Model Dev Memo Podcast, Season 7, Episode 5: Ads in ChatGPT (with Rishabh Jain)

You know those channels your colleagues keep bragging about? The ones getting all the credit? Yeah, they might be doing squat. Attribution makes every channel look like a hero, even when it's a zero. Incremental tells you who's actually [music] doing the work. It's like a lie detector for your marketing budget. Start using Incremental today. Get your demo at incremental.com. That's i n c r m n t a l.com. Mention that you came through the Mobile Dev Memo podcast for a special 15% discount for [music] the first 6 months.
The problem is that the distinction needs to be drawn between the [music] competence of the economists and the correctness of their analysis.
Welcome to the Mobile Dev Memo podcast. This is your host, Eric Seufert, [music] and I'm joined again by Rishab Jain. Rishab, welcome back.
Thank you. I'm always happy to be here.

So, we're going to talk about ads in ChatGPT. That is the entirety of the topic set for this podcast episode, and I wanted to have a dedicated episode for that topic because I think it's important. And the first person I thought to invite onto the podcast to discuss that topic was you because you came up with the best answer of anyone for the question that I've been asking for the last year on the podcast when it came up, which is when will ChatGPT introduce ads? And you said they will introduce ads before Fiji Simo's cliff is up, which has happened. That's exactly what happened. So, Fiji, I think, joined in May, and we we are getting ads, I guess, in February, or maybe even now, I don't know, but soon, very soon. And so, you were right, and that was a great answer, and I think that touched on a lot of the motivations there, too. So, first of all, congrats on on on getting that right. And second, like just give me your brain dump of how

you responded to that news and and what you think about the proposition of ads in ChatGPT.
I think I actually found my earliest tweet. I like was curious, what was the earliest thing I said publicly about ads in ChatGPT? And like I found this tweet that I had in 2022 that said like, you know, can't wait for the time when like I I got the format wrong, but I said like can't wait for the time when like the new ad units come in ChatGPT that are going to be natural language. And so I like definitely got the format wrong, whatever 3 and 1/2 years ago, 3 years ago. But then when they hired Figuicimo it became very apparent to me that they understood the economics of the business they were building and the types of capabilities that they would need to develop underneath that. And so I think that they are both economically rational and want to build the best product for consumers possible, which is

why they hire someone like her. And I think that ads are the best way to do those two things. Like to get the best economics and to build the best consumer product. Because I I think this is an area where you and I agree, which is ads get you a more consumer aligned product than not ads. And have less perversions than other economic models, basically. And my instinct when I saw the ad format was this is like not dissimilar to what you were dis- I mean, basically they just took space on the screen and said like, "Okay, now this space on the screen is where we're going to put sponsored things." Where you have to see what the final format looks like, obviously, so it could obviously this is going to you know, possible to change. But I think that the initial format of being called out in a separate section makes sense. And I think that the initial way that they're rolling it out also kind of makes sense on like a CPM basis because today they don't know anything else. Which yeah, we should dive into. I don't

want to like go too deep, but my my initial reaction was like this makes sense. This is great. I was like genuinely happy. And I genuinely have the most excitement that I have had in a very long time that like the American consumer is going to get surplus as a consequence of these decisions. So.
Yeah, I want to talk about that. I want to talk about the launch strategy. Let's put that aside for a second. I want to talk about the benefit here to consumers, right? So you know, predictably you got the what I think are mistaken the predictable hostile reactions. Right? Which are well, here we go. Like they're going to you know, whatever just you name the kind of conspiratorial framing. The word I'm anchoring on, I don't want to say it cuz I don't like, you know, giving this word power or this phrase power, but the enshittification argument is the one that was circulated kind of the most prominently, which is like oh well, that's the idea. That's like it's these tech companies they give you something

for free and then they continuously degrade the experience because you're hooked and you are weak and you have no agency and you can't disengage yourself from the products, which I just disagree with on a philosophical basis, but you know, that ignores the fact that in offering ads they're giving more capacity to the accounts that the ad supported users will will experience, right? I mean, they're getting the access to the better models. You're getting a better experience in exchange for these ads being placed in the product. Now, you might argue in the medium term maybe that like it's a trade-off cuz I'm getting something, but I'm giving something because oh this is an ad, right? Um that's distracting or, you know, it's something I don't care about or it's, you know, but we've seen with Instagram and Facebook and and YouTube and you know, people can pick out very specific instances, but in the main I think you can make the argument that like when these ad systems get really good, when they have a lot of

data to work from and the systems are sufficiently sophisticated, they're showing you stuff that you're actually happy to be exposed to because you buy it and it's stuff that you wouldn't have otherwise discovered. I mean, this happens to me all the time. And so in that sense, it's all upside. Right? I mean, if you get to the power of the Instagram ads engine, it's all upside. And people just ignore that and I think it's disingenuous to do that.
Totally. I mean, my version of that same argument is I basically ask the question, what products would get recommended if you did not have ads? And like now let's go down that trajectory. So like I think what becomes easy to understand is let's just pretend we have a system with no ads and like you're doing your searches and all that sort of stuff. Some amount of those searches are going to be commercial and they said very clearly that the ads are not associated with the query. So and I understand that. For the sake of the sort of thought exercise, the only products you will ever discover

if you did not have ads are the ones where when you are doing some sort of organic query of some variety, the products that show up in the context of the conversation you are having because that is the organic thing that happens, right? And like you have to have a response that has products because so much of how you operate in life is like products that you have to purchase. So it's like, okay, that's what's going to happen. Whose products are the ones who are going to show up? The companies that have the ability to have the most amount of their content show up in the places that are trusted by the LLMs, whatever that means. I'm not I'm not trying to debate whether or not you can game the LLM. Like I'm actually saying like let's pretend you cannot game the LLM no matter what. Whatever the true trusted sources, that's where it's going to get it. In that world, there is a 0% chance that a new company who has just started will ever find that customer because there is no trusted data about

that company because it's new by definition. And what we are then saying is we prefer an internet where a new company will never have a chance to actually show up to a customer who could want their product. And that to me is a squarely worse internet. And like to me it is like that simple toy example is why I think that consumer surplus gets generated and why it's like an economic good, a experience good that ads get created because you are giving an opportunity that previously did not exist and could not have existed if you did not have an ad system. And I think that that totally gets lost and I think we can like bookmark this to the 4% fee for shopping because I think that they are related. But yeah, I I don't agree at all that ads make products worse.

I think it makes the total experience better. I think it is actually better for the consumer and I think that yes, it also allows for the economic engine that allows for the product to continuously get better over time. And so the best products are the ones that have the best economics because all of those profits get poured back into making the product better and so on and so forth. So yeah, I I I have a really hard time interacting with the idea that the product degrades because of ads.
So but you called that a toy example. So first of all, I wrote this piece a while back, affiliate links, personalized ads, and chatbot revenue optimization. I made the argument that just from an economic standpoint, the affiliate model that chat GPT implemented with the instant checkout was just suboptimal relative to ads. It's like a really long piece, right? I made this argument, but I invoked that idea that like you're just going to get existing stuff. I mean, it's you're you're not going to have an ability to explore exploit. Or I mean, you you could, but you might not do

that. And anyway, it's probably going to preference the stuff that's just like lower cost cuz that's going to be correlated with the conversion rate, right? And so, without the bid to mediate that, you won't have any ability to sort of penetrate that dynamic with something that's new, right? That's how new products get discovered is with ads, but it's not It's this is demonstrable. This is This has been studied. There's a paper that I cite from the from researchers, they published it with the Federal Reserve with the Atlanta Fed. And it's called the expansion of varieties in the new age of advertising, and they find that growth in internet advertising spend coincides with an expansion of the availability of products to people. Like it cuz it just creates the opportunity to get new things in front of people that they can then buy and try, right?
I agree with your affiliate take. I I I want to add one I mean, I've read this piece and other people should read this piece, and the one thing I would want to add to it is you can actually already see this in practice. So, when you look at

editorials where there you have every confidence in the world that the editorial piece is not an advertising piece, the consequence of those types of editorials where they're doing product reviews is that all of the links go to Amazon.
Right. Right.
And so, this is like not a mystery. And so, the thing that I would just add to like your case is that look, you can like and still not like ads, that's totally fine. But just go to any editorial and ask yourself, why is it that I always get directed to Amazon? It's because what they are doing is they're optimizing for the checkout. And in optimizing for the checkout, as you said, it's price, and then the other one is distribution. And so, it turns out that consumers have the highest trust in both cost and distribution with Amazon. It also turns out that Amazon's affiliate model gives you credit for the entire cart, not just the product that you came into purchase.

And so, you end up with everybody biasing toward Amazon, which I don't know if that's better than just having ads in the article.
[laughter]
And and the perversion to me is much worse, because all you're doing is feeding basically one of the biggest companies on the planet. Right? And so, I I struggle a lot when with the this like 4% affiliate fee, people are saying like, "Oh, yeah, that's better than ads." I'm like, I struggle with that. Like, the current world of affiliate just pushes everything to Amazon. Like, do you guys want everything to go to Amazon? If that's what you want, then okay, then you know, I'm not here to like apply judgment to the fact that you want everything to go to Amazon, but just understand that's the outcome.
Well, don't get me started. I want to stay I want to stay on topic with ChatGPT ads, but let me just let me just make a couple points about why it's sort of category error to look at the 4% and compare that to what you might be paying in ads as a percentage of call it GMV, right?

For that one product. So, So, here's why, right? So, first of all, when you get the affiliate link from ChatGPT, I mean, that's not under your control. So, you can't control that. That's not a lever that you can tune for growth. That's just, you know, a a decision that's made in by this mechanism that is is completely opaque to you. Right? So, that's not it's obviously not a a method of growth, right? You can't rely on that. You know, it's not something that you can systematically influence, which means it's not a strategy. So, that's one big flaw there. Whereas, ads are, you can influence it. You can influence it by your bid and your budget. Like, those are the ways you influence it. You can control your growth in that way. But, putting that big flaw aside, that 4% is not a So, first of all, you you when you're talking about you can't talk about percentages when you're talking about customer acquisition. That just doesn't make any sense, right? Customer acquisition is set with an absolute dollar term. It's not a percentage, right? You can't think about give yielding a percentage in exchange for customer acquisition. That

just That just breaks the economics, first and foremost. Even if you can calculate your CAC as a percentage of the overall GMB, that doesn't mean that that's the way you priced it, right? Because you can't really price it in that way. Then, put that aside, that 4% is not for CAC. It's for a one-off transaction. You have no relationship with that user following the transaction. If I pay for an ad and that person comes to my Shopify storefront, I get their email address. That is a relationship that I have. I can remarket to them. I can send them promotions discounts. So, whatever I spent on that initial touchpoint, that can that's sort of like is amortized across the lifetime the LTV of the user, and that's how I think about it. It's CAC to LTV. It's not CAC to that singular first purchase or transaction. And if that is, well, okay, that's a different story, but that's not true for most products. So, there's just very fundamental differences here. But, I think, and I know you said this, I agree, that 4% is going to go to zero. I think

this was just a means to to bootstrap the data that they need for doing advertising targeting. And once they get there, they probably won't need the affiliate model, and it'll probably give way to advertising anyway, because guess what? What's going to happen in the affiliate scheme when you've got a bunch of products that you could theoretically reasonably fill in that spot. How do you just how do you determine which one to show? You use the bid. So the affiliate model naturally is going to give way to the advertising model. Talk talk to me about that. What what do you think is going to happen in this affiliate scheme, this whole product uh scheme, the instant checkout when they get ads ramped?
Okay.
[laughter]
I think there's lots of reasons why the checkout on in platform is like a red herring. It is certainly for the data. It is certainly for the data. And I was at this Agendtic Commerce dinner last night and someone was asking me about like what I thought about the like checkout within ChatGPT.

And I asked them like well, what do you think about the fact that TikTok Shop has just said that merchants cannot fulfill their own orders that like TikTok has to fulfill it. Like what why do you think that happened? And they were like, "Yeah, I don't know. And are those connected?" I'm like y- y- yes. It cuz it turns out e-commerce is like more complex than we give it credit for in this like type of transaction where you have like a store in platform, especially when they're trying to take an affiliate fee and own the customer in that context. And it is just an incredible amount of pain for the platform to hold. But they will take that pain in order to actually understand the consumer behavioral data for what leads to the transaction. But they don't want to deal with it because they now have to think about the fact that from the consumer's perspective said, "Hey, ChatGPT, I trust that you're

going to make this transaction on my behalf." And now the now ChatGPT is in the middle of this situation essentially where even if they clarify things like the merchant of record is the brand and all this sort of stuff, they're in the middle of this messy messy thing where they are relying on the merchant at the back end to fulfill that transaction and for the consumer to get it and all that sort of stuff. It turns out it's a pain. And the way we know it's a pain is Meta stopped accepting payments on their platform and TikTok said, "We're going to fulfill it, but we have to fulfill it." Because when you accept the payment from the consumer, whether you like it or not, you are somehow responsible for that consumer's experience. Whether you clarify you're not the merchant like you know, all of these things are sort of in the details of what how the consumer has now shifted their expectations. And we can just see it because TikTok has made an affirmative decision. And then that affirmative decision by TikTok

changes the economic model. And so the all of these things are related. They are not unrelated because when you do commerce on the internet, the places that gather consumer attention are exceptionally gathering attention and merchants are merchants. And you need to have the merchant to be the merchant and you need to have the attention be the attention. And so what's going to end up happening is yes, you have this 4% fee. I don't even know why it's four. Like it could have been three, it could have been five, it could like the number is certainly irrelevant in the grand scheme of things. And then it's just going to dwindle to zero because they're not going to want to deal with that transaction responsibility. They're going to want the merchant to be the merchant and they're going to find a elegant way to do the handshake to pass the person to the merchant site to complete the transaction so that the merchant can then fulfill and the consumer knows that they are interacting with the merchant.

Because that's how like the whole thing gets disentangled. And I I think that people I don't know why one would expect that just because there's a 4% fee that it is a model that will like grow over time when we have seen enough examples where the best model is to hand it over. Like like I I can't tell what's different this time.
If you're swimming in dashboards, but still arguing about what actually drove [music] installs, this is for you. Branch is an AI-powered MMP built for growth marketers who care about signal quality and outcomes, not just reports. You can quickly answer questions like, "How is my TikTok spend really performing?" Or, "Which [music] partners are driving net new users?" And even launch campaigns that move users from offline to app without breaking attribution. Branch's AI proactively surfaces [music] what's working, flags issues early, and takes care of the busy work like link creation and tagging, so you can move faster and spend smarter.

Learn more at branch.io. That's branch.io.
I want to talk to you about that. Why do you think TikTok is leaning back into social commerce? Cuz they were pulling back. They had rounds of layoffs. I mean, Facebook, you know, abandoned it completely. Why do you think TikTok's leaning back in? I found that pretty odd. I was on TBPN yesterday. They asked me about this, and I didn't really have a good answer. I mean, I other than, "Well, they are." What's the reasoning? I don't I don't really get it.
I think they want to compete on product search. Like the reason you start fulfilling products the way Amazon does is that you have an expectation that product search is going to originate on your site. And it has actually less to do with the content feed. And so and of course, we don't have any of the like I don't have enough of the like private information held inside of TikTok to know how this is shifting. But for example, WeChat their agentic systems in China,

they're already doing like closed-loop transactions basically through an agent interface because the search is now natural language. And so if they believe that they are going to capture enough audience where search intent within the context of TikTok is high enough and then they can fulfill against that, then they can own the product search interface. And I think that that that space is extremely highly contested. And I think even Amazon is scared. And the way that you can know Amazon is scared is because for the first time in their history, Amazon allows link out directly to brands. Right? And so we we know that there is motion in the product search space and we know that there are multiple apps that are trying to compete for it. And people are willing to give up on many other things in order to be the place where people prefer to do product searches. And I think that that's fundamentally why you end up doing fulfillment, too.

Yeah, so you got that there's the Amazon is shopping brand sites directly. That's I wrote about that a while back. Yeah, that's that's really fascinating cuz they're doing the link out. So they're not selling on their site, but it's it's part of the ads product. What and so it's yeah, that's a great point about TikTok cuz I actually I did write about that a while back where TikTok the developments with Shop were really more of an attack on Amazon's business cuz they were trying to drive a lot more discovery through there. That's good That's a great point. I didn't remember that yesterday when they were asking me about it. But yeah, that's I think that's probably right. Okay, I want to go back to to ChatGPT ads though because I think this was like a watershed right moment. I think it it's a really important moment. Let's talk about the state of the product at launch. So they said $60 Well, this has all been kind of leaked, right? So we can let's assume that this is right, but you know, this is coming kind of second hand. $60 CPMs, less than a million dollar commitment, essentially no measurement, essentially no targeting. Talk to me about that. What do you How do you think this is going going

What Who do you think is the right advertiser base for that product? You just kind of riff on the state of the initial product.
Yeah, I think So, first of all, I like to view this I always give the benefit of the doubt. Meaning, if I was maximally intelligent, why would this be the single best way to launch the ads product? So, like why is it that this would is better than like a bidding system out the gate, for example, right? And the best explanation that I could come up with is they don't know yet how effective their targeting algorithms are going to be given that they are not giving any sort of like targeting capability to the merchant. And so, a different way of saying the same thing is Facebook today and Google today they came from a heritage of saying like put the pixel and we're giving you these targeting levers and slowly taking them

away. And Open AI doesn't have any of that baggage. So, if Facebook could start a fresh today, would they actually just have an interface where you actually can't you don't have a access to any of those levers? And that's how is that actually better for them? And so, is this actually better for Open AI where they are like intentionally giving you the minimal possible levers because they know something we don't and they actually have confidence in the capacity of their ad system to deliver extremely good matches between what the ad is and what the person is. But they don't want to reveal any of that. If that were true, then the best way to launch the ad system is just to say, "Hey guys, we don't have anything yet. It's a $60 CPM and we're going to launch it." Okay, and you kind of like play dumb, right? As it were. Which by the way, I'm not saying that they're doing. I think they're playing very smart. And so, I think to assume that they don't have

like sophisticated systems could be a mistake. Like we cuz we don't know. And I also think that charging a CPM is very smart because by charging a CPM, they're not devaluing any part of their future monetization structure because they're doing something that is so obviously not going to be the future of how this thing gets priced. And so, I actually think it's like the best way for them to get the direct data of how effective their systems are without making any promises whatsoever to the advertisers. I actually think that likely who they are targeting is like all of the largest advertisers on Meta and and TikTok. Because I think that what they're doing is again maximizing data. And so, like what you want is actually people who are in those DR type of systems or who are spending in DR systems even though this is a CPM model

because what you're doing is trying to maximize signal, basically. And you don't need to maximize signal at the end transaction, you can maximize signal within the context of your platform. That's a great sort of analysis. I I I just don't know how else they would start this. I mean, you've got you've got to start somewhere, right? I mean, what what are you going to do? Like you have no conversion data to use in optimizing the campaigns, right? Like you have nothing. So, so just do exactly what Netflix did, and Netflix did charge did the exact same thing. They charged $60 CPMs and that right out of the gate $60 CPM. They had no targeting, no measurement. And they had to they had too much demand. They had more demand than they had supply. No, they had to return money to advertisers. They couldn't spend it all. So, I mean, they know it's going to be successful. People are lining up to buy this. You see a bunch of like, you know, kind of the DTC crowd saying that who would ever buy this? Not you, but that's not who it's for. Like they needed to start somewhere.
Totally.
The real question is the rate of change. How quickly does this evolve? That's the That's what they will be evaluated by, not the starting point. The starting point can be the same for everyone. Facebook started the same way.

Right? It's not like they they could have built some intermediate product. They would have been making a lot of assumptions that might not hold up.
Totally. And all of their actions are consistent. Which I think is I think we're both agreeing is all of their actions are signal maximizing actions.
Right.
And promise mitigating actions. So it's like they're keeping maximum leverage for themselves and maximizing the signal that they can get. Which is why you do things like this like 4% affiliate thing and get you get transaction data from there. Right. You get all of the signal on how people are interacting with your ads and how good the match is. Like you're just like maximizing your signal really fast. You're trusting that AI is good at presumably I again I don't know. You're trusting that AI is good enough that it's going to be able to react to those signals really fast. And we're going to wake up in like four to five months and this thing's going to be awesome. Like it is going to be awesome at like matching the right ad to the right

person. I made this comment on like how much revenue I think could come from it, but I think like it is the million dollar thing is also I think it's a signal maximizing tactic. Which is like we want to make sure we get as much breadth of signal as possible so that we can train our systems appropriately. We don't want to over index in one way or another. Right? But we want to get enough signal. And so it's like I think every move and every action should be viewed from the perspective of assume these people are really smart and they're going to maximize the inputs to build the best possible system as fast as possible. And then it kind of makes more sense.
Yeah, and I think that's a sort of logical flaw or just a process flaw that I'm seeing be committed. It's the not assuming these people are really smart. Because they are and they all come from Meta, right? And so like they know what to build. It's not like they they have to envision some new paradigm here. I

mean, they know exactly what to build. I think people you know, it's let me just I want to I'm going to I'm about to I'm about to monologue here, but let me let me correct myself in just a second ago. So the shop brands sites directly, that's actually not an ad placement. I thought it was cuz it shows up in the search, but that's not they don't charge for that. You can just opt out. If someone searched for something that doesn't exist on Amazon, Amazon will just direct you to that brand's website. They're not charging for that. I thought they were, but I just looked it up they're not. Uh it says it's called buy for me. But yeah, the shop brands sites directly, that's not an ad unit. That's just free. That's a giveaway. That I mean, it's in beta. I think they probably will charge for that at some point. But right now it's not anyway. So that's Amazon just pushing people out like, "Hey, look, if you came with a need we can't fulfill it, it's better for us to be the person that routes you to that destination cuz then hey, when you have another need you'll come back to Amazon instead of going to Google or whatever." But okay, so let me talk about the the team. They've been there, they've done that, they built the tech, they know what needs to be done. They probably have a road map here, but

I think you know, I got a lot of pitches for like middleware type ad network companies for like LLMs, right? Or sorry, I I always say I make I make the mistake of saying LLMs. When you are going to chat GPT, you're not interfacing with LLM, you're interfacing with an agent who will then go and query, you know, a model that's usually for the mega models that they're a mixture of experts and they do routing within them. So it's not, you know, it's an LLM too, but it's a family. When you interact with that agent, there's no inherent necessity of like AI in the ads layer. Right? Like you could imagine and we talked about this. Remember with the podcast we did with Mike? And I said, "Look, Trade Desk could fill that inventory. You could buy that inventory with Trade Desk. There's nothing inherently like AI native about filling an ad placement. You could do that with the existing pipes." Right? I mean, what you're building is like I mean, I you know, you people talk about like you know, the meta application of AI to to ads. And

2 years ago, I used to have a hill that I would die on, which was like a really pedantic position to take, which was I would say this was not AI. Like this is just plain vanilla ML. Like, you know, they're doing ML. They're doing ML ads optimization. Why are people calling it AI? Well, they're calling it AI cuz that's a great narrative for Wall Street. But also, I mean, I guess you can make the case, look, they're building these really big models. It sort of extends past what you consider to be like traditional vanilla ML. Whatever. Let's call it AI cuz that's more exciting. But nonetheless, when you're building, you know, just some tech to intermediate, you know, a buyer and a seller, you know, an advertiser and a publisher, that doesn't need to be any more sophisticated than what already exists, right? So, like if you look at the playbook, like the basis sort of like MVP of this, it's really you need an ads manager, you need an ability to upload creative, you need an ability to set, you know, a bunch of parameters on your ads. That's you could get that done with enough people in 6 months to where

you've got this fully, you know, sort of like a self-serve ads manager stood up and buying impressions against whatever you want, right? Whatever targeting you want, and whatever they make available, right? Where they could probably get more sophisticated is like parsing information out of the context of the chat. I think that might be like a mistake. I think if you go down that path, you run into the conflict, the sort of the treacherous conflict territory, which I think you probably want to avoid. I think it's better if you actually say, "Look, we're not going to let you target that because that might then create some tension. It might still this doubt in the user that their best interests are being served by the core, you know, agent output. But talk to me about that. Like how how could they handle targeting? I mean, you know, you could imagine if they just replicate Facebook, well, they just say, "Look, you know, we've got a CAPI, we've got a pixel, and that's what we use for targeting."
Totally. I I mean, I think that's going to be a big part of it. I also think that the kinds of product I think that they're going to build an interest graph because they're going to

know enough about the different ways that you think and interact and need to like like worry about things and blah blah like the amount of things that ChatGPT knows about me, I mean, it's not stored in memory. This is a separate issue of like how they store in memory versus how they'll take advantage of it in ads, but the amount of things that it knows about me is astonishing. It doesn't need the context of that chat to serve me relevant like it does not, okay? It knows that I'm a CEO of this company, blah blah blah blah like all this stuff. It can serve me plenty of products without caring one ounce about what my interaction is.
[laughter]
Like it is a non-issue. And I think that that's just what they're going to do. They're going to be like, "Hey, I bet you this guy is going to keep buying jeans, you know, because he's a like brown tech guy in San Francisco. That's all they wear. So [laughter] like So like let's get him some new jeans, you know? Or and and it's just going to be stuff like that. And I'm never asking ChatGPT

about jeans, right? And so there will be no conflict at all. And so it's like the beautiful thing about this is it knows so much that it can recommend useful products to me without worrying about the So like why bother with the conflict issue, right? Like that's actually the core thing is like there's no upside. The other way of saying is like, "Oh, it it knows so much about what's happening in the chat. It could serve you the perfect product." It's like no, it can serve me the perfect product regardless of what's happening in the chat. So, I The upside is like de minimis if it exists at all. And the downside is massive. And so, yeah, I I think they're like advertising principles are and and they released whatever some set of principles, and I think that they are like roughly guided, again, based on economic rationality of just like what's going to make a great experience here. And they probably have enough confidence that their models are going to learn enough about what products are good and not

good. Again, because they're not charging CPC, they're charging CPM. Guess what's going to happen in those first 6 months? They're going to check what people click on and not. They're not going to care about the click-through rate because they're maximizing signal. And so, their system is going to get so good at learning, oh, it looks like you clicked here, didn't click here, clicked here. And it's going to do this search in those first 6 months to understand where it's wrong about what products to show me. And then, after that, they'll switch the economic model. But right now, they don't want an economic model that messes up their ability to learn how to build the best possible system.
Yeah, and I think they don't want to be serving niche products, either, cuz that's just that creates a an incredible opportunity to misfire, right? I mean, you'd want the big brands coming in cuz like think about the least controversial ad you could see. It's for, you know, Colgate. Right? I mean, like it's not You're not going to You're not going to offend someone or creep them out, right?

With a a deodorant ad. But you would, you know, I don't know. You can imagine a number of ways you might, right?
Mobile game developers no longer [music] need to pay up to 30% in major app store fees. With Xsolla Web Shop, you can create a direct [music] storefront, cut fees down to as low as 5%, and keep players engaged with bundles, rewards, [music] and analytics. Start today at xsolla.com. That's x s o l or use the link [music] in the episode show notes.
Okay, let's say that I you know, what's today? Today's January 30th. I bring you back January 30th, 2027. And we're like, "Wow, what a blowout. That was even bigger than we expected. Oh my god. You know, Chat GPT did 10 billion in revenue last year. I can't believe it." Or whatever, some large number. Some seemingly improbable at the as of this moment number. They did some incredible number of ad revenue for the

year. Why? What did they do right? What did they get right to achieve that?
Yeah, I mean I think it's actually only the mundane things that get them to that blowout number. Meaning they have made it easy to onboard people and to enable them to put a budget in and to enable them to understand like not in the first 6 months, but in the back half 6 months to enable them to understand some amount of effectiveness of that spend. And the reason is this year the budget of every like I was on Tuesday I was at this conference and I was asking this like head of marketing of this well-known brand "What are your priorities for AI this year?" And he was like "Honestly, it's two things. And I'm going to take budget from other places to do it.

How much am I showing up in AI interactions? And like how what can I do to affect that? And testing and spending as much as I can on Chat GPT ads." He like said it straight up, okay? Straight up. And the answer for everyone else is the same. Okay, like he is not some special snowflake in his way of thinking. He understands where I mean like him and everyone else understands where the puck is going. And so they're going to everybody's budget to spend into this especially this year is going to be extremely large. Because they can't afford to not understand. And so then the only constraint is ChatGPT able to absorb that. But just like with Netflix, the problem became they were not able to absorb it. And so that will be the only problem will be ability to onboard and ability to

supply. And that will be the only problem. And those things are way harder than they sound. So if they can get those things right, we're going to wake up in a year. Yeah, I mean I'll say something potentially too optimistic. I think we could wake up in a year to a number closer to 20.
At the run rate at the run rate. Wow. Okay, run rate. Right. Okay.
Yeah, yeah, yeah. So so like so like the December spend is like 1.8.
Yeah, I mean I don't think that's unrealistic to be honest. I mean that that's that requires everything to go right. That requires exceptional execution, exceptional unrelenting execution over the course of the year, but I don't think that's not impossible, right? And I had written in my predictions for 2025, so I wrote them at the end of at the end of last year or at the end of the year before last year that OpenAI would launch ads in 2025 and it would hit a half a billion run rate by the end of the year, but I was you know, I was writing at the very end of the year, so I kind of thought well the launch would be, you know, middle of the year. So, you know, call that, I

don't know, a billion run rate and that was like I thought a kind of a lowball number. I I think you know, this and that was before, you know, they acquired Statsig, that was before, you know, Fiji Simo joined. So I think, you know, given that the density of talent now from Meta, I think they just know exactly what to do. But I think if I would ask this sort of opposite question, like we wake up and it's, you know, January 30th, 2027, like this was a dud. I think the issue would be lack of focus because they do seem like a little bit of an undisciplined organization. They throw a lot of the stuff they throw a lot of stuff at the wall. They wanted to see what's going to work. They've got kind of an endless appetite for investors for their equity. So they feel like they could just, you know, endlessly try stuff and at some point they'll come up with a second revenue stream that matches what ChatGPT is doing with the subscriptions. My sense is there maybe is a little bit more discipline being applied now, probably by the investors. That was probably I feel like my hypothesis is that they were the ones who kind of force bringing in a CEO of applications.

I feel like Sam Altman probably wouldn't want to surrender that control, you know, of his own volition. And but so I think if they can just stay focused, stay disciplined, and execute, and not be distracted by a million other things, yeah, they could reach that blow out number where, okay, wow, this was obvious, but it what wasn't obvious was the scale of the opportunity, which was far beyond, you know, what people expected.
Yeah, I mean, I think that if they can't focus on this, I I I just don't know what to say. I have all the respect in the world for that organization, right? I didn't build chat I mean, like we did not build OpenAI, right? So like all the respect in the world, but it's such an obvious net good and net benefit to them that to not have the right people focused on it would be astonishing. Yeah.
I mean, I think the scale of this is such that it could have a perceptible material impact on the economy, right? I mean, if you get this right and you inspire other people to do it and you prove the model

in a sort of scalable way, yeah, this this is a massively, you know, accretive and this is an accelerant for economic growth, right? In the same way that just digital advertising is generally, but if you get this right, this use case, and you create the blueprint for replicating it, I do think it's a trajectory change for the impact of digital advertising on GDP growth, right? Which I think is generally under appreciated. You know, cuz when people talk about the effect of advertising on on on the economy, they tend to look at like, you know, the percentage of advertising that sort of the percentage of GDP represented by advertising is like pretty stable over time. It falls within a kind of a narrow band, like 1.5 to 2.5%. But that ignores the fact that it's actually it caused GDP growth to accelerate, right? If you think about the acceleration effect of the internet broadly, but digital advertising is a big part of the internet economy, it is the internet economy, well, then that caused GDP growth to accelerate. That was part of that acceleration. So, you can't just look at the overall percentage, you have to look at the

acceleration, too. And this could be another one of those inflection points. So, I I'm rooting for them. I hope they get it right. I hope they're given the latitude internally to do this the right way.
Yeah, I would provoke a step further. Like, yes, and I think like people forget that Shopify Again, Shopify, exceptional organization. And I assure you this is exactly tied to OpenAI ads. What's What most people don't appreciate about Shopify is the strategic thought of how that organization executes is like 150 out of 100. This is like strategy of the highest caliber. And Toby Lutke decided, "Hey, I'm going to build a system that allows you to sell stuff online, but I'm going to like constrain it a lot. But guess what? You're going to be able to stand something up and put ads on Facebook and sell extremely fast. And it gave rise to a new type of

merchant. Okay? Like, that type of merchant didn't exist. The only type of merchant that existed was retailers. And so, he had the foresight to realize a new type of merchant can now exist because of the way that you can acquire people online. He is doing that again, or he wants Shopify to be that again for LLMs and for AI. And you can see that in like he is on podcast talking about it all the time, in the executive changes that are happening, in how the organization is being run. If you talk to the large retailers and the merchants who they are selling to versus like not really spending much time with, he I think he fully appreciates that commerce driven by consumer shift through AI is going to lead to it the next set of new merchants and next set of new businesses that get launched online. He wants Shopify to win that. Whether or not they win is not relevant to the point of OpenAI ads outcomes.

But OpenAI ads outcomes will deliver that level of like growth. And and just to like put numbers on it, that level of growth it basically the GMV that today powers Shopify. Like that is the right way to think about the impact. Not the amount that was spent on ads. It's like yes, you spent that amount of ads, but look at the amount of stuff that got bought through those ads.
Right. And the the I call it the money multiplier effect of of performance advertising. It it's that gets recycled. It gets recycled. That's why people are under underestimating the impact of Meta's investments here. Like it's okay, 3%. Who cares? 3% First of all, 3% on that top line, give me a break. It's a lot. But second, that compounds. I recover that in 90 days and I reinvest. I reinvest the the larger amount. So, this compounds over time. Like that's what people don't get. And I know, I'm I'm super excited. I think this is great. I'm really happy they did this. I appreciate that, you know, this was probably like a fraught decision internally. I think it was good that they did it now. It probably would would

better if they did it a year ago, but, you know, better late than never. And I appreciate you coming on on pretty pretty short notice. This was great. This is a great conversation.
Yeah, totally. Hopefully we can get some more people excited about OpenAI ads. So, slowly but surely.
The right people are excited about it. The hostile voices tend to get amplified the loudest and it's not always an easy position to take to embrace the advertising economy, but it's I think it's the right decision and I think this is a really exciting development. I think even given how much coverage it's got, I think it's under-appreciated the impact it's going to have.
Totally. Awesome, man. Thank you again for having me.
Rishab, always a pleasure. Thank you so much.