Retailgentic | Consumer Behavior & Retail Trend

OpenAI’s Chief Economist Ronnie Chatterji joins Scot Wingo on the Retailgentic Podcast to unpack OpenAI's recent National Bureau of Economic Research working paper "How People Use ChatGPT".

From asking vs. doing to AI-powered shopping behavior, this deep dive reveals how generative AI is transforming the economy and how soon “agentic” commerce will redefine online retail.

This episode is more than a window into OpenAI, it’s a look at the data shaping how humans and agents will interact, shop, and make decisions in the years ahead. If you care about where AI is really headed, this one’s not just worth your time, it’s required listening.

Timestamps:
02:00 – Meet Ronnie Chatterji, OpenAI’s Chief Economist
06:00 – How the ChatGPT usage study began
10:30 – What surprised the team most
15:00 – Asking vs. Doing: the evolution of agentic behavior
19:00 – The viral “colored columns” chart and what it shows
22:00 – AI and the future of writing, learning, and work
26:00 – How Ronnie bought jeans with ChatGPT
28:00 – The 2.1% insight: early AI commerce signals
33:00 – Voice + vision = the next leap for agentic shopping
38:00 – The global view of AI adoption
42:00 – What’s next in OpenAI’s research
45:00 – Closing thoughts & how to follow Ronnie

👉 Connect with Ronnie: https://www.linkedin.com/in/aaron-ronnie-chatterji/
👉 Learn more about the paper we discuss: https://openai.com/index/how-people-are-using-chatgpt/

🧠 Want to stay ahead in AI commerce? Subscribe and follow along:
📰 Subscribe to the free Substack: retailgentic.ai
📺 Watch episodes on YouTube & Subscribe for updates: https://www.youtube.com/@Retailgentic

Creators and Guests

Host
Scot Wingo
CEO and Co-founder, ReFiBuy.ai
RC
Guest
Ronnie Chatterji

What is Retailgentic | Consumer Behavior & Retail Trend?

Retailgentic, presented by the ReFiBuy Media Network and hosted by Scot Wingo, is where Agentic Commerce meets retail innovation. We help retailers and brands prepare for the future of agent-driven shopping with actionable news, expert insights, dxis, forward-looking predictions, and practical guidance on Agentic Commerce Optimization (ACO).

And so that ordering surprised me,
both expressing being down there,

but also asking being ahead of doing.

And the thing I immediately wondered
as an economist is, as the agentic

capabilities both develop and are more
widely diffused so people know how to use

them, maybe that would start to change.

And I wanna keep that classification
that we have consistent so I can look at

those doing asking messages over time.

Welcome to the Retailgentic Podcast,
where we explore the intersection

of retail, e-commerce, and AI
agents, which we call Retailgentic.

Our mission is simple.

We are shining a bright light
on the new agentic AI shopping

megatrend, which will disrupt the
trillion-dollar digital retail market.

The stakes are high.

Google, OpenAI, Microsoft,
Perplexity, Amazon, Apple, Meta, and

Anthropic are all fighting to be the
consumer AI shopping agent winner.

There are gonna be some tough decisions
for online brands and retailers to make.

What happens to checkouts?

How should payments work?

Should you continue to invest
in retail media networks?

What does the world look like when
over half of your traffic is not

from humans, but from AI agents?

These existential questions
are the tip of the iceberg.

On this podcast, we're talking to industry
experts, analysts, and influencers

to get answers to these questions,
identify trends, and best practices.

This is Retailgentic, your
podcast partner in navigating

our AI shopping agent future.

And now here's your host, Scot Wingo.

Welcome to Retailgentic In this episode,
we have a very special treat for you.

In mid-September of 2025, OpenAI released
a working paper in collaboration with

the National Bureau of Economic Research.

Like many of you, I read that paper from
start to finish, and it analyzed millions

of conversations that happened on ChatGPT.

It was excellent, but many
of us in the agentic commerce

world had a bunch of questions.

I thought, "Who could we get on
the show to talk about this paper?"

And as fate would have it, the
OpenAI author is Ronnie Chatterji.

He is a professor here locally
at Duke University that started

back in 2006 and left in late 2024
to be OpenAI's chief economist.

It also turns out I know him pretty
well because I go over there to Duke,

uh, which is in Durham, North Carolina.

I'm in Raleigh, uh, where NC State
is, and then in the Triangle,

we're, we're fortunate to have
three tier one universities.

The third is University of Chapel
Hill in Chapel Hill, thus the name

the Triangle, the three cities.

Um, so I go over there and talk to the
students in both the business program,

undergrad and grad, master's, and the exec
program very frequently about startups,

entrepreneurship, startup financing.

Um, there's even a professor there
that goes really deep into marketplaces

and e-commerce, and I've, I've--
I probably go over there once a

quarter, if not more frequently.

Um, so it's a lot of fun, and I find
that it's just really interesting talking

to these professors 'cause they're out
there doing very interesting research.

So I figured, you know, he's-- I'm sure
he's a big shot now, and he's hanging

out with presidents and all kinds of
stuff of, you know, foreign dignitaries.

Uh, he…

the last thing he's gonna wanna do is
come on a podcast about the intersection

of retail and agentic commerce.

But nothing ventured, nothing gained.

So I shot him an email, and he responded
very quickly And today, it is my

pleasure and honor to welcome to the
humble Retailgentic podcast, Ronnie

Chatterji, OpenAI's chief economist.

Welcome to Retailgentic.

We have a very exciting guest today,
someone I've known for a long time.

He's, uh, was…

When I knew him, he was a professor
at Duke, uh, and now he is at OpenAI.

Ronnie, welcome to Retailgentic.

Uh, Scott, thanks for having me.

It's an honor to be here with
you, and, uh, I've learned a

lot from your amazing business
career and all the entrepreneurial

ventures you've been involved in.

Yeah, so imagine I sur- my surprise.

I was reading this awesome paper, which
is how people use ChatGPT, and I'm looking

and I'm like, "Duke, Aaron Chatterji.

Wait a minute, I know him."

So- Yeah … I reached out to you and
I was like, "Man, I love this paper.

Uh, I sleep with it under my pillow."

Um, not to be too, uh, uh, uh, too
fanboy on it, but, um, I do, I do

appreciate you guys putting out data.

It's really fascinating to see how people
are using such this new experience.

Um, and yeah, but before we jump into
the paper, um, tell me a little bit…

I, I know your background, but give
the audience a little bit of background

and, and specifically, how did, how
did a Duke professor end up at OpenAI,

and, and what are you doing there?

Well, thanks, and I'm looking
forward to talking about the paper

and, and why we wrote it, and kind
of the impact that it's having.

So I've been at Duke since 2006.

This was my first job out of grad school.

Uh, I did my PhD out in the Bay
Area, actually during the time of

another big boom in Silicon Valley.

And so when I was at Berkeley, I just
learned a ton about entrepreneurship

and innovation, talked to a lot of
successful entrepreneurs, and I realized

I really wanted to devote my career to
understanding the economic impact of

entrepreneurship and, and innovation.

And so when I came to Duke, that was
obviously a really hot topic, both

for research and teaching, and I found
myself in an environment where I could

do really cutting-edge research, but
also interact with folks like you, Scott,

and others, uh, in our area in a really
exciting entrepreneurial ecosystem.

So it was just a perfect fit for me.

Um, did that research, you know,
wrote a bunch of papers, and sometimes

when you're writing those papers
for an academic audience, you're

not sure that anyone's reading them.

But to your point about sort of the role
of data in informing us and society and

our decisions and just the usefulness
to, to practitioners, I always felt like

the research I did, if I could summarize
it for an op-ed, let's say, in The Wall

Street Journal, New York Times, I was
probably writing a worthwhile paper.

And if you can't do that, then maybe
you should think about your research

question or kind of the work you're
doing because it, it's hard to translate.

It's not gonna have the impact
maybe that you want it to.

And so as I wrote those papers and
found some success as an academic,

I started to get more interested in
how to apply that in the real world.

So I did a bunch of things, serving
in government, of course, and worked

with a lot of companies in terms
of innovation and entrepreneurship.

Uh, and then, you know, when I came back
from serving in the federal government

where I was running, uh, the CHIPS
and Science Act implementation, which

was this really interesting investment
program in semiconductors, OpenAI

reached out around that, and I think it
actually was the intersection that I've

had between academic work and economics
on innovation and entrepreneurship.

Really sort of deep interaction with
business in terms of how businesses are

implementing technology, and understanding
sort of the policy environment around, uh,

critical technologies like chips and AI.

And putting that all together, it
just made it the right time and the

right place for, uh, for OpenAI to
name its first chief economist and

an honor to, uh, to be in that role.

And also, you know, really proud
that, you know, Duke University

has supported me all these years as
well, uh, in, in these endeavors.

So really proud to, to be a member
of both those, uh, families.

So your official title is Chief Economist?

Yeah.

Economist.

Yeah.

Chief, chief economist.

Yeah.

It's, uh- Nice.

Very cool

yeah, it's good.

It's good.

I'm the first one, so I guess they
could give you that title, but, uh,

you know, mostly what it's meant
is I have two main roles, you know.

One is to conduct cutting edge research
on the impact of, um, AI on the economy.

And man, for a data nerd like me, um,
we have this amazing access to seeing

how technology is changing the world
through the use of ChatGPT, through

Sora, through, uh, Atlas, the browser,
through all these different mechanisms.

So it's been really interesting to
get to see some of that data, uh, and

be able to, you know, write papers
around those, um, and, you know, of

course, protecting privacy and making
sure we're using the data in the right

way, but, you know, tell the world
the information they need to know

about how AI is affecting the economy.

And the second piece is doing a lot of, of
speaking and engagement around the world.

It's, it's interesting to me, Scott,
like this technology in particular, and

AI, uh, generative AI, has just sparked
so much interest around the world.

So I feel like whether it's in
Delhi or Abu Dhabi or Brasilia or

Sydney, I, I've been to like 10
countries in the last 10 months.

You just see such an interesting
conversation around AI in all those

places and more, and it's quite
different across those places, which

I thought was pretty interesting.

So that's kind of how it's been shaping
up so far and what I've been learning.

I imagine your, your
chips knowledge helps.

I'm, I'm a…

This may be a huge leap, but I'm sure
OpenAI cares a lot about chips, so yeah.

If I, if I, if- Usually- I, I li-
I like the e-commerce oriented

news, but it's like s- you know,
feathered in with all the cool

chip stuff you guys are doing, so.

Yeah.

So you're, you're- Hardware and software
… I'm sure you're busy on that side too.

Yeah, it must be insane.

Yeah, hardware and software.

I mean, that's, I, I- Yeah … one
thing I'd say is, guys, like what's

interesting about AI is both those sort
of the, the innovation obviously in

software, um, but also in hardware and
kinda how they have to be complementary,

and that's gonna be a really important
part of, you know, the companies and

the countries frankly that are winning,
uh, in AI and other deep technologies.

I have a million questions on that, but
we're gonna keep it narrow for today.

Yeah.

Sure, sure.

Um, so let's talk about the paper.

Like, uh, what, what was, you
know, m- my understanding is

this is not really a series.

Uh, you, you may get, do more, but
this was kind of a first of its kind.

What was the impetus?

Uh, obviously as a chief economist
and a, a former professor, publish

or perish is always on your mind.

Yeah.

I don't know if that's in your purview
here or not, but, uh- … you probably

always have this in the back of your
head, "I gotta be publishing something."

It's great.

So there's probably, I'm probably that.

No, it's right.

No, Scott, so interesting.

Yeah.

I think all these things have mattered.

So what is, you're right, like when
you come into a new role from a

different environment, usually people's
advice is something like, "Okay,

this is gonna be really different.

You know, throw everything you
know out the window and swim

in a new kind of water, right?

And, and learn the ways."

And, and there's a lot to that.

Like, you know, I haven't worked
in a place exactly like OpenAI,

and so it was gonna be…

It was required that I would
have to adjust to things.

But actually it's a pretty research
driven organization, and research is

the origin of all the advancements.

If you think about why people are
at OpenAI, it's because of the

research in large language models,
the advent of the transformer, the

scaling laws, things that have taken
this to where you see it is today.

And so me coming in as a person who said,
"Look, I wanna do economic research,"

wasn't that foreign of a concept as
it might be for the typical professor

coming into a typical organization.

So that was number one,
a lot of encouragement.

Number two is the economics
portfolio was something that

people were really curious about.

So in addition to being a researcher,
I'm also very demand driven.

So if people say, "Hey look, you know-"
How is this affecting the future of work?

How is it gonna affect my company?

I started to get these
questions all the time, Scott.

It's like I often joke, you know,
economists aren't usually the

most popular person at a cocktail
party or or something like that.

Yeah.

But when it comes to this,
people now ask me all the time,

you know, "What do you think?"

You know, "What do you think
about AI and the economy?"

So I saw that there was a lot of
interest in answering those questions.

And finally, I just think it's really
important for the world to have the

best forecasts and indicators that
they can about how AI is affecting

the economy, 'cause it'll be a, you
know, a very major change, I think,

in terms of how we organize, uh, work,
how we organize our leisure time, and

plan those things, and how we think
about the way the economy works.

And I don't know if that is going to be
faster or slower than people's timelines.

Depends on your listeners, but it
is going to happen, and I wanted to

make sure that the research I was
doing was really relevant to that.

So writing a paper about how people use
ChatGPT seemed like a great first step.

And then, uh, tell us a little
bit about the methodology and

where the data comes from.

You're, you're obviously in the belly
of the beast, so- Yeah … so you have

access to all the data, so it must have
been hard to pick what to, what to, to,

yeah, to pick the piece you wanna go with.

Yeah.

I think the two requirements here are,
like, how do we protect user privacy?

How do we do this in a way where we can be
really transparent with all the methods?

So that's another reason we wrote, um,
a paper and we, you know, we intend to

submit it for peer review so people can
look at that, and that we published with

the National Bureau of Economic Research
as a working paper so people could look

at the appendices and the robustness
checks, and all those kinds of things.

There's obviously more work to do.

We can always do better, but
that's why we took that approach

versus, like, a typical report.

First, we had to find a way
to think about, you know,

what are you gonna focus on?

What do people wanna know?

There's lots of different
kinds of data one could look at

when it comes to AI adoption.

We chose usage by consumers.

So when you and I are using ChatGPT
on the free plan or the Plus plan or

the Pro plan, think of us as consumers
who have messages that we're sending

to ChatGPT, and understanding sort of
wh- what those messages are at a high

level, not actually reading them, is
a really important indicator of how

consumers are using this new technology.

Just for, you know, for
reference purposes, you can

think about enterprise data.

You can think about sort of
Codex, our coding product.

You can think about Sora.

Lots of different angles.

Even within enterprise, there's chat,
but there's also the API, which people

like you, Scott, probably build on,
uh, more so than even the enterprise

chat side, which I know is important
in a lot of other kinds of firms.

So we focused on the consumer side,
and that's gonna be important later

when I tell you about the results.

So, so

that's all out?

Yeah.

Right?

That's, that's- No

API, no enterprise

links.

That's exactly right.

That's just- Not, not
in this paper … yeah.

Mm-hmm.

And, you know, I think as a…

This is where it helped
being a researcher.

I thought, you know, it'd be great to
do everything all at once, but it's,

um, all those data sets have different,
um, restrictions associated with them,

how you might wanna use them, how
you might wanna interpret the data

from Codex, let's say, versus Chat.

And so I really felt like it was important
to sort of focus on something where

people could understand what it was
and, and also we could get deeper in

terms of thinking about these things.

The second thing is, you know, we're
not gonna read the messages, right?

That's, you know…

So our a- angle was to try to create a
technique to try to classify the messages,

um, you know, rather than read them, and
of course, to protect privacy, but also

just, um, for efficiency purposes, right?

You can't read all those messages,
nor would you be able to or want to.

So we constructed a method to use
classifiers, and that's a lot of work

that's been done inside OpenAI that we
built on, and we tried to classify these

messages into different categories.

So we didn't need to read them.

We could basically understand,
is this a message where

someone's asking for something?

Is it a message where someone's trying
to do something, complete a task,

or are they just expressing their
opinion or their belief on something?

So that was kind of how we
thought about it to start.

Uh, and we sampled over the, like,
basically the last year, 2024 and

2025, or the two years that went in
our sample for most of the paper.

And we said, "Let's take, you know,
a million or so messages from this,

classify them, and try to understand
how people are using ChatGPT."

So that's how it all started.

And so the surface area
is like prompts, right?

Um, is that kind of like the atomic-

That's the right way to think

about it … classified prompts.

Yeah.

But you're not looking…

I'll call them classified prompts 'cause
I want to be really specific that, and I

understand you're not looking at any…

It's all totally anonymized.

You're not looking at the prompts,
but- Oh, we- … I'm trying to

understand the surface area.

So that's the surface area.

Yeah, de-identified.

I'd say the surface area is
de-identified conversations, so it

would- Conversations … still be the
prompt, but then conversations, exactly.

Okay.

So that was one of my questions, 'cause
I, I got a little confused 'cause it

seemed like prompts in part of it and
then doing convers- So, so if I'm,

like, you know, the topic of what we're
doing is buying products, and- Yeah

so if I'm having a conversation
about fixing drywall and, um-

Right … you know, uh, some, some
of them can actually be c- Will

you actually pick that conversation
apart and multi-categorize it?

'Cause I may be DIY kind of category.

This is- And then I may go into the, well,
I'd want to buy a hammer or a drywall

tool.

We'll sample, we'll sample, so it's not…

Y- you're right that, like, if it's,
um, it's a kind of a multi-layered

conversation where sometimes you're
asking and sometimes we're doing,

we're not necessarily gonna pick
that up at every conversation level,

and that's something we can do-
Okay … actually in future work.

But for now, kind of sampling from
conversations is the way we did it.

And- Okay … to your, to your point
though, as you think about the future of

how people interact with these tools, the
conversations, you know, might go from

asking to doing pretty quickly to say,
"Well, you know, how would I do this?"

And I might say, "Why don't you do it?"

You know?

Yeah.

ChatGPT.

And that is interesting and something
we've been thinking a lot about, but

for now, limiting just sampling would,
would kind of prevent us from doing

some of the things you're talking about.

Yeah.

In my world, we're seeing data, survey
data, not, not your data, but it's showing

that the things you're doing, you're
offering to the consumer in retail,

where they're usually doing research,
find, and buy, you're compressing that

down very quickly, where, you know…

And, and as they trust y- because
you guys have earned this high trust

bar, again, survey data, you, you
haven't said that, that's me saying it.

Uh, it, you know, people are kind of
like, "Oh, it's recommending this.

I'll…"

You know, so they're, they're going
through that very quickly, so it will

be interesting to see that pulled apart.

That- Right … that's my one rec-
my one, uh, ask for the next paper.

Oh, I'm, uh, you know, I'm
taking suggestions 'cause we're

gonna keep working on this.

It's so interesting, Scott, 'cause
I actually, um, I bought a pair

of jeans, actually I'm wearing
them right now, they're my fav- my

favorite pair of jeans, using ChatGPT.

And, uh, you know, I don't shop that
often for jeans and something I don't

really enjoy, and so my wife has always
encouraged me to go trying a bunch

of things, and I, I really enjoy…

I don't want, really wanna
do that usually, right?

Yeah.

And, but of course, like, you need
to get, you know, uh, jeans that fit

and, and you're comfortable with.

So, you know, I basically talked
to ChatGPT about, "Hey, here's the

kind of brands I like, you know,
that's worked for me in the past.

Here's the sizing on those brands."

I kind of do some of this
off the top of my head.

And I said, "You know, I
want something different.

I don't wanna buy the same
brand that I had before.

Um, I wanna buy something
that is related."

And I, I thought, hey, maybe, you
know, I'm a, I'm a, a guy now who works

with a bunch of people who are 25.

So I said, like, "Get me something
that not everybody has, so I can kind

of, you know, feel like I'm keeping
up with the young people," right?

Yeah.

And it recommended a lot of great brands,
and I just went and investigated them,

and ended up buying the pair of jeans,
and it worked out really well, you know?

And so I feel like I can see that- That's

how Ronnie ended up with skinny jeans.

That's … That's, that's the, that's
how I ended up with skinny jeans.

But I feel like, you know, I, to me,
that shopping experience was much more

enjoyable than going to the same store,
asking the same questions, looking at

the same brands, trying to be in a hurry.

And, and I did it, you know, in, in a
waiting room, something for one of the

kids while I was just waiting for them.

And so I feel like, wow, like,
that, that made my life easier.

I don't know if I'm the typical
shopper, but I saw some of the

gleams of what you're talking
about in commerce through that.

What surprised you the most as
you were looking at this data?

You know,

uh-

Uh, the- … as a, as our chief
economist, what, what did you say?

Yeah.

"Wow, that's really, really interesting."

T- two things.

One is there's been a lot of talk
about how people are using ChatGPT

to express themselves and ask for
advice about personal relationships.

Actually, you know, that's
not the largest category.

The largest category is just asking
for information the way you would with

a web search to say, "Hey, you know,
what's the capital of North Carolina?"

That's like the, the kind of question.

And if you think about those use
cases, uh, people want a specific

answer often when they're asking.

The second one was doing, which is,
"Hey, can you do this task for me?"

And I think what surprised me is I tend
to use some of the agentic capabilities,

uh, pretty deeply in a lot of my work.

But I think for most people, asking
is still higher than doing in

terms of percentage of messages.

So if you rank our use cases,
asking is highest, doing is a close

second, and then, uh, expressing
is a distant third, I'd say.

And so that ordering surprised me,
both expressing being down there,

but also asking being ahead of doing.

And the thing I immediately wondered
as an economist is as the agentic

capabilities both develop and are more
widely diffused so people know how to use

them, maybe that would start to change.

And I wanna keep that classification that
we have consistent so I can look at those

doing and asking messages over time.

Well, the topic of our podcast is really
specifically in the product world and,

and it's good you've already used ChatGPT.

You've seen the little
product cards and, and-

Yeah

… I'm sure you know about this
instant checkout thing that has-

Yes … us all excited in our world.

Um, the other thing is,
uh, you know, this is the

When people, when I talk about this
report- Everyone's like, "I don't

know what you're talking about."

I'm like, "You know the one with
the colors and the columns?"

They're like, "Oh yeah, I saw that."

So, so this kind of went viral in
my world because a lot of people…

You know, it's, A, a good
chart, and it's very colorful.

So, uh, did that surprise you?

Have you…

Yeah, that this chart-
Yeah … was very popular?

Okay.

Yeah.

'Cause you have a lot of other cool
charts that I'm sure economically- Yeah.

No, I think- … you were
more excited about this

… no, no.

This- … this chart basically in
terms of the breakdown and the colors

we spent a lot of time thinking about.

Yeah.

And I have to credit my co-authors, and
I wanna call out Kevin Wadman, uh, Carl

Shan, Chris Ong in particular for, like,
sort of thinking about this kind of chart.

And, uh, it didn't surprise me.

I think we knew that some charts
were gonna be more interesting in

terms of, like, let's say LinkedIn
or people who are practitioners.

We knew other charts would be
more interesting to economists.

And so for me, like, this chart I
kind of knew could be interesting.

I didn't realize how viral it
would go, but, um, but I knew

that this one would be popular.

And to be honest, um, sometimes
economists, and if I didn't have

some great data scientists working
with me on this paper, we don't think

enough about making charts that are
readable and digestible and punchy.

Uh, at least I think I don't do enough
of that in my research, so having

people who think about that as part
of their job, and that's why I'm

calling out Carl, Kevin, and Chris.

Really, really important.

So yeah, this is good.

Yeah.

But let me ask you, Scott, like, when…

What do you think about this chart
made people excited, or wh- why do

you think it was interesting to them?

I think it was interesting
'cause the columns, so you've got

multimedia, other and unknown.

Yeah.

It's usually…

So that's interesting that's, that's
second, 'cause usually people would

put that at the end, so I don't, I
don't know if that means anything.

Yeah.

And then practical guidance, uh,
seeking information, self-expression.

And now you've said it, I kind of
understand what you're going for.

Um-

Yeah

… so I thought that was interesting.

Yeah.

Um, and- I, I, I think practical guidance-
Yeah, good … seeking information,

really important sort of use cases.

The other one that comes up a
lot is writing, and that's, like,

something I want to emphasize too.

The right column, yeah.

It's the purple.

Yeah.

So in the purple, right, sorry.

I think this…

It's narrow on my chart right now, but
the- Yeah … the writing, which is good,

but, um, the writing is really interesting
both in the consumer applications, but

especially when you look into other
applications, let's say enterprise,

writing is really, really important.

And for some people they say, "Oh,
well, that's, that's sort of boring.

Like, I'm not interested in writing.

That sounds very simple."

But writing is actually the foundation
of most of the jobs that people are

doing in an office, let's say, like what
traditionally called white-collar work.

And so you think about the ability
of ChatGPT to help people write,

that's really, really important.

The second thing I'll say is, um- A lot
of people, and I'm a dad of three, so the

first thing I thought was, gosh, like, I
hope that people aren't just outsourcing,

like, this writing process to ChatGPT.

And what it turns out is two-thirds
of this is actually people submitting

things that, you know, is, are already
written and asking for edits or summaries.

And, you know, there are challenges to
that too, but one thing I started to

think about is, like, for people who
don't speak English as a first language

or might not be the best writers, it
allows them to refine their writing

and polish it and make it better.

Or maybe make it co- more cogent or punchy
and get to their, their point across.

I thought this is a big deal, and
I think as we refine writing with

AI tools, and we gotta do it in the
right way, and I talk a lot about

this with educators, and I'm asking
a lot of the questions of educators,

I think it has tremendous potential.

And it seems like a pedestrian
category compared to all the other

things people think about, but I
think it's really, really important.

And then the one that caught my attention
the most is in this red column, which

is column number four at the bottom,
and it's a tiny, tiny slice, which ma-

made me a little sad, but we'll talk
about that, is purchasable products

at, at a, a, a, a very tiny 2.1%.

So that was you shopping for
your jeans though, that day.

That's probably me shopping for jeans.

Uh, if, if that was in our data set.

That's exactly, that's
exactly right if we do.

But I think, you know, to me, Scott,
what I've learned from this research,

and it might be useful for folks
who are listening who are super

interested in retail, you know, a
lot of it's about awareness, right?

It's, it's…

I'll give you an example.

Deep Research is one of these parts
of ChatGPT, a tool you can use that,

you know, in the previous versions of
ChatGPT, it was sort of something you

had to use, click on a dropdown menu, and
you could sort of like queue it up, and

Deep Research was really useful for me.

What it does is it goes out and really
does, like, a more, um, sort of serious

and rigorous investigation of, let's
say, the prior literature in a particular

field, like monetary policy- Mm … and
comes back with a really well-organized

summary, if that's what you're asking for.

And as someone who's an academic who wants
to get smart on a couple different topics,

that is, um, really, really valuable.

And I go to people all the time.

I do these trainings around the
country and around the world where

I'll try to help people use ChatGPT.

I don't know if I'm the best person,
but I know a little bit about it.

And I'll say, "Hey, Deep
Research can really help you."

And they say, "Oh, what's that?"

And they're really- Yeah

experienced users, but they don't
know, always know the features.

And I think when it comes to, like,
thinking about personal products, I'm

sure there's lots of things like that.

As awareness increases, as people
understand the use cases, I

expect that to, to, to change.

So that's something I'll be watching,
but I think it's super interesting.

As an entrepreneur, I'm a, a
eternal optimist and I was, I was

thinking, you know, I don't really
understand the categorization thing.

I bet it's something in there.

And, and you've given me hope and,
you know, so, so if someone started,

I do think most people start with
these kind of like, "Hey, I'm

thinking about getting some jeans."

And, and that may show up, uh,
especially, you know, like the

whole home improvement category.

You almo- you don't start with
like, "I need a compression drill."

Uh- Yeah … unless you're like an expert.

Yeah.

Yeah.

You start like, you know, "Something's
gone horribly wrong in my driveway," and

you end up getting a compression drill.

Um- Yeah … the way you're sampling
works, that could get caught up and put

into more of the, the education bucket.

So-

That, that's correct.

Yeah.

That's correct.

Yeah.

And I think you're right.

Someone might be asking a question,
but it's actually sort of about

intent to purchase downstream-
Yeah … and our sampling methodology

would undercount that, which is
super interesting to explore.

I'll tell you one more thing.

It's obvious to you, but just
for folks who listening who

might also be interested in this.

I mean, the video and the voice ability
changes the game here, too, because

sometimes, particularly like let's say
a tool around the house, like, I don't

know what tool I need to buy, you know?

You might be better than me at this,
Scott, like knowing the right tool, but

if I can show ChatGPT a picture of here's
the thing I'm trying to fix, they might

tell me, "This is actually what t- what
tool you're looking for"- Yeah … which

then gives me the cue to buy that thing.

I would've been completely lost without
it, and, uh, that's very helpful, and

I find that to be sort of in terms
of, you know, awareness to intent

to purchase to, like, the sale, a
really, really important aspect there.

Yeah, I use voice.

I, I, I've used both those together.

It's like I had some weird growth
on a wall, and I took a picture,

and it's like, "That's black mold.

Do this immediately."

And it, it took me to buying products
very quickly, which I- That's

interesting … I did instantly.

That's really … Yeah,
that's very interesting.

It got me freaked out.

I was kind of urgent about it.

I was like, "Oh-

Yeah.

And this, I mean, this- … ChatGPT's

urgent"

… yeah, this is amazing, right?

I think we end up taking
this stuff for granted.

Now I use the video all the time, but if
I would've told you, I don't know, pick a

time period, 18 months ago, 12 months ago,
that you could basically talk to ChatGPT

as you were looking at the wall, figure
out what it is, and then buy something

to deal with it, like you would've told
me that was, uh, science fiction, right?

It's, it's pretty cool.

So I'm- Yeah … I'm interested in

that.

The sense of wonder, I, I still have it.

I feel like I'm in a Star Trek episode.

It's amazing.

Yeah.

It's like amazing.

Yeah.

I love it.

I, I imagine you didn't really do
demographics- Mm … so there's no

way to kinda see what the younger
audience is, because you're not,

you're not tying it to user sessions.

So there was no way to look at that as so

… Is that true?

We, we do, um, in, in a data clean
room, we were able to match to

some, some user characteristics.

And so as we continue to release
more research, we'll tell you,

you know, which users are more
likely to do one or the other.

Mm.

We did use those demographics to try to
understand what people were doing that was

work-related, let's say, compared to their
job, uh, compared to not work-related.

So we do have some limited demographics
in there that we were able to match,

and that's one of the innovations
that we had, uh, in this paper

that no one else had done before.

A lot of the data on usage
has been really interesting.

I learn a lot from all the
other people doing this.

They haven't had demographics.

And so we were able to say, for
example, that, uh, users with typically

masculine names, uh, versus users with
typically female names or feminine

names, that gap between those two
kinds of users had closed, uh, almost

entirely by the time we did our study.

When ChatGPT first came out, you were
really … The platform was dominated

by users with typically masculine
names, a big gender gap at the

beginning, inferred by name, of course.

We don't exactly know people's genders.

We have to infer it by name.

Really important to say that.

Yeah.

But over time, we're the first ones to
report this, that gender gap closed.

And so really, really interesting
to put some of those demographics in

there, um, and could do a lot more with
the age variables as well, I'm sure.

Yeah.

So, so one thing I've, I've had a
lot of discussion on LinkedIn with

people, um, number one, people love to
argue over what agentic commerce is.

So I won't- Yes … I won't, I
won't tie you up on that one.

You can keep me out of that one.

I won't get down that

road with you.

I'll, I'll keep you out of that one.

Uh, but then, you know, there's survey
data that shows in, in Q1, like maybe

33% of the folks kinda said, "I've used,"
uh, you know- a, a, a gen AI for shopping

or researching products, and now it's
starting to show up as high as 75%.

Wow.

And I just kind of
wonder, like, so the 2.1

and the 75 don't line up.

Yeah.

So it sounds like two
things could be the cause.

Number one, you're, you kind of pointed
out with deep research, maybe people

just aren't as aware of the shopping
thing, and it's kind of a lagging…

This stage is a bit of a lagging
indicator of where behavior is gonna go.

Yeah.

And then the second one would be my, my
theory that there's a little sampling

thing going on in there, that it's hard
to, depending on where you're sampling the

conversation, if you sample the top of it,
it's gonna feel way more like information

gathering than buying and, you know.

So but as people get more familiar with
the abilities and, and a lot of this

is new, so the cards, the product cards
that pop up, and then certainly, you

know, we're, we're only like, what?

45 days out from the instant
checkout announcement.

Yeah.

Um, people even knowing and, and,
you know, not everyone's live yet.

So the people being able to do this
and, and faster, so are, are those two

theories that kind of jive with your

economics?

I think those are two good theories.

Yeah.

And the one I'll add, it's not gonna
explain the whole discrepancy, but

I think it's possibly relevant here,
which is, you know, we classified

the percentage of messages, right?

You could also think, uh, the
surveys that you're citing

are at the user level, right?

Yeah.

So it could be a user sending lots of
messages, but they do more messages

asking questions like, "What's
the capital of North Carolina?"

than they do about shopping.

Yeah.

And if that's true, that
could create spread.

Now, the, the spread is larger than
probably would be explained by that.

But that, I think, if I'm thinking
about the numbers right, is another

way to think about this, is percentage
of messages versus, versus users.

Do you guys plan to put out more…

Is this, like, gonna be a series
or, or is it- We could try

like, enough work that
it's hard to kind of like-

Both, both Chain together.

Yeah I mean, my team will tell you both.

No, I, I love this kind of work, 'cause
I mean, Scott, at the end of the day,

you nailed it, uh, at the beginning.

As someone who's a researcher, I
think the most valuable thing I can do

inside the organization is do research
on the impact of AI on the economy.

And while some of that research, you know,
obviously is really useful internally,

I think that OpenAI's mission to benefit
all humanity, it requires that as much of

this data that I can share and research
that I can share, I put out there.

And of course, I never can do, do it all.

Like, it won't be perfect.

There'll be some things I
won't be able to do that with.

But continuing to write research papers
that people read and learn from, like this

one, is definitely a goal, and what you'll
see much more from us going forward.

The other part of your question
is also, yes, it's a lot of work,

and I can see why it's something
that's hard to stay committed to.

Because, you know, in a lot of
organizations, they do it once.

Maybe they don't wanna do
it again, 'cause, uh, they

don't think it's worth it.

For me, I wanna keep doing it, but,
you know, as a manager, allocating

the resources to get it done, uh,
data science, econ resources, but

of course, also comms, legal, all
the other people we have to rely on

who help us make these, uh, projects
work, uh, it taxes them as well.

So I just try to think about that
from a leadership perspective and

make sure that we can organize
ourselves in a way to do the good work.

And, and people like you and others who've
run companies and groups and things like

that know that's a big part of it as well.

Well, I know you're super busy.

You, you're probably off to jet somewhere
to go solve, uh, you know, answer

people's economic questions about-

uh, what's going on.

Uh, really appreciate you
coming on and, and helping us

understand what's going on here.

It's super helpful.

Uh, you know, would love more
data that you guys put out.

You know, thanks for doing this, you
know, from, from just anecdotally from,

uh, a little, little corner of the world.

We appreciate it, and this kind of
data's amazing and very helpful for

us to understand what's going on.

Well, Scott, thank you, and, uh, I just
say, uh, your, your work, as we talked

about, is an inspiration to me, and it's
made a huge impact on our community,

so it's an honor to be with you.

Thanks.

And if folks want to…

Are you active?

Do you, do you kind of like,
um- LinkedIn … have a…

Just LinkedIn is the best

way to- Yeah.

Yeah … reach you?

LinkedIn's the best way to reach me.

That's…

I'm trying on all mediums, but really
LinkedIn's the one that works the best

for me right now in terms of putting
my messages out there and trying

to get, uh, people to see the data.

So we'll, we'll keep posting stuff about
the research findings, so if you can

follow me on LinkedIn, that'd be great.

Awesome.

We gotta get Sam to have you on one of
those little podcasts where they- … sit

there and, and talk about stuff.

That'll be fun.

Yep, exactly.

Thanks, Ronnie.

Really appreciate it.

Thank you, Scott.

Appreciate it.

Follow, rate, and share to stay up
to date with RetailGenic, your guide

to the future of AI shopping agents.