Marcus Johnson (00:05):

Hey, gang. It's Friday, May 1st. Grace, Nate, and listeners, welcome to Behind the Numbers, an EMARKETER podcast. I'm Marcus and joining me for today's conversation, we have two people. One of them is our technology and AI analyst living on the West Coast in California. It's Grace Harmon.

Grace Harmon (00:19):

Hey, Marcus. How are you?

Marcus Johnson (00:20):

Very good. How are we doing?

Grace Harmon (00:22):

Good.

Marcus Johnson (00:22):

Welcome to the show. We're also joined by our principal AI analyst living on the other coast some of the time. Right now, he's in New York. It's Nate Elliot.

Nate Elliott (00:30):

Happy mayday, Marcus.

Marcus Johnson (00:31):

Hello, fella. Thank you very much. Same to you. Today's fact. So who invented chess? I'm not a hundred percent sure, but chess.com, which is like the official place I hear, they've said that the game was born out of an Indian game, which sounds a lot like a yoga pose, but is... Is it the same word? Chaturanga. I think Chaturanga is what they say when you do it. Anyway, Chaturanga is the game, 600 AD in Northern India during the Gupta Dynasty. So that's where the game apparently came from. I was wondering, because I just watched Chess Mates on Netflix. Have you guys seen that? It's one of the Untold documentaries. It's about Magnus Carlsen, the Norwegian chess grandmaster, the best in the world.

Nate Elliott (01:27):

The erratic genius.

Marcus Johnson (01:29):

Has been, yeah, for over 10 years. And he plays against someone called Hans Niemann, who was accused of potentially cheating. Did he? Did he not? That's what the documentary's about. So definitely watch it. It's really good. [inaudible 00:01:42].

Nate Elliott (01:41):

So chess is less than 1,500 years old?

Marcus Johnson (01:45):

Yeah. However, there are theories, this is as far back as we can trace it confidently, but there are theories that it came from different military folks years before that. So who knows what the original, original, first ever version? But this is as far back as people agree it could have gone. Yeah. It's absolutely fascinating. I remember learning a long time ago, but this, I'll just quickly say, because I think it's so interesting, the number of moves possible in chess. So in 1950, Michigan-born mathematician Claude Shannon came up with the Shannon number, which is a number of possible unique chess games. It's a number one followed by 120 zeros, which would far exceed the estimated number of atoms in the observable universe. This is why I don't play.

Nate Elliott (02:34):

That's a lot.

Grace Harmon (02:34):

That sure is a lot.

Marcus Johnson (02:36):

Games. Yeah, shockingly. And it ramps up real quick. So after the first two moves, white and black, there are 400 possible board positions after the first two moves. After one more each, there are 71, 72,000 board positions. And after three moves, there are nine million different possible board variations.

Nate Elliott (02:55):

Wait, so this mathematician could have been curing diseases or solving global warming and he did this instead?

Marcus Johnson (03:01):

Mm-hmm. Time well spent.

Nate Elliott (03:03):

All right.

Grace Harmon (03:03):

I would say there's far fewer moves if you're playing me because I am terrible.

Marcus Johnson (03:09):

Same. Anyway, today's real topic. AI assistants that sell you things on the side. All right, guys. So ChatGPT maker OpenAI expects to make $2.5 billion in ad revenues this year. But by 2030, in four years time, or less than, they estimate that they will be making a hundred billion dollars from ads. Seems ambitious. Nate, I'll start with you. What do you make of this projection? OpenAI can make $100 billion in ad revenue by 2030. Fanciful or absolutely on the cards?

Nate Elliott (03:48):

I'm not sure I'm allowed to call it banana pants crazy, but it sounds banana pants crazy to me. They were bragging about their $100 million annual run rate a couple of weeks ago. Now they think they can scale that into 2.5 billion by the end of the year. Not run rate, but actual cash taken in the next, what, eight months, and then quadruple that and triple it and double it and basically grow from nothing to a hundred billion dollars, which took Google two decades to do. They think they can do it in four and a half years. That's not ambitious. That's wildly unrealistic from where I'm sitting.

Marcus Johnson (04:29):

It seems high, Grace. Yeah, this hundred million that Nate's mentioning is based off their ads pilot. In the six weeks after launching, they made that, or quoted this, cited this number. The projection, it assumes that OpenAI products will have close to three billion weekly users by 2030. They're currently at 900 million. Where do you land on this rather ambitious, it seems, goal?

Grace Harmon (04:54):

I think banana pants crazy is a good way to say it. I mean, it's a huge statement to be making so soon into their ads offering. In terms of the ARR announcement, it felt to a degree like something you would more expect to see from a seed stage startup talking about its own numbers. So to see that big of a projection from OpenAI was definitely a big surprise. Considering the tens of billions of dollars it needs to generate to write its balance sheet, I think it makes sense that they're having such a big target. They also just, so far, they've been offering advertisers very little insights into how things are performing. They don't have the infrastructure or the efficacy of ad platforms like Meta, like Google, which like Nate said, took a long time to get to these kinds of numbers. So I don't think the infrastructure is there, let alone high enough demand.

Marcus Johnson (05:45):

We have a AI search ad spending number. It's a lot lower than what they're expecting to get, is two billion this year for all of AI search ad spending in this country, two billion, five next year, and 13 the year after that. 2029, you're looking at $25 billion market.

Nate Elliott (06:04):

Yeah, I've been working with our forecast team on updating those numbers over the last few weeks. We should have those published pretty soon. They will not validate OpenAI's expectations on this one. Now, I will say the numbers you're quoting are for US and they're talking about global numbers, so there is a difference there. But if you think about all the things that have to go right for them to come anywhere near these numbers, I mean, number one, AI as a category needs to continue to grow exponentially, both in terms of the number of people using AI chatbots, the number of those people who are choosing ChatGPT over competing products and the volume of the conversations, the number of turns that people are having with those AI chatbots. All of those numbers need to continue to grow year on year on year. At the same time, the addressable audience within ChatGPT needs to continue to grow.

(07:03):

And so that's not just the number of people using it, but it's the percentage of people who are not paying for ChatGPT because they're only showing ads to free users and to users on the cheapest pricing tier. But also, those conversations need to be monetizable. They need to be able to put ads into those conversations. And there are a lot of things people are doing with ChatGPT, with chatbots in general that don't necessarily lend themselves to monetization. And then on top of that, the ads actually need to work incredibly well. I mean, they're asking for $60 CPMs during their trial. They since backed off that. I've heard $45 CPMs thrown around, and of course they're now offering a cost per action basis, which is where I think most folks think this will end up.

(07:50):

But those ads have to deliver incredible value to advertisers to keep the ECPMs high enough to drive anywhere near this kind of growth. At the same time, they need to build a technology stack that they don't have. They need to launch in 200 countries, whereas right now they're in one. I mean, the list goes on and on. Grace, what am I missing in terms of things that need to go perfectly for them to grow anywhere near this rate?

Grace Harmon (08:15):

I think one of the points you made that really sticks out to me is that you need to be thinking about what intent are behind those queries. If you're looking for healthcare advice, which a lot of people are going to ChatGPT for or chatbots for, there isn't really a place to insert a sponsored result there without really breaking the conversation and user trust. I guess one thing I think is that I would point out that rivals like Claude don't have ads. So for the people who are genuinely interested in chatbot ads, ChatGPT is the place to go. But I agree that you are expecting just exponential growth that continues, for ChatGPT to never slip, for there never to be a growing preference for other services, which we've seen start to happen a couple times this year already, let alone over the next three and a half years.

Marcus Johnson (09:00):

I want to talk a bit about what the ads are starting to look like or what could look like to get them to 20% of this goal, let alone the 400. The Economist writing, "For the ad industry, ad bots, as they're calling them, are a possible answer to the existential question of how advertising will work if users move away from conventional search engines. Although it is early days, the outline of a new kind of marketing is emerging."

(09:26):

It says, "Early evidence suggests OpenAI and Google have different approaches to how a chatbot should sell. According to Similarweb, nearly 99% of the ads in Google's AI mode appear in response to user's first query, rather like a conventional search engine. However, ChatGPT, by contrast, bides its time. Less than half of its ads come in the chatbot's first response and nearly a third come after the 10th turn in the conversation, like a shop assistant waiting for the customer to see the customer's intent and then to become clear and then to make their pitch. This is not search. This is a new category of advertising," argues Harel Amir of Similarweb. Nate, have OpenAI pointed their advertising efforts or strategy in the right direction?

Nate Elliott (10:07):

I don't know that they have, but I think more concerningly, I don't think that they know whether or not they have. One of the things that vexes me about this trial that isn't really a trial, right? They announced this earlier in 2026. Within a couple of weeks, they had set up a trial, and within a month or two of that, they seemed to be going full blast ahead, launching in different countries, offering different pricing formats, expanding not just the number of users who see ads every day, but the number of advertisers who are buying ads every day. It looks less and less like a trial and more like a full-blown launch of the ads product. But in amongst all of that, from what I've seen, the screenshots, the trackers, everyone else who's paying close attention to this, they're looking at effectively a single ad format put into a single placement within the conversations.

(10:59):

And the point of a trial really should be to experiment at least a little bit to try different formats, different shapes and sizes, different placements. I mean, these are all things that they should be thinking about and it doesn't seem like they are. So I understand the desire to say if they're putting the ads in the third turn or the fifth turn or the 10th turn, that that's fundamentally different from search advertising. But let's be honest, the ads that are showing up look pretty much the same as a traditional search ad. And if they're going to try to make a hundred billion dollars in the next three and a half years, they're going to be putting ads in every turn of conversation pretty soon.

Marcus Johnson (11:41):

There's a lot of ways this can go and a lot of different camps emerging in terms of what might be the right approach. We just talked about Google's AI mode showing you the ad straightaway after the first. We're talking with Nate about is it going to be after the third, the fifth, the 10th? Et cetera. There's also this conversation or this pull between performance and brand. The Economist also noting OpenAI and Google seem to be going after different kinds of advertisers. Google's direct offers allow companies to promote a discount code for a product the user is researching in AI mode. OpenAI, on the other hand, focus on brand-building ads designed to burnish a company's image rather than trigger a click. So far, 81% of its ads fall into that category according to Similarweb. It may reckon that users deep in a coding or editing session are unlikely to break off to pursue an offer.

(12:29):

So I mean, this behavior is so new using these models. And so it does seem as though, to what Nate's saying, they haven't quite figured out where ads are going to be most effective and what those ads are trying to even do.

Grace Harmon (12:41):

Yeah. I think to the latter part of what you said, I think that the big decision it's making is the focus awareness or is the focus conversion. I think that a lot of its advertising direction just still feels really cautious and really underdeveloped. Some of that might be a little bit intentional. I think like Nate said, they're testing the waters, how diverse those efforts are to test things. I think that it's falling short, but they still kind of need to figure out how sensitive the chat interface is to user trust. You need to figure out how to evaluate user response and behavior just to avoid eroding that relationship, which again, we're still hoping, or they're still hoping that chatbot adoption really spikes. And if you are interrupting everything with an attempt to send you off to a website, an attempt to shove a discount code in your face, I think that's really going to become an abrasive user experience. So I think awareness seems to be its biggest focus rather than conversion. I think that that's rational.

Marcus Johnson (13:34):

Nate, what do you think about this idea that they found? So at the beginning of April, OpenAI partnering with ad tech firm Smartly with the aim of eventually launching interactive chat-based ad formats. Our senior briefings editor Danny Constantinovich explaining that in this format, clicking on an ad opens a separate window where users can chat with a shopping agent to tailor product recommendations. Thoughts on this chat-based ad format partnership?

Nate Elliott (13:58):

I mean, it's great that they're testing it. This is exactly the kind of testing I was asking for a moment ago. I think they need to do a lot more of it, and I think they need to do a lot more of it in the core product that they're telling us is going to generate hundreds of billions of dollars of revenue over the next five years. We don't know if what they're talking about here is going to work. As Grace was saying a moment ago, we don't know the answers to lots of questions. We don't know if branding makes more sense than direct response. We don't know which position or size or format of ad works.

(14:31):

We don't know what kind of targeting is going to be most effective or which pricing model's going to be most effective. We don't know what kind of impact this is going to have on users' consumption of ChatGPT. All of these are really important questions. All of them bear out research and investigation and trials, and this is where they should be two or three months into their advertising trial, not racing full speed ahead to try to make it to a hundred billion dollars.

Grace Harmon (15:00):

I think I'd also still come back to one of the biggest risks here is user trust. I think that the chat-based ad format partnership makes sense. It's leaning more into utility over interruption, and that's really where the risk is lying here that ads, if they're looking too much like answers, that's going to blur trust boundaries really quickly. Having that pop-up window where it's very clear to the user what's happening, what the AI is doing, I think that that is going to help for some people, make suggestions feel more helpful than they're subtly being steered towards something.

Nate Elliott (15:30):

I got to say though, I actually think users will have more tolerance for things that look like answers than a lot of us are thinking they will. I mean, we've had this conversation in all the other new digital media formats. We had this conversation about search, we had this conversation about social media. And the concern in all of these channels was if you make the ads look too much like the content, that will confuse people, it will upset them, they'll stop using that channel. And search found a way around that challenge. Facebook and other social platforms found a way around that challenge. I think at the end of the day, stuff that provides something like an answer or a part of an answer to a user prompt or query is probably going to be one of the formats that performs best. I mean, this is the kind of stuff that ChatGPT should be testing right now.

(16:22):

And I think that if you demarcate it carefully enough, if you put it on a different colored background or try some of the other tricks that helped Google and Facebook and other platforms get around user concerns about ads that look like content, I think it might be threading the needle, but I think there's room to make that happen as an ad format. And I think that that format would be likely to deliver a lot of value for the advertisers.

Grace Harmon (16:50):

When you're getting to ads that look like answers though, and I'm not saying that search doesn't do that, I think that there comes a line of whether users know it or not, manipulation, there's a lot more vulnerability for users who are using chatbots just in terms of the personal information they hand over and whether they see it as a person or not a person for a lot of people. And it seems like the amount of context they have, things get a lot more targeted. And again, I don't like using the word manipulative, but I guess I can't think of a better one.

Nate Elliott (17:18):

It's an open question, I think is the key thing here. So I may be right that there's a window to make this happen. Grace may be right that it's just that these know too much about people and it's too personal to make it happen. Either way, they should be testing this. They should be finding out the answer to the question and they don't seem to have any interest in doing that.

Marcus Johnson (17:36):

They're hoping that this test with Smartly is going to be one of those answers, because Danny was noting in the piece that a new chatbot conversation, a separate window could help ChatGPT avoid the pitfall, he says, of cluttering its conversations with ad space, maintaining user trust. You've mentioned trust a few times, Grace, and it is so important. There's this chart here we've got up on the screen, data from Ipsos. 63% of consumers agreeing that ads decrease trust in AI outputs. 24% of people say it has the opposite effect. However, the other side of that coin, there's another study from Forrester suggesting ads won't change consumers' AI habits, finding that 83% of online adults would continue to use free tiers of answer engines even if ads were introduced. Only 6% would switch to a paid ad-free tier.

(18:24):

So they've got to try and figure out if ads are going to work, what they look like, where they look like, how they look like, all of the things, but they're doing it because they need money. OpenAI, they have a lot of money.

Nate Elliott (18:37):

Lots and lots and lots of money.

Grace Harmon (18:37):

Lots of money.

Marcus Johnson (18:39):

Yeah. Under significant financial pressure already, a company generating about 13 billion in revenues last year, according to CNBC. Internal documents seen by [inaudible 00:18:49] show they expect a $14 billion, 1-4 billion-dollar loss this year. And then Danny mentioning further straining its wallet, plans to spend as much as 200 billion through the rest of the decade on training and running AI models. So they need money. They're trying to use ads to get that money. But if not, the alternative could be subscription. So Grace, in February, you and one of our analysts, Marisa Jones, co-authored a piece explaining that AI-powered answer engine Perplexity is fully stepping away from in-chat ads over concerns that ads risk eroding user confidence in AI-generated answers we've been talking about. And the company is focusing on monetization through subscriptions and enterprise sales, prioritizing paying users over growth metrics like total queries answered. So Grace, is there a world where chatbots remain ad-free and rely solely or at least in large part on subscription revenue?

Grace Harmon (19:36):

Yeah. I mean, you said it's serving millions of free AI users is expensive and ads help cover that cost. That is a reality. I think that in terms of staying ad-free, that works in only some cases. They work to stay ad-free if people are willing to pay for them for the most part, just in terms of company's ability to, again, have an even balance sheet. I think that works best for work tools, so writing, coding, research, and then it works really well for power users, businesses, enterprise customers. And that's part of why it's working for Claude being ad-free because it pulls in so much of its revenues from enterprises. But for mass use, I think it's a lot harder. Obviously, there's the trust factor that could slow down ad adoption. And I think that that risk is what might keep ads constrained or more clearly separated for a lot of platforms. And I think OpenAI is just still kind of walking that line.

Nate Elliott (20:26):

Grace, I think that the word you used there was for mass use or mass consumption, and that's the most important thing from my perspective too. I mean, find me a channel where the largest players don't have ads. There just aren't any. And it doesn't mean that there isn't room, Marcus, to your question, for some players to remain ad-free and make all their money from subscriptions. But by definition, in all the other media channels and platforms that we look at, those are niche players. And the biggest players may take subscription revenue, but they're absolutely going to take advertising revenue as well.

Grace Harmon (21:03):

And I have to ask, how well is Perplexity doing compared to OpenAI or Claude? Not as well.

Nate Elliott (21:09):

But that's why it makes sense for them to take a different tack, right?

Grace Harmon (21:12):

Totally.

Nate Elliott (21:13):

You're getting walloped by ChatGPT. You have nothing to lose. They have tens of billions of capitalization to lose, but-

Marcus Johnson (21:22):

Why not?

Nate Elliott (21:23):

... yeah, you have to try something.

Marcus Johnson (21:25):

Why not try something else?

Nate Elliott (21:26):

You have to try something different. And so go for it.

Marcus Johnson (21:28):

Marissa agreeing, saying, "Unless AI firms can generate meaningful revenues through other channels, advertising may ultimately become unavoidable."

Nate Elliott (21:36):

I mean, Marcus, the most important question for me is who were the 24% of people in that survey who said that seeing ads makes them trust AI more? That's a really confusing number for me. But I mean-

Grace Harmon (21:49):

Maybe the question was posed weirdly, I would say.

Nate Elliott (21:51):

The two data points that you put up are, for me, the entire story. People say that they don't want ads because they don't want ads, and yet they will accept ads. And again, this is the story that is always written and always told in any new technology platform, any new category of media. People don't want to see ads, but when ads show up and the question becomes you look at ads or you pay money out of your own wallet or you look at ads or you lose access to the tool, the answer is they will accept the ads.

Marcus Johnson (22:24):

Yep, a story as old as time. So we'll see where it all goes. That's all we've got time for for this episode. Thank you so much to my guests for now. Thank you first, Grace.

Grace Harmon (22:34):

Thank you, Marcus.

Marcus Johnson (22:35):

And thank you, of course, to Nate.

Nate Elliott (22:36):

Pleasure to be here.

Marcus Johnson (22:37):

Yes, indeed. Thank you to the whole production crew. Danny and Lance on this one, thank you guys for helping us out. Thank you to everyone for listening in to Behind the Numbers, an EMARKETER podcast. Subscribe, follow, leave a rating and review if you can. We're back on Monday, of course. Until then, I hope you have the happiest of weekends.