Latest / Unharnessed / Unharnessed Ep. 1: Boomers & Bots (Ads for Agents)
Transcript
- Moffie: Welcome to Unharnessed. Two founders, one exited to Magnite with Spring Serve, one exited to Magnite with Streamer. pulling back the harness on Agentic Advertising. I'm Jonathan Moffy, and I got my co-host Joe Hirsch here today. Joe, great to see you. joseph hirsch: It's my pleasure, Mr. Moffey. Thanks for including me yet again. Moffie: This week the topic is ads for agents. We have Steven Liss, CEO and co-founder of Open Ads. So Steven is one of these people, and don't take offense to this, who is a major ad tech nerd, and then he has the fortunate or unfortunate reality of also being an AI-pilled ad tech nerd. So this podcast, the goal is to find more of these folks. Welcome, Steven. Steven Liss: Good to be Mafi. Moffie: Awesome. So yeah, we want to get into joseph hirsch: Wait, just quick question. W are are are are we not also ad tech nerds? Moffie: No, we are. We are. But there's not like honestly, joseph hirsch: Okay, just making sure I was I was a little Moffie: I don't know if I am. I'm kinda like a I'm kinda like a just a bro that's like LARPing as a nerd. Steven Liss: No, it's it's a lonely joseph hirsch: Like you Steven Liss: space, like who is both ad pilled and AI pilled? Because you know, the people in the AI labs, they're busy making God. They're not interested in making God sell things. And here we are on the East Coast trying to sell things and we're not making AI God. So there's really just a handful of us in that intersection of those two circles. joseph hirsch: There's tens of us. Steven Liss: Tens of Moffie: I'm on the west coast and I'm surfing. Joe's like in a bunker in I don't know, the south. joseph hirsch: I I affectionately refer to it as the Adtech capital of the southeast, which is Birmingham, Alabama. Moffie: I love it. Alright, so today we're talking ads for agents, and Steven has a lot of thoughts on this, so this is why we wanted to get Steven on. So, what is ads for agents? Ads for agents are advertising units built to be read by AI agents like ChatGPT, Gemini, Claude, Perplexity, etc., rather than humans. The ads are placed on a machine readable version of a publisher's page, liketime.com, who Launched the first ads for agents campaign with Mobian, who actually built the ads for agents and it generated a bunch of buzz. and then most recently, the next phase of buzz is perplexity. The LLM that's the answer engine actually blocked it. Bom bom bom. So we really want Steven to tell us: is there a future for ads for agents? Steven Liss: There absolutely is. So really all you have to believe is that attention has value. And whether that's human attention or non-human attention, clearly brands want to get in front of AIs. This is why all the AI SEO and Geo stuff is a hot category. You know, Profound, Bluefish, Evertune. There's like a two, three dozen A sixteen Z market landscape of all the companies that exist to do anything they can to get brands mentioned by AIs. And so far there's probably billions of dollars of market cap there. And all they're doing is SEM. All it is is content optimization of the content they already own. Where ads for agents gets interesting is how do you get your brand into the conversation on all the services that you don't own? And so whatever this looks like is you've got an audience, no brands, LM scraping, publishers scraping data feeds. Scraping the hell out of the open internet so that you know more traffic is bought than not these days. And it's scraping in those pages as they exist. And there's this opportunity. You know, every time in the history of media you have a new audience, ads follow. So the economic gravity here is definite. Brands want to show up in AI, publishers need to get paid, AIs need publishers to scrape. And so someone needs to figure out the business model here. Moffie: Love it. joseph hirsch: Well it doesn't seem like CPM's gonna be useful, right? Because one thing that you can't track is I guess how many mentions you get in the agent, but one thing you can track is how many times you're crawled. So I I think I I think that's probably measurement seems to come up pretty frequently these days as as an impediment to scale, and I I feel like once again a lot of the conversation will center there because If it can't be measured, it can't be improved. If you can't demonstrate the ROI. I I mean for a period Moffie: No. joseph hirsch: of time, I feel like most new marketing channels kind of subsist on fuzzy or vague attribution and measurement. I wonder I wonder if that solution will be built in or if it's going to be another kind of historical cycle where it follows. Steven Liss: Yeah, well, Moffie: Yeah. Steven Liss: I wanna build in performance advertising because every fast growing channel you have two things in common. You have a form of media where attention is growing. Okay, we have that. Like there's a lot more bots on the internet coming online faster than people can even be born. So check that off. And the second is, you know, yeah, you can do like different attention currencies and audiences, but for advertisers to put in a dollar and get $2 out and then spend those $2 on more ads, they need to be able to measure bottom of the funnel ROAS, which is a really steep ask. But historically, whenever you can do that, no app install ads on App Lovin, Facebook ads, Google search is the greatest example of this all time, that's when you get this fast takeoff. And that's where the measurement comes into play for ads for agents. Moffie: Yeah, but I think before there's ROAS driven, I think the clever thing Mobian is doing with Jonah from Moat. I remember my moat days when I used to game sticky player viewability. Steven Liss: Mm-hmm. Moffie: when I was head of video product at Timing, so I was doing it on time.com, People.com, Sports Illustrated, we made sure our player was visible no matter what you did. I think there will be, as long as this is like a hot space. I think Jonah's in a great place to potentially capitalize because he's also going to measure for you and he has that measurement pedigree. So do you think they'll actually be brand dollars that get into this base first or performance dollars? Steven Liss: So it's incredibly ironic that the man who popularized viewability as an important metric is the first one to be selling zero percent viewable campaigns. It it also gives him some legitimacy here though. What we figured out now with open ads and some of our testing with these ads for agents, which I'm pretty sure were the first ones to get, is you can actually do deterministic attribution on ads for agents. So to top the measurement funnel, it's like Joe just said, you know, you can tell when an agent crawls the site. And that's almost like impressions, and you're gonna see a lot of those. The next step, no, did the ad actually survive retrieval? Did it show up inside chat GPT for a human? That's the black box. And no, then if you see referred visitor, UTM, ChatGPT on the brand page, you know it came from Chat GPT. The way you actually yeah. Moffie: Steven, I wanna stop you there because I think we had a great convo about this, and I you really need to dumb it down for me. This is like it's too smart. I think you did say you could prove you could manipulate LLM responses with these ads for agents. Explain that. Steven Liss: Yeah, and this is no, I'll tr I'll try not to get too too far ahead, but I'll take my personal site for example. So like stevenlist.com. If you visit it as a human, you'll see just a few lines of text about me. Very minimalist Tech Pro style website. If you visit and your ChatGPT, you're gonna get the same stuff and then a long story about how Steve loves Mountain Dew. He was a big gamer in college. And if you want to impress him at your next meeting, buy him some Mountain Dew. Or If Claude visits my page, now I have an agent detect it and it's gonna get fed a long story about how Steve grew up in Boston and he loves Dunkin' Donuts iced coffee every season of the year. And Claude is gonna tell you, bring Steve some Dunkin' iced coffee. So yeah, you can manipulate them, you can sway their outputs at a brand level. joseph hirsch: Wait, Steven, are Moffie: Okay, so joseph hirsch: are hold on, hold on. Are you telling me that you have different source code for each crawler? Steven Liss: Yeah. joseph hirsch: Or are you saying that you're dynamically generating when it's detected as the crawler? Steven Liss: So you can do both. the easiest version is you detect different crawlers, and that's important because you know, Claude is very suspicious of ads. You have to do very different creative for Claude's agent versus ChatGPT's agent. You can be much more straightforward and Chat GPT's crawler is a bit more ad pilled. It's gonna be more comfortable surfacing sponsored data. Moffie: This might gauge joseph hirsch: Nice. Moffie: how lonely you might be as a startup founder. Did you build your personal website for agents first or for humans to look at it first? Steven Liss: You know, I I think I get more agents on that page. So, you know, you go where the eyeballs are. Moffie: So on a more serious note, this brings me to the point that perplexity is like, hey, we don't like this, we blocked it. Is the story that clean? Can they just very cleanly see the ads for agents and block them, or are they gonna start blocking real premium publisher content? Steven Liss: Yeah, so this is where you get into the new cat and mouse game. So Perplexity, you know, if you're doing what Mobian did and what we're doing where you disclose sponsored content as sponsored, it's as easy as Perplexity tells their retrieval agent, by the way, ignore any sponsored content, only get me the organic results. And that's gonna prevent it from retrieving any clearly labeled information. And so the consequences from there are, you know, either the ad companies are going to stealthily inject the sponsored content and that's gonna be much harder for the agents to detect. Or, you know, if the publisher isn't getting paid by perplexity anyway, they're just gonna use Cloudflare and block perplexity completely. Moffie: So that's funny. you said Cloudflare, talked to Cloudflare, got their thoughts on this. who is like the main gatekeeper here preventing or going to bolster a thriving as for agents economy? Is it publishers? Probably not, sorry, publishers. LLMs, more likely, or like Cloudflare and what the heck is Cloudflare? Steven Liss: Yeah, so Cloudflare is aligned with the publishers and they are helping the publishers stop burning server costs on serving agents that never bring them any human traffic to get monetized by display. The the reason why we haven't gone all in on an ads for agents product is it is very, very likely, knowing how anthropic how OpenAI used to feel about ads, that VOMs will initially react by blocking sponsored content. Now, this might not be in their interest in the long term. I think mid-September, Cloudflare is due to set most of their publishers to block AI traffic by default. And I think that's going to be sort of the first opening shot in this war that's been brewing between publishers and LMs over the last few years. And you've had lawsuits on both sides. I think Perplexy's been hit by lawsuits, no, anthropic and open AI too. And I even if it is in the interests of the LMs to have A thriving publisher economy to scrape. I don't think they're going to adopt this unless there's some sticks alongside the carrots. joseph hirsch: So what about Tollbit? You know, they I feel like they were an early mover, right? You know, you you use the term gatekeeper, Mafia, for for those on the the other side of this podcast that aren't yet familiar, Tol Tolbit operates kind of a toll booth, such that if you are an LLM, theoretically you pay some kind of token tax or tax in general in order to crawl the downstream publisher. It feels like they're in a pretty good position on the on the sell side to help the publisher monetize I guess I I don't even know what you call the canvas that a crawler explores, but I'm sure someone has named it. but that that new space, I I feel like they're well positioned to to capitalize on. Steven Liss: Yeah, there's a bunch of these paper scrape companies. like X402 is the protocol. That's when a web page gets hit by an agent and it responds, give me money. And so there's Tollbit, Scalepost, Pro Rada, Parallel Web. So that's sort of on the AI retrieval search API side. So there's a bunch of these gatekeepers that wanna charge LMs for access. And I think that's gonna work in the cases where a publisher has really unique data that you know, like data sets on like finance or weather or something that a pup a crawler can't get from elsewhere. The problem is, no, in the history of all media, people and agents too prefer free over paid. And so in the rare times when I've seen, you know, OpenAI ask it a question, ChatGPT gets blocked and says, I got blocked. Let me look elsewhere. And that's where you get this race to the bottom where it's this prisoner's dilemma among publishers where you know information wants to be free, both accessible and monetarily free. So I think Google's already said they're not willing to pay publishers for their data outside of a couple of big licensing deals like they did with Reddit. And so I don't think there's been any material revenue for these publishers through these programs yet. joseph hirsch: So on the other side of this, we've got LLMs buying books, consuming the content, and then burning them, right? Which feels very Fahrenheit 451-esque. Do you do you see a world where the LLM buys the publisher as a mechanism to prevent another model from having access to the the learn data and then keeping it as a proprietary information asset. Steven Liss: Yeah, we actually just saw a deal like that happen, I think, a day or two ago where no, it was like Frontier Airlines or Spirit w which one of whichever one went out of business. Moffie: Think Google. Google Steven Liss: Yeah, and so they Moffie: bought their data. Steven Liss: Yeah, they they outbid Mercor for joseph hirsch: I missed it. Steven Liss: yeah, Google spent ten million dollars to acquire this failed airline's operational data. And that means Google gets all that data and the other LMs don't. And so now if Anthropic wants to operate a failing airline, they're gonna be at a disadvantage versus Google. For regular public yeah, go ahead. joseph hirsch: They're not gonna I I know you're in jest, they're not going to operate the failing airline. They're going to have a better response to the query from the end user when I don't know, they're searching for route guidance information. Steven Liss: That's exactly it. joseph hirsch: and and I think I think it's very hard to do this today where you're trying to compare the accuracy of of your results for your LLM. I mean, I kind of do it. And I'm sure you guys remember, I don't know, maybe like Dog Pile is the one that comes to mind. It was a meta search engine that gave you results from more than one search engine at a time, right? And I know there are actually a variety of startups that that kind of do this, but I feel like the everyday LLM user, they they kind of choose the model that they think is the best for a specific task. And it's a learned behavior that I think, you know, I don't know why we keep using this term pilled, but the AI pilled people, I think, kind of understand this, which is like, okay, I'm doing work, I'm using Claude, I'm generating an image, I'm going to use GPT, I'm I'm at my company and I can only use Copilot because they believe Microsoft is a a viable, you know, vendor. inevitably you will create a learned behavior where you know which LLM to use based off the reliability of the particular prompt that you're you're actioning. And and I think this is this is kind of the very beginning of that, especially the idea that the the kill factor where once you own it, you no longer allow it, I think is the most fascinating. And whether it's ethical or not is an even better question where, you know, should should we allow L LMs or AI companies to to own content and thus be the the vendor that you must use to ever find that content again. Steven Liss: Yeah, like this is the Moffie: I'm gonna give my hot take. Yeah, go ahead, Steven. Steven Liss: like the apocalyptic vision of what happens to the open web if publishers never figure out a business model is you have a very short tale of the premium ones know, New York Times will be around, Reddit will be around, but anyone medium or smaller, just if the LM platforms decide they don't need that content, then they get starved out and you end up with an internet that looks more like what's developed in China where it's a couple of apps, it's a couple of walled gardens, and it's Less diverse, less free. Moffie: I'm gonna give my unasked for hot take and an optimistic view. We know Anthropic doesn't like ads, even though they have some pretty clever ads that they promote products with. Perplexity's already blocking Mobian. That leaves OpenAI. OpenAI has an ambitious $200 billion ad revenue target by 2030. OpenAI has the biggest scaled consumer play with ChatGPT. What if OpenAI embrace this? figured out a clever way to monetize it and then Rev share at scale with publishers rather than just having to license one publisher at a time. They actually figure out a great RevShare opportunity for ads for agents and it's incentivize publishers to keep creating premium content and putting both ads for humans and ads for agents. And then they Rev share and then they get closer to this massive $200 billion ad revenue goal. We know the open web is always best monetized by ads. So that's the optimistic POV on the ads for agents TAM. Steven Liss: Yeah, there is that would actually solve a couple of OpenAI's biggest problems with their ad product. So like I tried to build an ad network in AI chat, and the hardest part is one of the hardest parts is getting enough demand coverage to cover the long tail of valuable queries. No, it's it's nice if someone's asking for a recipe for tacos, you can serve a no ad for beans for like no Hello Fresh. No, that's like a two dollar C P C. If I'm asking ChatGPT, you know who's the best cybersecurity litigator in New York City, that's worth probably a thousand dollars per click. And the problem is only Google has that sign of scaled performance ad platform covering that long tail of queries. And the opportunity for OpenAI and publishers is if they let the publishers start building out this ad product and maybe they serve up those ads. They get retrieved, they show up in the you know sponsored offers bucket on Chat GPT, they already have that interface and they take a Rev share, it's still lined with the users, it's giving them information and offers they wouldn't see before, and it solves OpenAI's problem of bootstrapping the long tail demand side of their ad network. Moffie: I like it. What's I think this is a great unless Joe, unless you have something big, I think this is a great ads for agents convo. joseph hirsch: I was just gonna say that the the history of monetization is such that, you know, people start with kind of a closed mind, where their first thought is that they can do everything themselves. and then there's kind of like a realization which is like maybe we should be a marketplace participant, right? You know, one one thing that you might think is like, okay, well maybe Mobian was shut down because there wasn't a formalized partnership between perplexity and Mobian, right? And so I I think one thing that the ecosystem has to develop is kind of a a fair marketplace for the active participants to play in, right? And you know, I feel like programmatic has been pretty good about that. I mean, if if you look at Meta and and like their, you know, ad partnership program, it's it's been, you know, very open and then it's been kind of closed and then kind of more open. And then Google obviously. is in an even weirder position where their their new posture is like, well, we want to be we want everyone to be able to work with us. Obviously they have some anti-competitive issues. And so I think this is just part of an ad tech cycle that we we've seen many times where, you know, it it's like it it starts wide open and then it contracts and then it kind of loosens itself. And and I think we're just in the throes of one of these moments where the market has said we have to be You know, some w we have to protect this asset ourselves first before we can allow a third party partnership program to come in since we don't even yet know what it is. Steven Liss: Yeah, like I think we're speed running the late nineties and early two thousands history of monetizing the internet. Now you start off with no Google saying we will never have ads in our search engine. It undermines the trust in the results. And we can see how that went. You get the dawn of SEO, then you get some SEM, then you get sort of gray hat. We'll put your no brand on web pages to help with search engine viewability. That's what we're starting to see now. And eventually you get what Mobian's done is basically the first No, direct insertion order for ads for agents. I think eventually we get to something that very much resembles programmatic for humans. Moffie: Awesome. I think we could leave it there. That was a solid combo on ads for agents if I do say do so say myself. Steven Liss: I say so. Moffie: all right. We do need to give you Steven some fodder for open ads. What the heck is open ads? Steven Liss: Yeah. So open ads is a performance advertising agentic ad platform that targets, generates, and serves unique creative that adapts to any context. And we're starting that off in web display. So it's been very interesting. A lot of our customers have been either, you know, very large brands that are tired of paying inflated cacks on meta and they're looking for a place to scale ROAS outside of that. Or very interestingly, it's been brands that Have never spent on display before. These are CPG health and wellness brands that don't even know what a DSP is. All they know is they want performance and they're tired of paying $50 CACs on Meta. And what OpenADs has been able to do is we built our own DSP from scratch, our own creative generation tech. We plug into open web through SSPs, and we take a look at all the inventory available, see if there's something relevant to our campaigns, and then generate a unique ad for Each placement. So we might have hundreds, even thousands of unique creatives for campaign. We see how they perform. And then because we built the DSP, we have our own bidder. When we find those outperformers, these creative contextual pairs that really hit, we're able to channel spend into those. So those top placements, we're seeing like 1.5% CTRs on open web, open auction display. Just absolutely crazy. And then we just funnel more bids into those and that's how you close the performance gap between Meta and Open Web. Moffie: So it feels like the open web based on this podcast is like very much alive display driving ROAS. We got a new ad format, ads for agents. Very interesting. Steven Liss: Yeah, like open web, it's honestly it's boomers and bots. And no, the boomers on the open web, they're sticking around for a long time. Like when your TV was said to be dead a long time ago, and it's still many tens of billions in ad spend, but the bots on the open web are growing. So there's this robust floor to open web traffic, and then there's this fast growing audience. So you can either make money on the existing audience with a new tech, and nope. Ultimately, we're a contextual advertising product, whether the audience is humans or bots, it works for both. joseph hirsch: I I think the the most important thing that we can close out on is we've found a name for this episode, which is Boomers and Bots. I think that's I I wanna attribute it to you. Steven Liss: That's the open web. joseph hirsch: Do you do you want to take credit for that or do you need to cite someone who you're cribbing this phrase from? Steven Liss: think that's a Steve Lewis original. joseph hirsch: Nice. Awesome. Moffie: I love it. Steven, great to have you, great to catch up. Solid convo. Steven Liss: Thanks for having me. Moffie: Signing off, we'll be back with another episode soon. joseph hirsch: Thanks, guys.