah, just trying to kinda like
filter out the curiosity clicks, as
I heard someone, uh, describe it, and
get to the people, the right people at
the right time, in the right time in
the conversation, where they're ready
to click download and make a purchase.
And I think actually product ads, uh,
like, or D2C probably are gonna see
more success with ChatGPT ads than apps.
Um, unless you have a very, very
relevant, you know, high volume app,
uh, where a lot of people are, like,
really trying to, to, to solve that
thing and, and download that thing.
Um, when is the right time for a
person to, like, download and, and
start using a meditation app or a,
um, you know, couples question app?
Um, so context hints I
think is, is really the key.
Yeah.
It's, it's so interesting
to think about that.
Two th- two things come to mind.
Um, one is that ChatGPT ads
are only running for users on
free and Go plans right now.
Right.
Um, and so potentially this is
not The, I mean, this is, this is
still a huge swath of users, but
like these people are not paying.
Yeah.
Uh, uh, f- or, or Go, I
think is super cheap, right?
Really discount it.
Uh, and, and s- yeah.
And so maybe you're already targeting
price-sensitive people that are,
are, are gonna purchase at lower
rates, don't wanna pay subscriptions.
So- Yeah ⦠I don't know.
I don't know how that impacts things.
Uh, but then also in terms of this
targeting, like that's what Facebook
or, or Meta's so good at, right?
Of like figuring out the people that
will convert, and then, but you're,
you're sending back those conversion
signals for the algorithm to optimize.
So- Right ⦠potentially it's potentially
not the right users, but it's probably
l- less that 'cause there's so many
people m- potentially just they, they
haven't really fine-tuned or figured out
that targeting algorithm, uh, quite yet.
Right.
I think this, this is one of the, the
things that's been most interesting
to talk about with different people
about when is the right time.
Kind of backing up, I feel like
ChatGPT Ads is going to be a really
strong channel, and it's more a
matter of when will that happen.
And so this, this like conversions
history, conversions mapping, is what I
think is the thing that's standing in the
way from ChatGPT Ads working well now.
But by next year, maybe they
have enough data for it to work.
This company knows how to work with data.
That's obvious.
And they have a lot of data on their
users, but like you're saying, they, they
need more data on the conversion side.
What happens when a
user goes off platform?
Meta did a very, very good job of making
sure that every advertiser was tracking
data and passing it back to Meta so that
their, you would get better targeting
data to optimize your campaigns.
And so then Meta had a ton of data.
It knew, it knew what people were
the right fit for each product in
what point in their life or day, and
at moment when they were ready to,
to convert to a downstream action.
Um, and so it knew who also
were subscribers, because
you would label this data.
This, you know, is a trial.
This is a purchase.
This is a content view.
So ChatGPT is new to the game with
direct response or advertising.
It's only been running since May of
2026, so it's, you know, it's still
not, it's not half a year at this point.
And they, they move very fast.
They've shipped a lot of features
in a very short period of time.
But I think they just need more time to
gather that post-back data and confirm
of those one billion weekly active
users, who's interested in what products.
Um- And eventually they'll
get there, I think.
Yeah.
It, it also, you know, I, I'm trying
to think about parallels between the
type of, um, the type of intent, the
type of action on the platform versus
search ads versus social ads, right?
Where, you know, there, there's
certainly some types of products that
work much better with social ads and
s- and some types of products that
works, work much better with search ads.
You know, the obvious parallel is
that it's much closer to search ads
where you have a high intent person.
Um, but it's gonna be much harder for
you to, uh, potentially drive traffic
for your fitness app with search
ads than your, um, travel app maybe.
'Cause so- someone has a specific
goal in mind with search.
They're, they're coming in,
they want their problem solved.
Right.
And so it feels like maybe there'll be
learnings there as well about, you know,
potentially there's certain types of
products that just won't work as well.
I also think, like, you know, f- Facebook
or, or Meta has built out this, um,
uh, knowledge graph or, uh, um, you
know, profile graph of all the, of
this, all these people, who they are.
The, the more usage over time that OpenAI
gets and understands about this person,
who they are, potentially there's, there's
multiple types of ads they can create
where some are- Yeah ⦠more based on who
you are, and you can target people there.
And then some are more
search-based, right?
That'll come up in chat, that'll,
that can be more higher intent.
And so, um, yeah, it, it feels
like, to your point, like super
early, the- not enough data.
Um, how do you actually
target people correctly?
And, and, and we'll, we'll
learn a lot, uh, over time.
Yeah, absolutely.
I, I think just there, there's so
much interesting, like, opportunity
that ChatGPT a- uh, has here.
Also, they borrowed or they
brought over a lot of meta talent.
I think their, their CTO of,
um, ChatGPT Ads came from Meta.
Their head of monetization
for ChatGPT came from Meta.
Several other VPs and
high-ranking folks came from Meta.
Like, they know how to
make an ad platform work.
So it's just having enough time to
plug things in and get things working.
Um, but yeah, I, I, I think that isâ¦
A, a couple things you mentioned, like
one, like, okay, it's, it's probably
closer to, to search than to pure, you
know, display or even, like, um, social.
Um, and, and yet it's, you know,
it's, it's not all the way down
the funnel like, like search.
Um, and so, like, kind of w-
when is that right moment?
It is a person who has an en- intent
to answer a particular question.
And so actually one thing that I've
seen in, in research is this concept of
turns, and that a turn is when you do one
back and forth of a conversation unit.
So I write a question, ChatGPT answers.
That's a turn.
So turn depth also is gonna have a
significant influence on, like, is
it the right time in the conversation
when the user is ready to take action?
And that's something that I'm sure
ChatGPT-- Like, that's something ChatGPT,
ChatTBT can work with a lot to see where
do we, where do we put that, uh, that ad?
Depends also on how long conversations go.
People have said they see an ad
right away, like, you know, like a
s- uh, the classic example of travel.
Like, hey, like, what's the best blank?
What's the best flight price?
Okay, that turn automatic-
like, already is ready for, for
an Expedia ad or, or whatever.
Um, or I was looking the other day,
like, what's, what's the he- I'm looking
for a blood pressure monitor machine.
Should I use this one or that one?
Immediately, you know, if the product--
I'm like, fine, if, if it comes up and
it's like, this is recommended, you know,
okay, like, then, then I'm, I'm ready.
But, you know, for apps,
yeah, it's like when isâ¦
Again, this, this form the context
of like, when is the right time for
ChatGPT to show the ad to the person?
And in, in the deck that I, the
presentation I did and the conversations
I have, it's like, okay, like, think-
Put yourself in the shoes of a person
who's having these conversations, and
like, when is the right moment for them?
A lot of times people, at least I use
ChatGPT, I'm like, "Give me ideas."
It's like, "Hey, I want
five ideas for blank.
Hey, I want 10 ideas for that,
and then I want more ideas.
No, those aren't good enough.
I want more ideas.
Give me more and more."
I'm not really, I'm not really,
like, putting too much effort
in, but I'm open to suggestions.
And so you, you don't, you, you
probably get high volume if you're
like users who want ideas for kind
of blank topic, but is it the right
time to serve an ad to that person?
So, like refining users who,
um, want ideas for meditation,
uh, you know, for instance.
Okay, one of the things that I've, uh,
seen that's worked a bit better is, for
instance, like, um, uh, rejected ideas.
So user has rejected ideas for
blank, and they want better ideas.
Okay.
Well, if then ChatGPT is like, "Hey,
here are some better ideas," and your ad
serve is right there, it's part of this
like, okay, I'm, I still want to solve
this problem, but I want a better idea.
Okay, I'm more open to considering this.
In Google, you don't have that ability.
You just have, you know, a bunch
of websites that are showing.
Even, you don't even have a bunch,
you're just at the top, like one
query, one question With whatever
Gemini's integration and everything
is, like I'm sure Google's working
out, like where in the conversation
do you, you know, insert the ad?
So too with ChatGPT.
Or another one that, um, has worked
well is this, uh, I'm calling it the
like layup insertion, and that is
where, uh, it's kind of a template.
User claims blank didn't work, and
they want better ideas, more ideas.
So they're like saying, you know, they're
qualifying themselves, like, "I'm open
to better ideas," uh, you know, "I, and
I'm, I'm ready at this, at this point."
So when you serve an ad there,
that's when I've seen that, you
know, I've been getting conversions.
Um, so just using keywords, it's not,
it hasn't, it sometimes has worked,
but it's not, it's not specific enough.
Just using conversations where the user is
discussing something at the right moment.
Um, it'd be interesting to have
ChatGPT give you a, a chance, like
a, a tool or a feature to control the
number of turns in a conversation,
uh, before your ad shows up.
Um, but anyway, these are just,
uh, uh, a few interesting nuances
of serving ads in a chatbot, uh,
that don't come up elsewhere.
Yeah, that's so interesting, and
it makes a lot of sense, right?
Where someone is unhappy with the answers
they're getting, they still wanna solve
this problem, and so in that moment,
they're potentially open to going
somewhere else to get that answer or a
solution, where I'm going to ChatGPT to
answer my question, it's not answering it.
I'm getting frustrated.
Well, uh, enough of this.
I'm, I'm, I wanna go somewhere else.
Hey, here's an option, uh, uh, for that.
And, and I think, yeah, that extra
level of context, uh, in the, the
user's mindset, I wonder if you'll,
over time, get things like sentiment.
Hey- Yeah ⦠I wanna serve ads where
with this question, with a positive
sentiment, and I know p- when people
are happy, they're more excited,
or hey, someone's frustrated now.
Um, someone's unhappy.
They're lookingâ¦
A- and this is typically when a user,
uh, uh, is gonna be more receptive
to, to using my product as well.
There's like so many possibilities,
but it also goes back to the problem of
targeting and conversion optimization.
When there's so many more possibilities,
optimizing who to show an ad to
when, based on the algorithm,
uh, clearly gets more complex.
So it'll be so interesting to see
how they, how they solve this well.
Absolutely.
Um, yeah, and, and it's probabilistic too.
So it's not you put this in, and
it gets interpreted by ChatGPT.
So like that's another thing that
I'm, I'm trying to look at, like do
I see the same results, you know,
for the same context hint over time?
People have mentioned, and
I've seen it as well, the, the
campaigns are very volatile.
One day they'll be performing okay,
and then it's, you don't really change
anything or you change the bid by
like, you know, 5%, and suddenly like
things change very significantly.
Um, so that's, that's a nuance of-
for ChatGPT to deal with, like where
there's no de- What parts of their
ad serving system are deterministic
and what parts are probabilistic?
Yeah.
The, it'll be so interesting to see
how it evolves and I, I think it's,
um, seems like it'll still be a little
bit before potentially it's high
performance, high volume, predictable.
You, you also, and, and we- we'll,
we'll sh- we'll show this, um, in
the video version, but you had other
kind of context hint categories
you thought were good to test.
Any others there that you think,
uh, you wanna call out that,
that were, uh, interesting?
Uh, yeah.
I mean, I think, like, one interesting
one is this, like, um, decision support.
So, you know, we were talking about
rejected ideas and being open to new.
It may or may not be good because a
person might or might not be just in that
mode of like, not good enough, not good
enough, not good enough, like nothing
that they get is going to be good enough.
Or it might be that
eventually they're ready.
But this decision sh- decision support,
maybe that is when they're more open.
They're-- So this one is like, um, users
choosing, you know, between several
different ways to solve problem X.
Users choosing between, or users
trying to decide between and understand
several different ways to manage stress.
And maybe what I think will be interesting
in, in your testing is like, how do you,
you know, add a little more context?
So can you say, "And they've
accepted suggestions before"?
I don't know if this, if ChatGPT is going
to read this and understand that the
person has clicked, you know, the result,
or they've thumbs-upped the conversation,
or they said, "That was helpful."
These are all things that are kind
of fair game with context hints,
but it's like s- it's limitless.
Like, where do you kind ofâ¦
In a way it's, um, it's overwhelming
'cause, like, there's just so many ways
you can target people, and theoretically,
like, the better, th- there's always a
better way to, to make things work more.
So you just have to put a lot of like,
you know, budget, um, behind like testing
before you really feel like, "Yeah,
I've tested a lot of different things."
But yeah, um, decision support, you know,
help me with this, you know, is, is one.
Like recency, recent issue trigger.
User just went through blank
bad thing, and they're looking
for advice or suggestions.
Uh, again, for our like, you know, uh,
dating app, um, questions app, um, user
just went through, um, a ba- like a,
a bad breakup or like a, an, uh, a,
even like a boring weekend, and this is
like a kind of negative or not wanted
thing, and they're looking for, you
know, new ideas or new suggestions.
So, um, a lot of these like focus around-
It's- Oh, yeah, like the, the ideas that
the person is asking ChatGPT to give them.
It's such a new paradigm of how to
think about, um, ad targeting and ad
placements, where really trying to-
empathize and, and think through what
the user's going through in that moment
and, and how do you capture on that,
and you're not even sure if you have
that context available to target, but-
Mm-hmm ⦠uh, i- i- it's, it's quite
interesting and, and completely shifts the
way you can think about, uh, advertising.
Oh, yeah.
Um, it'llâ¦
Yeah.
So it's fascinating.
Uh, how- And I think a lot of peop- like,
people should try it, and should also
know that just because if it doesn't work
for you downstream, you know, that's th-
that's, uh, most people's experience,
but your product might work well for
it, and you might find the context
hint that works really well for it.
And I'm kinda like suggesting to people,
like, try it every so often, you know,
every couple of months or something.
It might, likeâ¦
You don't, you don't wanna miss the boat
when it's actually rocking and rolling.
But right now, for most advertisers,
you know, it's probably not gonna work.
But as we're discussing, like, the,
the, just the conversion mapping,
as more advertisers use it, ChatGPT
will have more conversion data.
So it might go from being like,
"Meh," to, "All right," to, "Oh,
yeah," you know, within the span
of, like, three, four, six months.
But for, you know, there, there are some
people that are making it work right now.
But, um, yeah, it's very, very
fascinating- One- ⦠use case One thing
I, I was just thinking about as, as you
were saying that is a lot of times you are
customizing your acquisition or conversion
funnel based on where the traffic is
coming from and the type of intent, right?
Where if your product is driven mostly
by meta and social ads, you often have
this longer onboarding funnel, um,
trying to kind of build up intent.
Some- you've interrupted someone's flow,
so you have to build up this excitement.
Uh, you've, you've taken them away
from, you know, scrolling their feed.
Now you have to convince them that
your product is worth buying, and
so it's this longer funnel- Mm-hmm
and more emotional, more
investment building.
Where with search, oftentimes if, if, if
Google is a main channel for you, um, or
SEO, someone was looking for something
to get done, and a lot of times your
onboarding in your app is super short.
You get right to the value.
Someone purchases immediately.
Yeah.
Potentially, we're gonna see this
maybe hybrid and, and maybe more
personalized conversion funnels based
on intent, where you're coming from.
Like, can we pass this traffic or
c- can we pa- like, you know, for,
for search ads, right, we can, Apple
search ads, we can pass keywords
in and we can potentially customize
experiences based on those keywords.
Maybe this is also part of, um, driving
actual better conversion that we'll learn
as well, and it's not just the targeting.
It's also the, uh, uh, kind of
post-acquisition experience of how
do we actually convert this user
based on where they're coming from.
Like, if we could get the term or
search or context passed through into
the app based on where they came from,
like, you could create amazingly, uh,
personalized experiences that would,
that would probably convert better.
Yeah.
Yeah, I, you know, the, the, the shift
in, in mobile UAs towards web UA and in
landing pages creative, landing pages
onboarding, landing pages product.
Um, so yeah, I think you're right.
That's another lever that, that would
get kind of more engagement, more bottom,
more conversion out of the same users.
Um, even warm them up, uh, and,
and get them ready if they're from
that pa- that particular context,
or drop them immediately, um, if
they're from that particular context.
Or drop immediately at the purchase point.
What information
can you passâ¦
Yeah, what information can you
pass today and kind of, is it
all just UTM parameter based?
And if you, so you put in different
links through different ads, and so
if it's on web, you could s- you could
create different funnels or different
flows today based on your different ads?
Yes.
Um, it's interesting, there's onlyâ¦
I'm looking in the UI right now.
There's only five parameters that
you can pass, um, from ChatGPT.
So you can't, for instance, put
like source, uh, for instance,
but you can put campaign ID.
Uh, or I guess these are, um,
yeah, the, the dynamic ones.
Ad group ID, ad ID, ad account ID, and
one called opref, which I'm actually
not exactly sure what that means.
I'm actually gonna look it up while
we're, while we're doing this, um, and
see if I can get a definition on it.
Cool, cool.
Um, yeah, that, that'll be interesting
to understand and how this evolve.
I guess with the ad IDs, you can
get pretty specific and pretty
nuanced if you have different ad
IDs and are, you know, customizing
experiences based on that ad ID.
But, um, yeah, the, the, the more nuanced,
the more, the better you can understand
where this traffic is coming from.
Um, and, and so kind of, kind
of related to this, you know,
experience elements, um, whatâ¦
On, on the design side of ads, you know,
we're targeting context windows and,
you know, questions and conversations,
but, you know, what, um, what design
elements have you seen work in your ads?
And I guess maybe, maybe just start
by sharing like what design elements
are possible, like what, uh, uh,
what customization do you have and,
and then what has worked well or
at least better that you've seen?
Yeah, and, and I just, uh, found out that
op ref is, uh, ChatGPT is like GClick ID.
It's like the, the- Ah, okay
click or their, like, refer, I guess,
I don't know, opportunity or something.
So I guess it's good to
include that as, as well.
But what ads work well, um, at first
I, I, you know, tried to use ads, the
images, as a way to like, you know,
really try different things, see what
works, and it was interesting that just
the, the ad or the image that works
the best is very simple, just an icon.
Put your app icon in there, and
that works best as the image.
I was porting some over from Meta.
I was bringing, like, creating some
new ones, like trying screenshots and
different things, and it was kinda no
contest, um, from the tests that I ran.
And it, it kinda makes sense
if you, if you look at a,
what a ChatGPT ad looks like.
It's small.
The image is like a small
little square thing that you
really can't read that well.
Um, and so if you're trying to
put, like, images and stuff and all
these different elements in there,
like, or text and all this It's,
it's not a, it's not a banner ad.
It's more of just like a companion image.
So you can't read, you know,
what you're putting in.
The user can't read what
you're putting in there.
So it's either wasted space or it's just
complex and the person's like, "Eh, you
know, I don't really-- That doesn't look
like, you know, it doesn't look like
the answer to my, to my question here."
But a, an icon, people
are used to seeing icons.
You know, it's like the, the, the like,
I guess, the preview image a lot of times
is the brand, um, icon or something.
Or, you know, it's the exact
like product or something.
So probably matches, for a website,
probably matches what a person is
expecting to see when they like see an
ad pop up, and they don't want like,
you know, a banner ad or something
you would see in Meta when you're
scrolling through Instagram or Facebook.
It's, it's very different.
Or you're scrolling in the App
Store, you expect screenshots.
So, so put an icon in, in ChatGPT ads.
In terms of the, the headline
and the description, still
trying to figure that out.
I think I'm generally more focused on
context hints, um, to try to kind of using
standard, um, ads to see w- I think more
of the, the, the lever that you can really
get the most out of is the context hint.
So I'm focusing there, and
text, um, is kinda coming next.
But, um, I think looking at actual
examples of ChatGPT ads, thinking, like
having conversations yourself and seeing
how does ChatGPT reply might give you
some ideas, you know, empathizing with,
with the person on the other side of this.
And it's probably simpler, you know?
It's, it's probably not going too
crazy, just, you know, kind of a simple
answer, um, in nudging them to just, you
know, click and like get the solution
or the idea that works for them.
But you don't have much room either.
Like it'sâ¦
Let's see.
I think on the, um- I also get this, uh,
error message that it's like even when
your character, your text is within the
character count limits, it still says,
"May be truncated on some placements."
So you don't have a really good idea
of how your ads will actually show
up, um, unless you, you know, you
have the, the preview image there.
But yeah, there's just a, a,
a few quirks of the platform.
So title is up to t- 50 characters.
Description is up to 100 characters.
Got it.
Got it.
And, you know, I, thinking about this kind
of search analogy, uh, in terms of search
ads analogy of why this works, potentially
it's more about your ad appearing at the
right time for the user versus the perfect
creative that captures them, right?
Theyâ¦
Y- you have to figure out which
context hints are gonna generate the
right intent at that right moment,
and then it's just being there.
Similar to search, you're figuring
out the right keywords, and as long
as your listing appears and it's
relevant and it solves their goal, you
know, it's simply a text placement.
It, itâ¦
There's less creative optimization.
Sure there's some, you know, kind of
creative text, but it, it's more, uh, um,
kind of being there at the right time.
Right.
And, and that's really
what drives performance.
And, you know, potentially people,
you know, general- are, are, um,
resistant to ads, so if you feel
less like an ad and more just like,
"Hey, here's the, the product."
Um, and, and then I also wonder if
that creates more, a more cohesive
flow when they go into your app, right?
So they see the app icon in the ad, they
go to the App Store, they see the app
icon, you know, subconscious confirmation
I'm in the right place, and, and continue
down the funnel potentially as well.
Yeah.
I think that's, that's spot on.
Absolutely all of that.
Yeah.
So I guess big question, you know,
is it worth testing right now?
Are, are you doing more testing?
Are you keepâ¦
Are, are, are you, uh,
pausing your testing?
Should people try this out?
I think in your presentation you said
80% or more of tests will, will fail this
year and, and maybe wait a little bit.
I think you said you ran
tests across eight apps.
Um, anyway, I think there's,
there's a lot of times alpha in new
platforms and, and just figuring
out if there's opportunities there.
Yeah.
Um, but yeah, w- what would you say?
What, what are your recommendations?
Yeah.
It's, it's, it's been interesting.
Uh, since doing that, that talk, uh,
I, I've talked to people afterwards,
and pe- like, people working in, in app
companies or, you know, direct in the
publisher, most of them have expressed
an interest to keep go- to do it anyway.
They're like, "Yeah, well,
like I, you know, I wanna see.
I'm curious."
And I think a lot of people also
have this like, "I would love for
another platform besides Meta or
Google or Apple, so I wanna try,
you know, the, the latest thing."
And- The people that have said like,
"I don't know, I'm gonna wait,"
are, are more on like the, the
agency side or the service provider.
They were like, you know, is it-- I
understand, like I'm not gonna put
my credibility and put the budget and
like have this whole test set up if
it's probably not gonna work right now.
So they're gonna wait.
So I think it's probably
not gonna work for you.
You know, that, that is
the, the just average thing.
Um, but it's still very, you know,
interesting to test, and if you have, it's
ve- Like one of the things that's great
about it is it's super easy to get set up.
There really is-- Like
you don't need creative.
You know, you don't need really much.
You could even-- You don't even--
You could use like a tracker
link and just, you know, drop
people, use that in the ad URL.
So you don't need, you
know, a landing page.
You could integrate it through your
MMP and get the tracking going.
That's very easy to do.
And it's 25 bucks a day,
um, is the minimum ad group,
uh, or campaign, um, budget.
So you can test it very easily.
Um, and I think if you're just curious to
see like how does this work and get like
an early head start on how do I, how do I
use a context hint, how does this platform
work, then it's worth it for the learning.
That you're saying, "Okay, here
I'm, I'm familiar with this.
I have some idea.
So when I see or hear that it's
working well, I'm ready to go."
Um, and I think there's
a benefit to, to that.
If you have limited time and you're
like, "I need another-- I-- When
I test channels, I expect them to
work," it's not the right time.
Don't run a test.
But again, I think like every so often,
you know, I, I, I personally have, have
conviction that, that it's going to work,
and so it's like you don't, you don't
wanna miss the, miss the boat when it,
you know, really does start to work.
Who knows what ends up happening?
But yeah, that, that would be my sense.
I'm gonna keep testing it, you
know, at low budgets like here
or there, like with new apps that
I work with or what have you.
Like I think for certain apps,
like it can work quite well.
Um, and I'm trying to be more
selective about which apps,
um, it can work well with.
Um, but there were, there was
something else I saw in a LinkedIn
post the other day that was, that
was interesting as well, and it was
that this, uh, this other advertiser
has said they put 600 bucks behind it
and they didn't see any conversions.
And I've seen that a lot as well.
Um, like you put hundreds behind
it, zero or one conversion, and
you're like, "Oh, that sucks."
But this person said that they
saw a lot of organic ChatGPT
referrals and traffic coming in.
So they were like, all right, you
know, it, it seems like it's h-
looking on, on the, on the other
side of this, the platform, uh,
engagement, that's benefited.
So it could be that you're driving
more algorithmic like benefits
to your ranking in ChatGPT.
Like search ads, when it first came
out, I was like one of the f- like the
first on it to test it, and I saw there
was a full correlation between search
ads traffic and organic keyword ranks.
They fixed that, you know, after a
little bit, so, you know, who knows?
Like maybe this is kind of like
early benefits and you do get better
visibility by running ChatGPT ads.
Uh, so if you do test it, check your
organic, you know, referrers, like
whatever tool you're using for that,
or just the how'd you hear about us,
and you might find that people, um,
are coming in that aren't attributed.
Um, so that was just another
interesting thought there Yeah.
Was, was this also a app, um, the,
the LinkedIn post you mentioned, this
n- was an app, uh, advertiser or not?
Uh, I don't remember.
Okay.
I was, I was just curious because thinking
about, I think a, a decent amount of,
um, chat, uh, uh, ChatGPT usage is still
on your desktop, and you see an app ad.
You go, "Oh, I'll just open my phone and,
and- Mm-hmm ⦠and go search it there."
So, like, you see this a
lot with UGC ads, right?
Like, you, you can't really track
things through latent clicks.
This is only a small percentage
of the attributable traffic.
Yeah.
But there's this large halo effect
of, of kind of impressions, people
going and just doing things,
uh, uh, naturally on their own.
So, but m- maybe getting more
sophisticated with this, and does this
come down to, uh, maybe impression,
uh, attribution is more valuable there?
Unclear.
Uh, we'll, we'll figure it out with time.
Um, okay.
So- Yep ⦠you, you,
you've experimented a lot.
Uh, what would be your wish list, uh,
of features if, if, hey, this, these
things like would get me pumped and
w- would get me more motivated to, to
invest more budget, and you think would
actually result in better performance?
Yes.
I have a, uh, a good wish list, um, here.
I think I, I also just looked it
up and, uh, it was not an app.
Uh, it was kind of like a-
Okay ⦠SaaS, um, platform.
Actually, I think SaaS is one of the
ones that works the best, um, on ChatGPT.
Product ads and apps, you know, is
probably where it's more difficult.
But wish list one is just give
us more insights and suggestions
on ca- uh, context hints.
Tell us some kind of v-
like, volume estimation.
Uh, it doesn't have to be, you know, exact
number, but even like a low, medium, high.
Like, no traffic, uh, 1 to 10,
something like that would really help
when you're deciding what to test,
'cause you do have to put budget
behind it all in order to, to test.
And, um, yeah, it'd just
be good to have that.
The other suggestions, you know,
in Meta, you go in and you can
like start typing something in,
and you already have suggestions.
And if you select one idea, then
suggestions on related ideas pop up.
I imagine that that would be very
easy for ChatGPT to do, and, you
know, those can be independent.
You don't have to give volume
estimations if you give suggestions
on, you know, context hints.
Um, it'd be great to have both, but those
two are definitely at the top of the list.
Um, demographic targeting would be great.
You can't do age, gender targeting
right now, um, so that would be nice.
Uh, you know, there's the whole data
privacy and everything there, but-
As an advertiser with a wishlist,
you know, that's, that's up there.
I think, uh, kinda also s-
like, rules would be nice to
have on the platform side.
Actually, like, ChatGPT and Codex, they
work really well for managing ChatGPT ads.
You, you do everything in bulk, that the
UI still leaves a lot to be desired, um,
with, like, duplication features in that.
But to have, like, rules where, for
instance, you know, I'm doing a lot
of context in testing, and I just
want to get a certain amount of
spend, you know, like with creatives.
Context hints are your
creative in a, in, in a way.
So, for every creative in Meta, I
want them to spend up to a certain
amount, and then I evaluate.
Did it drive conversion
to the right level?
In ChatGPT ads, you don't have a
setting for, like, spend evenly.
Um, so you just have to put the
budget, and you know, have a bunch of
ad sets with different context hints.
And I, I'd probably burn a lot of tokens
or usage credits if I had ChatGPT or
Codex, like, constantly monitoring and,
like, pausing context hints once they
had, once they hit a certain spend level.
So, it'd be great to have, um, rules
automations in ChatGPT ads as a
platform so that you can kind of work
with some of the limitations, um,
that are there to do your testing.
So, I think those are a few ones.
Like, again, like, turn,
uh, base targeting.
That would be interesting.
If I could say, "I only want, you know,
for this context hint, ads to show up
after 5 turns, 10 turns, 20 turns."
Um, probably you would see,
like, a correlation with
performance and quality users.
Um, at least it would be
something else to test.
Um, yeah, I think those are,
uh, a few of the, the ones.
Maybe kind of an unformed idea
just coming up now is, like, you
know, Google Ads, search ads have
dyn- dynamic keyword insertions.
Maybe there's some kind of context
hint insertion in a way that it
adjusts the ad text to kind of fit the
conversation or the context a little more.
I don't know exactly how that would
work, but this, like, pairing, you
know, the context hint with the
ad, you know, is probably something
that, that makes, um, a lot of sense.
And you can do it, you know, by having
a different ad for each context hint,
but maybe there's some kind of feature
that could help align more, um, with the
user's conversation that ChatGPT could
expose or OpenAI could expose Yeah.
I, I think that would be so cool,
and that would be, create a lot of
value for their actual customers
there if they can do that.
Um, where hopefully they think about, you
know, taking the, um, the Meta approach to
advertising, where like how do we actually
create more value for our advertisers,
versus maybe the Google approach or
Apple search ads approach, where it's
just the bare minimum, good luck.
Um- For sure ⦠we're, we're
extracting money from you.
Uh, a- and then also I was thinking,
you know, they ha- like there should be
no reason why OpenAI doesn't build out
complex automation into the platform.
Like- Mm ⦠I think they have
the technology to build some
pretty sophisticated agents to
help- Yes ⦠you do these things.
And- For sure ⦠and so that should
definitely be an integration that,
that, uh, uh, hopefully comes soon.
Uh, but yeah, so interesting
and, and I think you're right.
Like, everything's gonna evolve
so much over the, the coming
months and, and, and years.
So, um, uh, any, any last, uh, um, tips
or advice for everybody, uh, getting
started with, uh, uh, ChatGPT ads?
Um, one more feature
request is lookalikes.
You can ha- upload custom audiences,
so you can do remarketing.
But it'd be nice to, you know,
take that step forward, and then
that helps with targeting as well.
But advice, like the, uh, I don't
know, my, my kind of approach
is, is to have fun with it.
And it's like you get to explore
a new channel, you get to like try
things out, get back, you know,
put your experimentation hat on
again, and it's like, "Oh yeah, I
could try, I could try this out.
I, I could try that out."
And then it's about learning,
it's about experiencing.
And if the campaign doesn't work, you
know, you still learn, you still kind
of enjoy your craft as a marketer.
And I, I really doâ¦
It is nice to have something you
can work with, um, and it's not UAC.
I think that was one of the, the
most kind of sad things, you know,
as a marketer, that you just have
more and more being taken away.
And it makes sense, but this is, you
know, like a, a coming back to like
using your skills, using your mind, like
thinking about how to get the right ad
to the right person at the right time.
Like, it's, it's cool.
So have fun with it.
Yeah.
That's, that's good advice.
And I think the way everybody improves
is people sharing their knowledge and
tests and experiments with everybody.
So I, I, I'm super thankful for you kind
of putting this together, taking the
time to share your learning so far, even
if it's super early and, and sharing
requests and suggestions, and this is
how, uh, e- everybody gets better here.
So really appreciate you coming on and,
and sharing all your learning so far.
Absolutely.
It's great.
Thanks for making an
opportunity for me here.
Awesome.
All right.
Thanks Gabe.
I'll talk to you later.
Gotcha.
All right.
Ciao.
Thanks for listening.
Hope you enjoyed.
Please go to pricepowerpodcast.com
to see all the episodes.
Go to Spotify and YouTube and
give us a subscribe and follow
so you don't miss any episodes.
All right.
Talk to you next time.