September 30, 2025 • 36 mins
Ever wished you could clone yourself to handle all the demands of running your Amazon business? That's exactly what our team has accomplished using custom GPTs to create a "Digital Amazon CEO" that thinks like us, responds like us, and dramatically accelerates our workflow.
In this tactics-packed episode, we pull back the curtain on how we've built an AI infrastructure that's transforming our brands and client businesses. Matt shares how he went from using other people's custom GPTs to creating his own in just 30-40 minutes, resulting in specialized AI tools that slash work that once took hours down to mere minutes. You'll discover how an "AI Command Center" with departmental GPTs for marketing, sales, and specific campaigns creates a powerful ecosystem where AIs collaborate to solve problems more effectively than any single model could.
The game-changer? These custom GPTs remember everything about your brand, eliminating the need to provide extensive context with each new task. Imagine crafting comprehensive Amazon listing image briefs in five minutes or producing an entire email sequence in three minutes—all while maintaining consistent brand voice and strategy. We explore why specialized AIs hallucinate less, produce more accurate results, and why using AI aggregator services gives you the perfect tool for every job without breaking the bank.
Whether you're AI-curious or already incorporating these tools, we provide actionable steps to start leveraging this technology immediately. Begin with existing custom GPTs in the marketplace, learn to create your own, and remember you can always ask AI itself for guidance on using it more effectively. The future of Amazon brand management isn't just using AI—it's creating an AI system that thinks like you do, only faster.
🚀 Transform your brand on Amazon by building a powerful customer list with the After Purchase Funnel Blueprint course. Click here to get the full course for free.
➡️ Ready to go deeper into your Amazon FBA journey to accelerate your success? Get your hands on ALL of the Brand Fortress HQ resources, mentorship, and knowledge base by visiting us at BrandFortressHQ.com
⭐️ Want to help our show grow so we can continue bringing you the very best of guests and actionable content for your Amazon FBA business? We'd greatly appreciate if you took two minutes to give us a five star rating and review. Thank you!
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
Welcome to another episode of the Brand Fortress HQ
podcast.
On this Tactics Tuesday, we'regoing to talk about how to use
AI to create your own Amazon CEOand just other ways to actually
leverage AI in order to moveyour brand forward on Amazon.
And so with this, I'm actuallygoing to turn it over to Mike,
you and Matt, because you guysare kind of the big brains on
(00:20):
this and I've got a couple ofexperiences and thoughts towards
the end, but you guys havereally been kind of pushing the
boundaries with AI and how youleverage it to be essentially
kind of that Amazon CEO.
So love to hear your guys'experience.
Speaker 2 (00:34):
Yeah, I'd say go ahead and take the reins, Matt,
and then I'll pick up after you.
Speaker 3 (00:41):
Yeah, so for about the last I'm going to say eight
months or so I've been usingother people's custom GPTs.
So there's a GPT marketplace.
For those of you that aren'tfamiliar, I think it's only on
the paid plan so only the $20plus plan that you can search
for GPTs, Similar to likeFacebook groups.
There's a custom GPT that wastrained specifically on pretty
(01:04):
much any topic that you canthink of and, like I said, I was
using other people's for quitesome time.
There's one that I used a lotfor marketing a quick
clarification on that map.
Speaker 1 (01:14):
So you I don't know if you need to be on the paid
plan in order to search for them, but you don't need to be on
the paid plan in order to usecustom GPTs.
Got it, Okay?
Speaker 3 (01:23):
Oh, good to know.
Speaker 1 (01:23):
Okay, that's people out there that are like hey, I
ran across this custom GPT.
First of all, just be you know,know where it came from, cause
obviously somebody hasprogrammed that and they
might've done it.
Well, they might have not allkinds of different things that
could happen with that Just likea piece of software.
But in order to use thosecustom GPTs, you can.
You can use them on the freeversion of ChatGPT.
Speaker 3 (01:45):
Got it.
That's a very goodclarification and that's
actually a really good segue too, because I got a lot of use out
of the GPTs that I was usingfrom other people.
But I went through a trainingthat taught you how to create
them yourself, which, now that Iknow how to create them myself,
create them myself.
(02:05):
But I actually I use chat GPTto help me create and we'll kind
of talk about that as we go on.
But it's not as heavy of a liftas it sounds to like.
It sounds overwhelming tocreate a custom GPT and it is
like 12, 12 to 14 differenttraining files.
But, like I said, I leveragedchat GPT to help me walk through
that map and also walk throughthe creation of them.
But anyways, I went through atraining that was called AI
(02:28):
Command Center, and basicallythe idea of the AI Command
Center was using a collection ofcustom GPTs, and what I learned
and didn't know at the time isthat custom GPTs are powerful,
but bringing multiple customGPTs into the same conversation
is a thousand times morepowerful.
So that's what this AI commandcenter concept was essentially
(02:50):
is creating a basically adigital CEO.
They called it theorganizational GPT, but to me it
makes more sense to call it thedigital CEO, but essentially it
knows everything that theactual CEO.
I created one for ProTuff andit knows everything.
And actually I used Mikeactually extracted everything
that his chat GPT knew aboutProTuff, extracted it out into
(03:12):
one big giant report that I thenused top of a deep research
report that I created and kindof synthesized all of that
together to create the trainingfiles for the ProTuff digital
CEO, which knows as much as Mikedoes about ProTuff and actually
has the power of chat GPT to goout and do more stuff and learn
more stuff.
So that's kind of where itstarts is this digital CEO.
(03:34):
But then there's likedepartmental CEOs, if you kind
of.
If you continue on with thiskind of organizational structure
of them, the digital CEO knowseverything the CEO knows.
Then you have your departmentalGPTs, like your marketing and
sales and data analysis andthings like that.
So marketing was the one that Istarted out because that's what
we needed it for.
I created a chief marketing GPTthat was trained on how to
(03:58):
market specifically a physicalproduct on Amazon, tiktok,
walmart, all the differentplatforms.
So again 12 different trainingdocuments on exactly how,
specifically on Amazon andTikTok and PPC and DSP and
things like that.
So then under that you haveyour specialty GPTs.
I have, for example, a TikTokaffiliate success manager one.
(04:20):
We have a creator connectionscampaign on Amazon that we
launched and we have about 20 or30 creators that we work with
on a regular basis and our adminuses this collection of custom
GPTs to talk to each other, torespond to these creators in a
very systematized way, in a verysimilar voice, which is our
(04:41):
voice, and it's been fed so manydifferent things to be trained
on.
So essentially, it's kind ofthe idea is that you're using
this collection of custom GPTsin a way that they're trained on
not only the brand andeverything that comes with that
all of the SKUs but then alsohow to market it, but then how
to have conversations withaffiliates with it.
Speaker 1 (05:00):
So, matt, quick question for those that aren't
super familiar withno-transcript.
Speaker 3 (05:30):
the brands that I worked directly with is that
every time I wanted to start aconversation, I had to go back
and start with a whole bunch ofcontext, whether it be because I
work with five, six, eightdifferent brands at a time.
So I have to start with okay,this is the brand that I'm
talking about, this is theirproducts, this all all the stuff
that basically a CEO would know.
And I had to start everyconversation with that context
(05:53):
and then give it the context ofwhat I wanted to do and then
give it the context of all theinformation that I was giving it
to help me, help, help it, helpme.
So that's a big long process.
So, essentially, what a customGPT allows you to do is
pre-train it on a very specificthing.
So breaking that up similar toat a company, like when you get
to be a big enough company torequire hiring more than five
(06:17):
people, naturally those peoplefall into different departments
customer service or sales orsomething along those lines.
So there's that separationbecause that person specializes
in that one particular thing forthat brand.
The same thing is true withchat GPT is when you give it a
very specific task and canpre-train it on it, you remove
that whole process of having togive it background in the first
(06:39):
place and you can just gostraight to town.
So now for me creating an email.
I use three custom GPTs in thatprocess, but I can literally
pump out a three email, a copyfor three different emails in a
series, in three minutes now, asopposed to the day that it took
before.
Speaker 2 (06:57):
And it's also, I think, part of what John
might've also been asking is youknow, if you're talking custom
GPTs, why not create just onegiant custom GPT that knows
everything, instead of all ofthese separate ones that you're
bringing into the conversation?
And I think the two things thatI would say that are the most
important reasons that you mightwant to do.
(07:18):
That is one, everything that Ihad.
Speaker 1 (07:23):
Oh, I'm sure Mike was gonna make a very good point,
but just as a reminder that thispodcast is live, he just cut
out, so I'm sure he's gonna saysomething really important.
Yeah, that said, so I again,and he's probably gonna say
something oh, here it comes back, so we'll let him finish.
So you just cut out, mike.
You were saying something veryimportant to have a job.
(07:44):
It was very suspenseful, soanybody who was listening, I'm
sure I was on the edge of myseat.
Speaker 2 (07:49):
It's a needle so what you have to say about it.
I think the government tried toshut me down, I don't know.
We'll have to edit that out.
And AI, at least right now.
You know we still getsituations where you know it
hallucinates, it still givesincorrect information sometimes.
(08:10):
You know, and I think that themore broad its information base,
the more likely it is to dothat.
And so in a sense, that's whywhen you're just talking with
the normal chat GPT and you'renot giving it a whole lot of
context, you're not telling itokay, pretend you are this type
sort of person or whatever.
It tends to hallucinate more inthose situations.
(08:32):
It tends to give poor responsesmore frequently when you do
that, because it's working on amuch more generalized knowledge
base.
When you do a custom GPT, havingit specialize in something very
specific and giving itinstructions that tells it, this
is the way I want you to think.
This is the kind of person Iwant you to be like.
This is the experience that Iwant you to pretend that you
(08:55):
have.
It kind of extracts, first ofall, it takes the information
you give it in the knowledgebase, which is very specific,
but then also this is my theoryis I think it also extracts from
its own generalized knowledgebase the parts of it that are
important for what you want itto do, and it kind of sets aside
a lot of the rest of thatknowledge base, and so it's much
(09:17):
less likely to make mistakes.
So that's thing one.
Thing two, though, I think isrelated to something that we
used to do in the early dayswhen Chet GPT first came out.
This is something that I wasdoing, and even back then it was
powerful.
Now it's 10 times more powerful.
But when you have GPTs or AIstalk to themselves, but you give
(09:38):
them different personalities ordifferent experience bases.
So I used to take and insteadof having a conversation myself
with Chad GBT, I would tell ChadGBT I want to develop this new
product in such and such acategory.
So I want you to have asimulated conversation between a
materials expert, anengineering expert and a product
(10:02):
launch expert or productdevelopment expert, or something
like that.
So these three people are goingto have a conversation about
what products we might launch,and what happens in that
conversation then is the.
So you tell it, this person hassuch and such experience in,
you know, product development.
Yada, yada, yada.
You tell it, you know, you giveit a little background on each
of those three people and thenyou say go and it just continues
(10:25):
this conversation all the waydown the page between these
three very distinct individualswho would come at this problem
from a very differentperspective and with a very
different knowledge base, andwhat happens is you get much
more creative and much moreuseful responses at the end of
it.
So the same thing is true here.
(10:46):
By creating multiple veryspecialized chat GPTs that you
then integrate into theconversation that are, in a
sense, kind of talking to eachother, you get a much more
well-rounded response towhatever it is that you're
trying to do, and I just thinkin the end you're getting a much
better result.
Speaker 1 (11:03):
Yeah, yeah, and I think that's such a good point
of you know being so specificwith the different ChatGPTs and
telling it like giving itcontext, because really
otherwise it has.
I mean, it's almost like havinga conversation with an
entry-level employee.
Like if you just out of theblue say, hey, I need you to
come up with a new product forthis brand, Like their brain's
(11:25):
going to explode.
There's a million differentdirections that are going to go
in and you're going to get awhole bunch of bad ideas before
you finally get a good one.
But if you said, hey, I need aproduct that does this, is made
of these materials, does thislike now they can actually do
something with that to researchwhat does that look like?
And, even better, with chat GPT, you can actually say okay to
what you were saying.
Mike, what would an expert lookat in this situation?
(11:47):
And I think where I've foundthis the most valuable recently
is and, Matt, you've beenworking on it a little bit with
some of our clients where youknow, developing a GPT that
looks at, for example, like howdo you build out, or quickly
building out briefs for aproduct, listing image stacks,
and what I found reallyinteresting out of that was is
(12:09):
that it pulled in some ideasthat like weren't really, you
know, on our radar or otheroptions that I thought was
really useful.
But then I also noticed thatthere were other things that
I've seen as best practicesaround, like, okay, make sure
that your key benefit or uniqueselling proposition, like is in
one of the first images, so thatway people understand, like,
(12:31):
why would I buy from your brand,you know, in this product over
you know some other brands,products and kind of those
things that I think are a littlemore advanced don't necessarily
come up you know pre-programmedin the box for ChatGPT, and so
you just really get the best ofboth worlds if you have somebody
that has that expertise andthen also using ChatGPT to kind
(12:52):
of combine those two thingstogether, and I think that's
where really the magic happens.
So it's not about like, hey,ChatGPT is just going to take
over you know all of Amazon andall of marketing.
It's really okay Identifyingwhat is the 70% of what ChatGPT
that gives you.
That is really good, and thenwhat is the 30% that it's
missing because it's not generalknowledge.
Speaker 3 (13:16):
That example that you gave that custom GPT.
I think I call it Amazon ImageOptimizer GPT, I think is the
name of it.
It's a really, really goodexample of how I feel like I
have a superpower now because Iknow how to create a custom GPT.
Or in the aggregator tool thatwe use, it's called a project
and actually custom GPT hasprojects now that you can
(13:37):
actually train very similarly toa custom GPT.
So, anyways, there's a lot ofdifferent ways that you can do
that.
But, like when I, we hadprobably about three or four
different clients that weregoing through an Amazon image
refresh or they were launching anew product or a product line
that they needed new images for,so we had a need for that.
(13:58):
So, because I had that skill ofbeing able to build a custom
GPT, it took me about 40 minutesand that's optimized.
Now I've gotten that down froma couple of hours to about 30 to
40 minutes to create a customGPT.
I use chat GPT to go out andfind everything that all best
practice and aggregate all ofthat data.
I think it, like for thatspecific GPT, it used 37
(14:20):
different websites for the datathat it pulled in for the
psychology of each image in thestack on Amazon.
And it came back and I builtthese.
I think there's like 10 or 11different training docs for that
specific GPT and we created it.
It went through the process.
So, for a client, I brought intheir digital CEO.
I told her what we wanted.
(14:41):
Then we brought in the imageoptimizer, we told it all the
USPs, we gave it their website,we gave it their brand
intelligence guide that we useto train the digital CEO, and
then we had it go out and say,okay, so, based on all those
best practices, ceo.
And then we had it go out andsay, okay, so, based on all
those best practices, give uswhat you would suggest for each
stack in or each image in thestack.
And what we learned and Johnbrought this example up is that
(15:03):
some of the jobs that it told usfor a particular image weren't
what we know to be best practice.
So we had that conversation andthen we went back to that
training document and we updatedit based on what we know to be
best practice.
For the second image in thestack, specifically, the two
main usps called out in oneimage.
So we changed it.
(15:23):
We updated the knowledge baseafter we ran it through for a
couple of clients and now itknows that for the next one.
So that again, that what I toldyou.
It would take me three minutesfrom start to finish to do an
email campaign.
It also would take me aboutfive minutes now to do a
campaign or a image stack brieffor our designer.
Where before that would take memultiple hours, now I can do it
(15:44):
in five minutes.
So that's the power of usingthese collection of custom GPTs
and being able to create ones tofit a specific purpose.
Speaker 2 (15:52):
I also think that one of the things that causes
people, the most problem withtrying to use AI is that I think
oftentimes the idea that wehave on our head in terms of
what we need ChatGPT to help uswith, I think oftentimes we're
in the middle of theconversation and we need to back
(16:15):
up and kind of re-engineerwhere we should have started in
the conversation.
What I mean by that is we likewhen we have an idea about
something you know we'll go inand we just kind of type in a
prompt and we're looking forChatGPT to make magic Right.
But the thing that that makesGPT ork or I mean like we keep
(16:38):
talking about chat GPT buthonestly, claude, is
exceptionally good.
You know, sonnet 4, opus 4.1are amazing.
So it's not like you have touse chat GPT, but any of these
AI models.
The thing that makes them mostamazing in my mind is oftentimes
the things that it tells methat I should be thinking about,
that I'm not thinking about,and so what I'll often do is my
(17:01):
first prompt really is basicallyexpressing to ChatGPT.
This is where I am in theprocess what did I forget?
Where should I go back to?
What questions should I beasking that?
Maybe I haven't already asked?
What information should I betrying to find that maybe I
haven't already looked for, youknow, like letting it ask me
questions about you know, thisprocess before I delve into the
(17:25):
actual conversation and tryingto move forward, I try to move
backward first.
You know, it is shocking to methe number of times that I enter
into a conversation with GPTand recognize I am I'm not even
in the ballpark of where Iactually should be in that
conversation and the directionthat I should be moving, because
(17:46):
there was these other piecesthat didn't even occur to me
that I should be thinking aboutand I'm like, oh wow, you know
like.
OK, let's start there.
Then you know, let's, let'smove from there.
So I think that's thing one.
Thing two is don't look at itas like a genie in the lamp.
You know where you rub the lampand you get your magic answer
the first try, or even in thefirst three tries.
Let's say it's you need.
(18:07):
You need to keep going withthat conversation and keep
iterating on that conversationuntil you get where you want to
go, and sometimes that takessome time, and so I think that's
a big value piece there.
One of the things that I wouldpoint out with that is that
that's one of the value piecesof that custom GPT is that a lot
(18:29):
of times, that process of goingthrough that prompt is giving
it all that background.
So having that custom GPT tostart with that already has all
that background, saves a massiveamount of time, and so it's
really a good idea.
Other platforms have similarfunctions to a custom GPT.
So if you're not using, youknow, chat GPT, figure out what
that function is and the otherplatform that you're using Do
(18:50):
they.
Speaker 1 (18:50):
Because I haven't seen this with.
I've messed around a little bitwith Gemini and that type of
stuff and I haven't seen it yet.
Speaker 2 (18:57):
Gemini had something.
I was in there the other dayand I could swear I don't
remember what gems.
Speaker 1 (19:01):
Are they called gems?
I think, yeah, I think they'recalled gems.
Speaker 2 (19:07):
Now I didn't play around with it that much because
I don't work in there that much.
Most of my time and this isprobably a good segue, most of
my time these days is spentusing an aggregation service
(19:30):
Matt has used in the past.
I kind of latched on to Abacus,which I like quite well.
I've never used Magi, maybeit's even better, but Abacus is
what we use as a company and toclarify, you know, an aggregator
service essentially what itdoes is it provides you with
access to all of the differentAI models at any given moment.
So anytime I start a chat, Ican do one of two things I can
(19:57):
either allow Abacus to decidewhich chat model to utilize in
response to my prompt, andbasically what it's doing is
it's using past data toestablish what we have
determined over time.
Is this type of question isbest answered by this model.
So maybe it chooses Chet, gpt5.0, or it chooses, you know,
(20:17):
quad, anthropic, you know maybeit uses Opus 4.1 or whatever,
and you get charged a certainnumber of credits based on you
know the amount of processingand the type of model that you
use, because some models aremore expensive than others, at
least by API, but you haveaccess to everything, and that's
not just, you know, grok andClaude and ChatGPT and Gemini,
(20:38):
it's also Midjourney and Fluxand all of these different types
of AI models that are eitherimage generators or video
generators.
Or, you know, all of thesevarious different models that
you could use deep seek and allof these more exotic things that
a lot of people don't touch.
All of them are accessible, andso you can either let Abacus
(20:58):
choose or you can specificallyselect which model you would
like to use for this particularprompt.
For me, whenever it's a promptthat I'm like I really need some
serious thinking power in this.
There are certain models that Igo to that I feel are better
with logic and reasoning.
I'll often use Clause Sonnet 4.
(21:19):
If I really need something good, quite frankly, I'll often use
Clause Opus 4.1, which uses alot of credits.
But I get to the answer.
I was actually having aconversation with one of my
employees earlier today, so oneof the.
But I get to the answer.
I was actually having aconversation with one of my
employees earlier today, so oneof the things that I think
oftentimes when people are usingthese aggregators because it'll
show you how many credits youused and you purchase a certain
(21:39):
number of credits.
You know, based on your plan,you get a certain number of
credits every month, and so alot of times what people will do
is they're like oh, like, Iknow that GPT 4.1 would use like
five credits for this prompt,and if I use Opus it's going to
be like 20.
So I'm going to use 4.1.
But the problem is you'reasking it a question that,
reasonably speaking, is a littlebit over its head, and so what
(22:02):
happens is it takes four or fiveor six prompts to actually get
to where you need to be in thatconversation, to where you get a
real answer.
Well, now you've already usedthose 20 credits, but you've
also now wasted the time ofgoing through multiple prompts
to get there.
So you're better off sometimesusing that more advanced model
(22:24):
that might use four or five, sixtimes the credits, but you're
going to get the right answerright out of the gate based on
that first prompt that you putin, rather than spending another
10 minutes, you know,reiterating the prompt and
changing this and going anddoing this.
So it's one of the reasons Ireally like aggregators, because
I'm not always convinced thatChatGPT is the best model for
(22:48):
what it is that I'm trying to do, and so it allows me to choose
the model that I want and forimage generation, like, if you
use ChatGPT for image generation, it's not bad, it can do some
pretty decent images.
But if you get more specific,you know, and I want to use
Midjourney or I want to use Fluxor I want to use whatever I can
do all of that right inside ofAbacus.
I don't have to go out and goto MidJourney, I don't have to
(23:11):
go out and go to Flux orwhatever.
It's all in that same systemand it's actually cheaper
because I don't have to pay forMidJourney and ChatGPT and Grok
and Claude and all of these.
I don't have to have a serviceon any of them.
I pay for Abacus and Abacususes their API to connect into
(23:32):
all of these other services andI don't have to pay for any of
them.
So if you want access to a lotof those other models,
especially at varied types ofmodels you know, for video and
images and things like that,using an aggregator could be a
way to do that in a way thatdoesn't break the bank, you know
, for your business but givesyou access to a lot of the tools
that you're looking for.
Speaker 3 (23:49):
I mean, it saved me money and it gave me access to
more things.
So I'm curious.
Speaker 1 (23:55):
You know kind of taking a little bit of a
sidestep here.
One of the other things,because we talked a lot about
custom GPTs but there are othertools now out there that
essentially act like custom GPTs, but across multiple platforms.
That I know.
That's something that we'rewell I'm starting to look into
with my agency and looking atleveraging that.
(24:17):
So that way, custom GPTs aregreat.
But again back to what you weretalking about, mike of like
okay, well, if we can get abetter result out of you know
Gemini versus ChatGPT, like wejust care what you know model
gives us the best result.
I'm not married to you knowChatGPT versus Grok or you know
whatever.
Speaker 2 (24:37):
Well, an abacus is actually nice in that regard and
again, matt can speak to Magiand how it functions.
But so within abacus, youcreate essentially a custom GPT,
and abacus is called a project.
Now, they're not exactly thesame thing, but they're similar.
So when you start a project inAbacus, you give it a knowledge
base and you give it a set ofcustom instructions, which is
(24:58):
pretty much what a custom GPT is.
When you do a custom GPT, yougive it a knowledge base and
then you give it its custominstructions, and then you go
from there, right, you're off tothe races.
So in that sense, they'reidentical.
What's nice, though, aboutAbacus projects is is that then
you have access to exactly whatyou're suggesting, john, which
(25:19):
is one.
You can open up a new projectchat, and now it automatically
has access to all of thatknowledge base and the custom
instructions that you gave it,which is kind of who are you,
how do you think?
What kind of you know like allthat kind of thing that you gave
it, which is kind of who areyou, how do you think what kind
of you know like all that kindof thing?
But then when you start theconversation again, same as any
other chat in Abacus.
You can either let Abacus choosewhich model it should use to
(25:39):
answer the questions and, by theway, I should clarify that
model can change at any point inthe conversation.
So I can start with like like Istart the conversation and
let's say I just let Abacuschoose.
I ask it a question and itdecides Gemini 2.5 is the right
model, so it answers thequestion.
Then I follow that with another.
(25:59):
You know, question or reply.
Abacus might decide okay, well,now in this point of the
conversation, claude Sonnet 4 isbetter, or ChatGPT this is
better, or you know whatever,and it will switch models at
will.
You can then also clickregenerate.
If it gives you an answer andyou don't really like it and you
think there's a different modelthat might work better, it'll
(26:19):
give you suggestions of othermodels that you could have tried
.
You can regenerate that answerwith a different model, but also
same thing you can choose.
So at any point in thatconversation I can change it and
say use QuadSonic 4.
Or you know, now I'm at thispoint in the conversation I'm
like oh, we need to use Fluxright now because I want to do
an image, or you know whatever.
So you're in that project.
It still has all the context,all of that information, just
(26:43):
like a custom GPT, but now youcan use any model you want
within that chat.
Speaker 1 (26:49):
Yeah, and I think there's a couple of just
important takeaways forlisteners out of that what you
just said, mike.
So the first one is is that itmeans you have the right tool
for the job you know in thesense of, or the best tool for
the job you know.
It's like you know, if you'llreach into your toolbox and
you're like, okay, I need to,you know, pound in a nail, okay,
(27:12):
and you're like, okay, I needto, you know, pound in a nail
Okay, well, if the only thingyou have is a wrench, like, yeah
, you can make it work, but youknow, a hammer would be a lot
nicer.
And so that's kind of, as we'restarting to see these different
tools, you know, kind of mature, they're good, good at
different things, and havingsomething like that aggregator,
aggregator, abacus or whatevergives you access to all the
tools in the toolbox.
I think the other importantthing, what we've found and
again we primarily use ChatGPT,but just you know from what you
(27:35):
bring up which is, it allows youto keep the results consistent.
So you know, okay, which modelgave us the best results?
And so then if you're havingyour team use that consistently,
you know that same project overand over again.
You know like, hey, here arethe parameters that give us the
best outcome, given this project, that we may do multiple times
(27:57):
a day, you know we may do, youknow, use tons and tons of time,
or do many, many times in orderto get a really consistent
result.
That actually does have a verypositive impact on the business
by allowing you to speed upprocesses and deliver a lot
faster than your competitors,which is super important,
(28:19):
especially on Amazon.
Speaker 2 (28:20):
Well, and if you're a team, I mean also just the cost
.
As you said, different toolsare good at different things.
You could easily be subscribedto 20 different AI services.
You know that do the variousthings within your business that
you need done and you know thistool is better for this, or
this tool is better for this, orthis tool is better for this.
(28:41):
Or you might be subscribed tofive of them and you're like,
look, this one is not the bestat this thing, but it's good
enough at this thing and I don'twant to pay for the one that's
the best for it, because then Igot another subscription that
I'm paying.
So we're going to stick tothese three or four or five
tools, whereas with theaggregator, it gives you access
to all of the tools under onesubscription price.
(29:03):
And if you're a team like ifyou got ChatGPT and you're a
team like ours, where you got 10people you know well.
I got to pay for a seat forevery single one of those 10
people, good Lord.
The price adds up, you know,and if I need multiple tools,
you know multiple AIs, that's anastronomical amount of money
(29:23):
and most of the time, some ofthose tools you're paying for
you only use occasionally, sohaving access to them through an
aggregator is way lessexpensive.
Speaker 1 (29:30):
Yeah, I think we've covered some good ground with
this conversation.
I know that there's a lot morethat we could talk about.
I do think this is probably agood place to wrap for today, so
, with that in mind, maybe I'llstart with you.
Matt, one kind of action itemyou'd have for brand owners who
are listening right now and arelooking for more ways to
(29:51):
leverage AI in their businesswhat advice would you have or
action item would you have forthem?
Speaker 3 (29:57):
So I kind of use ChatGPT as my security blanket.
So most of the two tips thatI'm going to give right now are
very ChatGPT focused.
First of all, if you've neverused a ChatGPT, that's where I
would start is go to the ChatGPTor the custom GPT marketplace.
First of all, if you've neverused a chat GPT, that's where I
would start is go to the chatGPT or the custom GPT
marketplace and find one.
For example, there's Alex Ramos.
He wrote a book called ahundred million dollar offers.
(30:18):
So go find that a hundredmillion dollar offer GPT and
have it.
Help create the grand slamoffer for your business.
That will take you five minutesto find that GPT and use it.
If you have used a custom GPTbefore and you've gotten value
out of it, learn how to createyour own.
It's not as overwhelming as youthink it is.
I've gotten the process downnow to about 20 to 30 minutes
(30:40):
for a decent custom GPT that Ican use over and over and over
and over again.
So that's what I wouldchallenge you is start using
them if you haven't, and startcreating them if you already
understand the power of them.
Speaker 2 (30:53):
And I guess for me, honestly, just kind of
piggybacking on that.
I think you know that idea ofcreating your own custom GPT
sounds, you know, maybe dauntingto some people.
To others it's old hat, youknow they've been doing it for
months.
But if it, if it seems daunting, I think, remembering that you
can use AI to help you use AIOkay, it's a very meta kind of
(31:17):
concept, right.
Like it's hard for people towrap their brains around that.
Like I was literally havingdinner with somebody the other
day some good friends of ours,and he uses AI a little bit.
Never occurred to him that hecould use AI to help him
understand how to better use AIjust didn't occur, and so we
(31:38):
were talking about that.
It was, like you know, mindblown.
So if there's something thatyou think you might want to use
AI for but you don't actuallyknow how to use AI for it, ask
AI how to use AI for that.
It will tell you and you'll beamazed at just how good that
response will be and it can walkyou through that process in a
very systematic kind ofmethodical way.
(31:59):
The other thing this is justkind of a quick tip is that it's
not uncommon that when you aska question of AI, it will give
you, like this giant especiallyif it's something that's fairly
involved Right, and it's like,if you're asking it, how do I do
X, y, z?
Right, and it's kind of a stepby step.
It'll give you the whole thingall at once and you're like, oh
(32:21):
geez, and then if there's somekind of maybe not mistake, but
there's some problem in therethat you can't work through,
then you're backtracking overand over again, right?
So in a situation like that,where I'm asking an AI to give
me a step-by-step, I am veryspecific to tell it only give me
the next step in the process.
As soon as we successfullycomplete that step, then give me
the next step and don't give meany more steps until we
(32:42):
successfully complete the step.
We're on it.
Really, it makes your chatsbetter.
First of all because a chat youcan only have so many tokens,
basically so many characters,let's say that go on inside of a
chat.
So the longer it goes, the morelikely an AI is to hallucinate.
So if it gives you this longanswer and all you really needed
was this little piece, you justwasted all of that, right?
(33:05):
So don't have it.
Do that, have it, give you thelittle piece, little piece,
little piece, and you'll be ableto get to the end of that
conversation without running outof tokens and having it start
hallucinating on you.
Speaker 1 (33:13):
Yeah, and then what I would throw out there just as
an action item for viewers isthink about something that you
do on a regular basis.
So the example that I think ofthat has just been incredibly
useful for me is that I'm the Inaddition to being the founder
and CEO for the Amazon agency Irun, I'm also the chief revenue
(33:35):
officer, which means I'mresponsible for all the sales
calls and follow-ups with peopleand that type of stuff.
I'm a decent email copywriter,but it's been so helpful to say,
okay, I actually have a littleGPT.
It's not super complicated thatI just take the email that I
write and put it into the GPTand say, hey, make this email
more engaging and persuasive.
(33:56):
And put it into the GPT and say, hey, make this email more
engaging and persuasive.
And really what I find out ofthat is that it takes the
pressure off from me of havingto write the perfect email,
because I can come up with, okay, here's kind of my ideas and
put out kind of what I want tosay.
I put it in chat GPT and I'mlike, okay, I really like this
that it came up with this otherpiece, and the seven m dashes
(34:18):
that put in there, you knowaren't a good fit, but I got
enough good ideas from itbecause there's been enough
times where I've looked atsomething to be like because I
customize everyone to make itengaging in that type of stuff
of like hey, you know, you knowwhy is what?
Does that next step look likeand being excited about you know
working together, so it justdoes a really step look like and
being excited about you knowworking together.
So it just does a really goodjob of almost being like that
(34:41):
writing partner, and so itdoesn't have to be something
where it's, you know, building anew product or, you know,
massive project in order to beuseful.
Sometimes, just you know thingsthat you do day to day in order
to give you almost like anotherperson to bounce ideas off of
and to, you know, really justmake things go faster.
So, yeah, I would encouragepeople to just look at you know
(35:02):
what are the day-to-day thingsyou do and can you know how can
you test, leverage it or usingChatGPT in order to solve those.
So, with that, I think this isa great place to wrap on AI.
Obviously, you know this willcontinue to change and we'll
have another conversation in sixmonths and that you know, the
entire AI landscape will havechanged and we'll have more fun
(35:23):
things to talk about then.
So until then, thank youeverybody for listening.
Welcome to another episode of the Brand Fortress HQ
podcast.
On this Tactics Tuesday, we'regoing to talk about how to use
AI to create your own Amazon CEOand just other ways to actually
leverage AI in order to moveyour brand forward on Amazon.
And so with this, I'm actuallygoing to turn it over to Mike,
you and Matt, because you guysare kind of the big brains on
(00:20):
this and I've got a couple ofexperiences and thoughts towards
the end, but you guys havereally been kind of pushing the
boundaries with AI and how youleverage it to be essentially
kind of that Amazon CEO.
So love to hear your guys'experience.
Speaker 2 (00:34):
Yeah, I'd say go ahead and take the reins, Matt,
and then I'll pick up after you.
Speaker 3 (00:41):
Yeah, so for about the last I'm going to say eight
months or so I've been usingother people's custom GPTs.
So there's a GPT marketplace.
For those of you that aren'tfamiliar, I think it's only on
the paid plan so only the $20plus plan that you can search
for GPTs, Similar to likeFacebook groups.
There's a custom GPT that wastrained specifically on pretty
(01:04):
much any topic that you canthink of and, like I said, I was
using other people's for quitesome time.
There's one that I used a lotfor marketing a quick
clarification on that map.
Speaker 1 (01:14):
So you I don't know if you need to be on the paid
plan in order to search for them, but you don't need to be on
the paid plan in order to usecustom GPTs.
Got it, Okay?
Speaker 3 (01:23):
Oh, good to know.
Speaker 1 (01:23):
Okay, that's people out there that are like hey, I
ran across this custom GPT.
First of all, just be you know,know where it came from, cause
obviously somebody hasprogrammed that and they
might've done it.
Well, they might have not allkinds of different things that
could happen with that Just likea piece of software.
But in order to use thosecustom GPTs, you can.
You can use them on the freeversion of ChatGPT.
Speaker 3 (01:45):
Got it.
That's a very goodclarification and that's
actually a really good segue too, because I got a lot of use out
of the GPTs that I was usingfrom other people.
But I went through a trainingthat taught you how to create
them yourself, which, now that Iknow how to create them myself,
create them myself.
(02:05):
But I actually I use chat GPTto help me create and we'll kind
of talk about that as we go on.
But it's not as heavy of a liftas it sounds to like.
It sounds overwhelming tocreate a custom GPT and it is
like 12, 12 to 14 differenttraining files.
But, like I said, I leveragedchat GPT to help me walk through
that map and also walk throughthe creation of them.
But anyways, I went through atraining that was called AI
(02:28):
Command Center, and basicallythe idea of the AI Command
Center was using a collection ofcustom GPTs, and what I learned
and didn't know at the time isthat custom GPTs are powerful,
but bringing multiple customGPTs into the same conversation
is a thousand times morepowerful.
So that's what this AI commandcenter concept was essentially
(02:50):
is creating a basically adigital CEO.
They called it theorganizational GPT, but to me it
makes more sense to call it thedigital CEO, but essentially it
knows everything that theactual CEO.
I created one for ProTuff andit knows everything.
And actually I used Mikeactually extracted everything
that his chat GPT knew aboutProTuff, extracted it out into
(03:12):
one big giant report that I thenused top of a deep research
report that I created and kindof synthesized all of that
together to create the trainingfiles for the ProTuff digital
CEO, which knows as much as Mikedoes about ProTuff and actually
has the power of chat GPT to goout and do more stuff and learn
more stuff.
So that's kind of where itstarts is this digital CEO.
(03:34):
But then there's likedepartmental CEOs, if you kind
of.
If you continue on with thiskind of organizational structure
of them, the digital CEO knowseverything the CEO knows.
Then you have your departmentalGPTs, like your marketing and
sales and data analysis andthings like that.
So marketing was the one that Istarted out because that's what
we needed it for.
I created a chief marketing GPTthat was trained on how to
(03:58):
market specifically a physicalproduct on Amazon, tiktok,
walmart, all the differentplatforms.
So again 12 different trainingdocuments on exactly how,
specifically on Amazon andTikTok and PPC and DSP and
things like that.
So then under that you haveyour specialty GPTs.
I have, for example, a TikTokaffiliate success manager one.
(04:20):
We have a creator connectionscampaign on Amazon that we
launched and we have about 20 or30 creators that we work with
on a regular basis and our adminuses this collection of custom
GPTs to talk to each other, torespond to these creators in a
very systematized way, in a verysimilar voice, which is our
(04:41):
voice, and it's been fed so manydifferent things to be trained
on.
So essentially, it's kind ofthe idea is that you're using
this collection of custom GPTsin a way that they're trained on
not only the brand andeverything that comes with that
all of the SKUs but then alsohow to market it, but then how
to have conversations withaffiliates with it.
Speaker 1 (05:00):
So, matt, quick question for those that aren't
super familiar withno-transcript.
Speaker 3 (05:30):
the brands that I worked directly with is that
every time I wanted to start aconversation, I had to go back
and start with a whole bunch ofcontext, whether it be because I
work with five, six, eightdifferent brands at a time.
So I have to start with okay,this is the brand that I'm
talking about, this is theirproducts, this all all the stuff
that basically a CEO would know.
And I had to start everyconversation with that context
(05:53):
and then give it the context ofwhat I wanted to do and then
give it the context of all theinformation that I was giving it
to help me, help, help it, helpme.
So that's a big long process.
So, essentially, what a customGPT allows you to do is
pre-train it on a very specificthing.
So breaking that up similar toat a company, like when you get
to be a big enough company torequire hiring more than five
(06:17):
people, naturally those peoplefall into different departments
customer service or sales orsomething along those lines.
So there's that separationbecause that person specializes
in that one particular thing forthat brand.
The same thing is true withchat GPT is when you give it a
very specific task and canpre-train it on it, you remove
that whole process of having togive it background in the first
(06:39):
place and you can just gostraight to town.
So now for me creating an email.
I use three custom GPTs in thatprocess, but I can literally
pump out a three email, a copyfor three different emails in a
series, in three minutes now, asopposed to the day that it took
before.
Speaker 2 (06:57):
And it's also, I think, part of what John
might've also been asking is youknow, if you're talking custom
GPTs, why not create just onegiant custom GPT that knows
everything, instead of all ofthese separate ones that you're
bringing into the conversation?
And I think the two things thatI would say that are the most
important reasons that you mightwant to do.
(07:18):
That is one, everything that Ihad.
Speaker 1 (07:23):
Oh, I'm sure Mike was gonna make a very good point,
but just as a reminder that thispodcast is live, he just cut
out, so I'm sure he's gonna saysomething really important.
Yeah, that said, so I again,and he's probably gonna say
something oh, here it comes back, so we'll let him finish.
So you just cut out, mike.
You were saying something veryimportant to have a job.
(07:44):
It was very suspenseful, soanybody who was listening, I'm
sure I was on the edge of myseat.
Speaker 2 (07:49):
It's a needle so what you have to say about it.
I think the government tried toshut me down, I don't know.
We'll have to edit that out.
And AI, at least right now.
You know we still getsituations where you know it
hallucinates, it still givesincorrect information sometimes.
(08:10):
You know, and I think that themore broad its information base,
the more likely it is to dothat.
And so in a sense, that's whywhen you're just talking with
the normal chat GPT and you'renot giving it a whole lot of
context, you're not telling itokay, pretend you are this type
sort of person or whatever.
It tends to hallucinate more inthose situations.
(08:32):
It tends to give poor responsesmore frequently when you do
that, because it's working on amuch more generalized knowledge
base.
When you do a custom GPT, havingit specialize in something very
specific and giving itinstructions that tells it, this
is the way I want you to think.
This is the kind of person Iwant you to be like.
This is the experience that Iwant you to pretend that you
(08:55):
have.
It kind of extracts, first ofall, it takes the information
you give it in the knowledgebase, which is very specific,
but then also this is my theoryis I think it also extracts from
its own generalized knowledgebase the parts of it that are
important for what you want itto do, and it kind of sets aside
a lot of the rest of thatknowledge base, and so it's much
(09:17):
less likely to make mistakes.
So that's thing one.
Thing two, though, I think isrelated to something that we
used to do in the early dayswhen Chet GPT first came out.
This is something that I wasdoing, and even back then it was
powerful.
Now it's 10 times more powerful.
But when you have GPTs or AIstalk to themselves, but you give
(09:38):
them different personalities ordifferent experience bases.
So I used to take and insteadof having a conversation myself
with Chad GBT, I would tell ChadGBT I want to develop this new
product in such and such acategory.
So I want you to have asimulated conversation between a
materials expert, anengineering expert and a product
(10:02):
launch expert or productdevelopment expert, or something
like that.
So these three people are goingto have a conversation about
what products we might launch,and what happens in that
conversation then is the.
So you tell it, this person hassuch and such experience in,
you know, product development.
Yada, yada, yada.
You tell it, you know, you giveit a little background on each
of those three people and thenyou say go and it just continues
(10:25):
this conversation all the waydown the page between these
three very distinct individualswho would come at this problem
from a very differentperspective and with a very
different knowledge base, andwhat happens is you get much
more creative and much moreuseful responses at the end of
it.
So the same thing is true here.
(10:46):
By creating multiple veryspecialized chat GPTs that you
then integrate into theconversation that are, in a
sense, kind of talking to eachother, you get a much more
well-rounded response towhatever it is that you're
trying to do, and I just thinkin the end you're getting a much
better result.
Speaker 1 (11:03):
Yeah, yeah, and I think that's such a good point
of you know being so specificwith the different ChatGPTs and
telling it like giving itcontext, because really
otherwise it has.
I mean, it's almost like havinga conversation with an
entry-level employee.
Like if you just out of theblue say, hey, I need you to
come up with a new product forthis brand, Like their brain's
(11:25):
going to explode.
There's a million differentdirections that are going to go
in and you're going to get awhole bunch of bad ideas before
you finally get a good one.
But if you said, hey, I need aproduct that does this, is made
of these materials, does thislike now they can actually do
something with that to researchwhat does that look like?
And, even better, with chat GPT, you can actually say okay to
what you were saying.
Mike, what would an expert lookat in this situation?
(11:47):
And I think where I've foundthis the most valuable recently
is and, Matt, you've beenworking on it a little bit with
some of our clients where youknow, developing a GPT that
looks at, for example, like howdo you build out, or quickly
building out briefs for aproduct, listing image stacks,
and what I found reallyinteresting out of that was is
(12:09):
that it pulled in some ideasthat like weren't really, you
know, on our radar or otheroptions that I thought was
really useful.
But then I also noticed thatthere were other things that
I've seen as best practicesaround, like, okay, make sure
that your key benefit or uniqueselling proposition, like is in
one of the first images, so thatway people understand, like,
(12:31):
why would I buy from your brand,you know, in this product over
you know some other brands,products and kind of those
things that I think are a littlemore advanced don't necessarily
come up you know pre-programmedin the box for ChatGPT, and so
you just really get the best ofboth worlds if you have somebody
that has that expertise andthen also using ChatGPT to kind
(12:52):
of combine those two thingstogether, and I think that's
where really the magic happens.
So it's not about like, hey,ChatGPT is just going to take
over you know all of Amazon andall of marketing.
It's really okay Identifyingwhat is the 70% of what ChatGPT
that gives you.
That is really good, and thenwhat is the 30% that it's
missing because it's not generalknowledge.
Speaker 3 (13:16):
That example that you gave that custom GPT.
I think I call it Amazon ImageOptimizer GPT, I think is the
name of it.
It's a really, really goodexample of how I feel like I
have a superpower now because Iknow how to create a custom GPT.
Or in the aggregator tool thatwe use, it's called a project
and actually custom GPT hasprojects now that you can
(13:37):
actually train very similarly toa custom GPT.
So, anyways, there's a lot ofdifferent ways that you can do
that.
But, like when I, we hadprobably about three or four
different clients that weregoing through an Amazon image
refresh or they were launching anew product or a product line
that they needed new images for,so we had a need for that.
(13:58):
So, because I had that skill ofbeing able to build a custom
GPT, it took me about 40 minutesand that's optimized.
Now I've gotten that down froma couple of hours to about 30 to
40 minutes to create a customGPT.
I use chat GPT to go out andfind everything that all best
practice and aggregate all ofthat data.
I think it, like for thatspecific GPT, it used 37
(14:20):
different websites for the datathat it pulled in for the
psychology of each image in thestack on Amazon.
And it came back and I builtthese.
I think there's like 10 or 11different training docs for that
specific GPT and we created it.
It went through the process.
So, for a client, I brought intheir digital CEO.
I told her what we wanted.
(14:41):
Then we brought in the imageoptimizer, we told it all the
USPs, we gave it their website,we gave it their brand
intelligence guide that we useto train the digital CEO, and
then we had it go out and say,okay, so, based on all those
best practices, ceo.
And then we had it go out andsay, okay, so, based on all
those best practices, give uswhat you would suggest for each
stack in or each image in thestack.
And what we learned and Johnbrought this example up is that
(15:03):
some of the jobs that it told usfor a particular image weren't
what we know to be best practice.
So we had that conversation andthen we went back to that
training document and we updatedit based on what we know to be
best practice.
For the second image in thestack, specifically, the two
main usps called out in oneimage.
So we changed it.
(15:23):
We updated the knowledge baseafter we ran it through for a
couple of clients and now itknows that for the next one.
So that again, that what I toldyou.
It would take me three minutesfrom start to finish to do an
email campaign.
It also would take me aboutfive minutes now to do a
campaign or a image stack brieffor our designer.
Where before that would take memultiple hours, now I can do it
(15:44):
in five minutes.
So that's the power of usingthese collection of custom GPTs
and being able to create ones tofit a specific purpose.
Speaker 2 (15:52):
I also think that one of the things that causes
people, the most problem withtrying to use AI is that I think
oftentimes the idea that wehave on our head in terms of
what we need ChatGPT to help uswith, I think oftentimes we're
in the middle of theconversation and we need to back
(16:15):
up and kind of re-engineerwhere we should have started in
the conversation.
What I mean by that is we likewhen we have an idea about
something you know we'll go inand we just kind of type in a
prompt and we're looking forChatGPT to make magic Right.
But the thing that that makesGPT ork or I mean like we keep
(16:38):
talking about chat GPT buthonestly, claude, is
exceptionally good.
You know, sonnet 4, opus 4.1are amazing.
So it's not like you have touse chat GPT, but any of these
AI models.
The thing that makes them mostamazing in my mind is oftentimes
the things that it tells methat I should be thinking about,
that I'm not thinking about,and so what I'll often do is my
(17:01):
first prompt really is basicallyexpressing to ChatGPT.
This is where I am in theprocess what did I forget?
Where should I go back to?
What questions should I beasking that?
Maybe I haven't already asked?
What information should I betrying to find that maybe I
haven't already looked for, youknow, like letting it ask me
questions about you know, thisprocess before I delve into the
(17:25):
actual conversation and tryingto move forward, I try to move
backward first.
You know, it is shocking to methe number of times that I enter
into a conversation with GPTand recognize I am I'm not even
in the ballpark of where Iactually should be in that
conversation and the directionthat I should be moving, because
(17:46):
there was these other piecesthat didn't even occur to me
that I should be thinking aboutand I'm like, oh wow, you know
like.
OK, let's start there.
Then you know, let's, let'smove from there.
So I think that's thing one.
Thing two is don't look at itas like a genie in the lamp.
You know where you rub the lampand you get your magic answer
the first try, or even in thefirst three tries.
Let's say it's you need.
(18:07):
You need to keep going withthat conversation and keep
iterating on that conversationuntil you get where you want to
go, and sometimes that takessome time, and so I think that's
a big value piece there.
One of the things that I wouldpoint out with that is that
that's one of the value piecesof that custom GPT is that a lot
(18:29):
of times, that process of goingthrough that prompt is giving
it all that background.
So having that custom GPT tostart with that already has all
that background, saves a massiveamount of time, and so it's
really a good idea.
Other platforms have similarfunctions to a custom GPT.
So if you're not using, youknow, chat GPT, figure out what
that function is and the otherplatform that you're using Do
(18:50):
they.
Speaker 1 (18:50):
Because I haven't seen this with.
I've messed around a little bitwith Gemini and that type of
stuff and I haven't seen it yet.
Speaker 2 (18:57):
Gemini had something.
I was in there the other dayand I could swear I don't
remember what gems.
Speaker 1 (19:01):
Are they called gems?
I think, yeah, I think they'recalled gems.
Speaker 2 (19:07):
Now I didn't play around with it that much because
I don't work in there that much.
Most of my time and this isprobably a good segue, most of
my time these days is spentusing an aggregation service
(19:30):
Matt has used in the past.
I kind of latched on to Abacus,which I like quite well.
I've never used Magi, maybeit's even better, but Abacus is
what we use as a company and toclarify, you know, an aggregator
service essentially what itdoes is it provides you with
access to all of the differentAI models at any given moment.
So anytime I start a chat, Ican do one of two things I can
(19:57):
either allow Abacus to decidewhich chat model to utilize in
response to my prompt, andbasically what it's doing is
it's using past data toestablish what we have
determined over time.
Is this type of question isbest answered by this model.
So maybe it chooses Chet, gpt5.0, or it chooses, you know,
(20:17):
quad, anthropic, you know maybeit uses Opus 4.1 or whatever,
and you get charged a certainnumber of credits based on you
know the amount of processingand the type of model that you
use, because some models aremore expensive than others, at
least by API, but you haveaccess to everything, and that's
not just, you know, grok andClaude and ChatGPT and Gemini,
(20:38):
it's also Midjourney and Fluxand all of these different types
of AI models that are eitherimage generators or video
generators.
Or, you know, all of thesevarious different models that
you could use deep seek and allof these more exotic things that
a lot of people don't touch.
All of them are accessible, andso you can either let Abacus
(20:58):
choose or you can specificallyselect which model you would
like to use for this particularprompt.
For me, whenever it's a promptthat I'm like I really need some
serious thinking power in this.
There are certain models that Igo to that I feel are better
with logic and reasoning.
I'll often use Clause Sonnet 4.
(21:19):
If I really need something good, quite frankly, I'll often use
Clause Opus 4.1, which uses alot of credits.
But I get to the answer.
I was actually having aconversation with one of my
employees earlier today, so oneof the.
But I get to the answer.
I was actually having aconversation with one of my
employees earlier today, so oneof the things that I think
oftentimes when people are usingthese aggregators because it'll
show you how many credits youused and you purchase a certain
(21:39):
number of credits.
You know, based on your plan,you get a certain number of
credits every month, and so alot of times what people will do
is they're like oh, like, Iknow that GPT 4.1 would use like
five credits for this prompt,and if I use Opus it's going to
be like 20.
So I'm going to use 4.1.
But the problem is you'reasking it a question that,
reasonably speaking, is a littlebit over its head, and so what
(22:02):
happens is it takes four or fiveor six prompts to actually get
to where you need to be in thatconversation, to where you get a
real answer.
Well, now you've already usedthose 20 credits, but you've
also now wasted the time ofgoing through multiple prompts
to get there.
So you're better off sometimesusing that more advanced model
(22:24):
that might use four or five, sixtimes the credits, but you're
going to get the right answerright out of the gate based on
that first prompt that you putin, rather than spending another
10 minutes, you know,reiterating the prompt and
changing this and going anddoing this.
So it's one of the reasons Ireally like aggregators, because
I'm not always convinced thatChatGPT is the best model for
(22:48):
what it is that I'm trying to do, and so it allows me to choose
the model that I want and forimage generation, like, if you
use ChatGPT for image generation, it's not bad, it can do some
pretty decent images.
But if you get more specific,you know, and I want to use
Midjourney or I want to use Fluxor I want to use whatever I can
do all of that right inside ofAbacus.
I don't have to go out and goto MidJourney, I don't have to
(23:11):
go out and go to Flux orwhatever.
It's all in that same systemand it's actually cheaper
because I don't have to pay forMidJourney and ChatGPT and Grok
and Claude and all of these.
I don't have to have a serviceon any of them.
I pay for Abacus and Abacususes their API to connect into
(23:32):
all of these other services andI don't have to pay for any of
them.
So if you want access to a lotof those other models,
especially at varied types ofmodels you know, for video and
images and things like that,using an aggregator could be a
way to do that in a way thatdoesn't break the bank, you know
, for your business but givesyou access to a lot of the tools
that you're looking for.
Speaker 3 (23:49):
I mean, it saved me money and it gave me access to
more things.
So I'm curious.
Speaker 1 (23:55):
You know kind of taking a little bit of a
sidestep here.
One of the other things,because we talked a lot about
custom GPTs but there are othertools now out there that
essentially act like custom GPTs, but across multiple platforms.
That I know.
That's something that we'rewell I'm starting to look into
with my agency and looking atleveraging that.
(24:17):
So that way, custom GPTs aregreat.
But again back to what you weretalking about, mike of like
okay, well, if we can get abetter result out of you know
Gemini versus ChatGPT, like wejust care what you know model
gives us the best result.
I'm not married to you knowChatGPT versus Grok or you know
whatever.
Speaker 2 (24:37):
Well, an abacus is actually nice in that regard and
again, matt can speak to Magiand how it functions.
But so within abacus, youcreate essentially a custom GPT,
and abacus is called a project.
Now, they're not exactly thesame thing, but they're similar.
So when you start a project inAbacus, you give it a knowledge
base and you give it a set ofcustom instructions, which is
(24:58):
pretty much what a custom GPT is.
When you do a custom GPT, yougive it a knowledge base and
then you give it its custominstructions, and then you go
from there, right, you're off tothe races.
So in that sense, they'reidentical.
What's nice, though, aboutAbacus projects is is that then
you have access to exactly whatyou're suggesting, john, which
(25:19):
is one.
You can open up a new projectchat, and now it automatically
has access to all of thatknowledge base and the custom
instructions that you gave it,which is kind of who are you,
how do you think?
What kind of you know like allthat kind of thing that you gave
it, which is kind of who areyou, how do you think what kind
of you know like all that kindof thing?
But then when you start theconversation again, same as any
other chat in Abacus.
You can either let Abacus choosewhich model it should use to
(25:39):
answer the questions and, by theway, I should clarify that
model can change at any point inthe conversation.
So I can start with like like Istart the conversation and
let's say I just let Abacuschoose.
I ask it a question and itdecides Gemini 2.5 is the right
model, so it answers thequestion.
Then I follow that with another.
(25:59):
You know, question or reply.
Abacus might decide okay, well,now in this point of the
conversation, claude Sonnet 4 isbetter, or ChatGPT this is
better, or you know whatever,and it will switch models at
will.
You can then also clickregenerate.
If it gives you an answer andyou don't really like it and you
think there's a different modelthat might work better, it'll
(26:19):
give you suggestions of othermodels that you could have tried
.
You can regenerate that answerwith a different model, but also
same thing you can choose.
So at any point in thatconversation I can change it and
say use QuadSonic 4.
Or you know, now I'm at thispoint in the conversation I'm
like oh, we need to use Fluxright now because I want to do
an image, or you know whatever.
So you're in that project.
It still has all the context,all of that information, just
(26:43):
like a custom GPT, but now youcan use any model you want
within that chat.
Speaker 1 (26:49):
Yeah, and I think there's a couple of just
important takeaways forlisteners out of that what you
just said, mike.
So the first one is is that itmeans you have the right tool
for the job you know in thesense of, or the best tool for
the job you know.
It's like you know, if you'llreach into your toolbox and
you're like, okay, I need to,you know, pound in a nail, okay,
(27:12):
and you're like, okay, I needto, you know, pound in a nail
Okay, well, if the only thingyou have is a wrench, like, yeah
, you can make it work, but youknow, a hammer would be a lot
nicer.
And so that's kind of, as we'restarting to see these different
tools, you know, kind of mature, they're good, good at
different things, and havingsomething like that aggregator,
aggregator, abacus or whatevergives you access to all the
tools in the toolbox.
I think the other importantthing, what we've found and
again we primarily use ChatGPT,but just you know from what you
(27:35):
bring up which is, it allows youto keep the results consistent.
So you know, okay, which modelgave us the best results?
And so then if you're havingyour team use that consistently,
you know that same project overand over again.
You know like, hey, here arethe parameters that give us the
best outcome, given this project, that we may do multiple times
(27:57):
a day, you know we may do, youknow, use tons and tons of time,
or do many, many times in orderto get a really consistent
result.
That actually does have a verypositive impact on the business
by allowing you to speed upprocesses and deliver a lot
faster than your competitors,which is super important,
(28:19):
especially on Amazon.
Speaker 2 (28:20):
Well, and if you're a team, I mean also just the cost
.
As you said, different toolsare good at different things.
You could easily be subscribedto 20 different AI services.
You know that do the variousthings within your business that
you need done and you know thistool is better for this, or
this tool is better for this, orthis tool is better for this.
(28:41):
Or you might be subscribed tofive of them and you're like,
look, this one is not the bestat this thing, but it's good
enough at this thing and I don'twant to pay for the one that's
the best for it, because then Igot another subscription that
I'm paying.
So we're going to stick tothese three or four or five
tools, whereas with theaggregator, it gives you access
to all of the tools under onesubscription price.
(29:03):
And if you're a team like ifyou got ChatGPT and you're a
team like ours, where you got 10people you know well.
I got to pay for a seat forevery single one of those 10
people, good Lord.
The price adds up, you know,and if I need multiple tools,
you know multiple AIs, that's anastronomical amount of money
(29:23):
and most of the time, some ofthose tools you're paying for
you only use occasionally, sohaving access to them through an
aggregator is way lessexpensive.
Speaker 1 (29:30):
Yeah, I think we've covered some good ground with
this conversation.
I know that there's a lot morethat we could talk about.
I do think this is probably agood place to wrap for today, so
, with that in mind, maybe I'llstart with you.
Matt, one kind of action itemyou'd have for brand owners who
are listening right now and arelooking for more ways to
(29:51):
leverage AI in their businesswhat advice would you have or
action item would you have forthem?
Speaker 3 (29:57):
So I kind of use ChatGPT as my security blanket.
So most of the two tips thatI'm going to give right now are
very ChatGPT focused.
First of all, if you've neverused a ChatGPT, that's where I
would start is go to the ChatGPTor the custom GPT marketplace.
First of all, if you've neverused a chat GPT, that's where I
would start is go to the chatGPT or the custom GPT
marketplace and find one.
For example, there's Alex Ramos.
He wrote a book called ahundred million dollar offers.
(30:18):
So go find that a hundredmillion dollar offer GPT and
have it.
Help create the grand slamoffer for your business.
That will take you five minutesto find that GPT and use it.
If you have used a custom GPTbefore and you've gotten value
out of it, learn how to createyour own.
It's not as overwhelming as youthink it is.
I've gotten the process downnow to about 20 to 30 minutes
(30:40):
for a decent custom GPT that Ican use over and over and over
and over again.
So that's what I wouldchallenge you is start using
them if you haven't, and startcreating them if you already
understand the power of them.
Speaker 2 (30:53):
And I guess for me, honestly, just kind of
piggybacking on that.
I think you know that idea ofcreating your own custom GPT
sounds, you know, maybe dauntingto some people.
To others it's old hat, youknow they've been doing it for
months.
But if it, if it seems daunting, I think, remembering that you
can use AI to help you use AIOkay, it's a very meta kind of
(31:17):
concept, right.
Like it's hard for people towrap their brains around that.
Like I was literally havingdinner with somebody the other
day some good friends of ours,and he uses AI a little bit.
Never occurred to him that hecould use AI to help him
understand how to better use AIjust didn't occur, and so we
(31:38):
were talking about that.
It was, like you know, mindblown.
So if there's something thatyou think you might want to use
AI for but you don't actuallyknow how to use AI for it, ask
AI how to use AI for that.
It will tell you and you'll beamazed at just how good that
response will be and it can walkyou through that process in a
very systematic kind ofmethodical way.
(31:59):
The other thing this is justkind of a quick tip is that it's
not uncommon that when you aska question of AI, it will give
you, like this giant especiallyif it's something that's fairly
involved Right, and it's like,if you're asking it, how do I do
X, y, z?
Right, and it's kind of a stepby step.
It'll give you the whole thingall at once and you're like, oh
(32:21):
geez, and then if there's somekind of maybe not mistake, but
there's some problem in therethat you can't work through,
then you're backtracking overand over again, right?
So in a situation like that,where I'm asking an AI to give
me a step-by-step, I am veryspecific to tell it only give me
the next step in the process.
As soon as we successfullycomplete that step, then give me
the next step and don't give meany more steps until we
(32:42):
successfully complete the step.
We're on it.
Really, it makes your chatsbetter.
First of all because a chat youcan only have so many tokens,
basically so many characters,let's say that go on inside of a
chat.
So the longer it goes, the morelikely an AI is to hallucinate.
So if it gives you this longanswer and all you really needed
was this little piece, you justwasted all of that, right?
(33:05):
So don't have it.
Do that, have it, give you thelittle piece, little piece,
little piece, and you'll be ableto get to the end of that
conversation without running outof tokens and having it start
hallucinating on you.
Speaker 1 (33:13):
Yeah, and then what I would throw out there just as
an action item for viewers isthink about something that you
do on a regular basis.
So the example that I think ofthat has just been incredibly
useful for me is that I'm the Inaddition to being the founder
and CEO for the Amazon agency Irun, I'm also the chief revenue
(33:35):
officer, which means I'mresponsible for all the sales
calls and follow-ups with peopleand that type of stuff.
I'm a decent email copywriter,but it's been so helpful to say,
okay, I actually have a littleGPT.
It's not super complicated thatI just take the email that I
write and put it into the GPTand say, hey, make this email
more engaging and persuasive.
(33:56):
And put it into the GPT and say, hey, make this email more
engaging and persuasive.
And really what I find out ofthat is that it takes the
pressure off from me of havingto write the perfect email,
because I can come up with, okay, here's kind of my ideas and
put out kind of what I want tosay.
I put it in chat GPT and I'mlike, okay, I really like this
that it came up with this otherpiece, and the seven m dashes
(34:18):
that put in there, you knowaren't a good fit, but I got
enough good ideas from itbecause there's been enough
times where I've looked atsomething to be like because I
customize everyone to make itengaging in that type of stuff
of like hey, you know, you knowwhy is what?
Does that next step look likeand being excited about you know
working together, so it justdoes a really step look like and
being excited about you knowworking together.
So it just does a really goodjob of almost being like that
(34:41):
writing partner, and so itdoesn't have to be something
where it's, you know, building anew product or, you know,
massive project in order to beuseful.
Sometimes, just you know thingsthat you do day to day in order
to give you almost like anotherperson to bounce ideas off of
and to, you know, really justmake things go faster.
So, yeah, I would encouragepeople to just look at you know
(35:02):
what are the day-to-day thingsyou do and can you know how can
you test, leverage it or usingChatGPT in order to solve those.
So, with that, I think this isa great place to wrap on AI.
Obviously, you know this willcontinue to change and we'll
have another conversation in sixmonths and that you know, the
entire AI landscape will havechanged and we'll have more fun
(35:23):
things to talk about then.
So until then, thank youeverybody for listening.
Popular Podcasts
CrimeLess: Hillbilly Heist
It’s 1996 in rural North Carolina, and an oddball crew makes history when they pull off America’s third largest cash heist. But it’s all downhill from there. Join host Johnny Knoxville as he unspools a wild and woolly tale about a group of regular ‘ol folks who risked it all for a chance at a better life. CrimeLess: Hillbilly Heist answers the question: what would you do with 17.3 million dollars? The answer includes diamond rings, mansions, velvet Elvis paintings, plus a run for the border, murder-for-hire-plots, and FBI busts.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.