← Marketecture: Get Smart. Fast.
Marketecture: Get Smart. Fast.
Episode 136: Will Jonathan Roberts become People’s Sexiest Man in AI
01:00:32
Loading summary
Transcript171 lines
- [00:01]
A
This episode is brought to you by Sovereign, a software and data company powering commerce media campaigns for the world's largest brands. With real time purchase intent. Data collected from over $3 billion in annual retail spend, Sovereign's commerce media deals are helping reach consumers at the time of purchase and deliver better outcomes for advertisers. To learn more, visit sovereign.com that's s o v r n.com to learn more. This podcast is brought to you by Incremental. You ever run a holdout test, wait between two and five weeks for the results and then realize you just wasted a big chunk of budget? By the time the data comes in, the campaign's over, the money's gone, and the insight's basically a post mortem. That's why I like what Incremental is doing. They give you always on incrementality measurement. No experiments, no holdouts, no delays. You get answers while the campaign is live. So you can actually make changes, not just analyze the wreckage. With Incremental, you don't just learn what worked, you do something about it. Check them out at the Future of Measurement. Welcome to the Market podcast. I'm Ari Paparo. I'm here with Eric Franchi and, and we have a great guest today, Jonathan Roberts, who's the Chief Innovation officer of People, which is the new name for Dot dash Meredith And Eric, you saw him, I imagine, at Architecture Live last spring. He's a really insightful, great speaker, don't you think?
- [01:41]
B
Yeah, absolutely. So he was at Market Live. We had him on the pod actually last year as well. He's a fantastic guest. People is extremely like Innovation forward on all things and particularly AI. So I think it's going to be a great conversation and there's so much to talk about with respect to AI with the news this week that, you know, as usual we're going to dive into it, but it's an interesting time.
- [02:03]
A
And speaking of Architecture Live, we're excited to start announcing some things that are happening there. October 27th is the event. The early bird ends on Labor Day and we're already, you know, selling tickets very rapidly. We're definitely going to sell out this thing. But Eric, tell us about your involvement.
- [02:21]
B
Yes, so I am going to be co hosting and co judging, which is going to be new this time, the Market Share Live Startup Showcase with my friend and colleague Sanya Partalo of S4S Ventures. So the first thing is startups, if you want to pitch us, if you want to pitch to a room full of some of the most influential people in ad tech and marketing and media. Just submit. So it's super easy to submit. Look for posts that I made this week on X and LinkedIn. Call for submissions. It's a simple Google form. Send us your stuff. And we're going to be selecting five startups to present. These startups will be series A or earlier. They'll be focused on AI or performance based in advertising and marketing because that's the theme of the whole conference outcomes and the CEO needs to be there to present in person. So submit. We're going to pick five startups. They're going to pitch for four minutes. We're going to have Sonya and I judging. We're going to have a People's Choice award. It's going to be so much fun. And you will get eternal glory if you are on stage, let alone the winner because you're one of the top five.
- [03:24]
A
Yeah, exciting. Well, as a winner, you will get a trophy. You will not get funding that is not guaranteed a fight. You'll just get a nice little trophy. And so there's one other thing which you may notice if you're watching this on YouTube, which is you see our faces. So we're starting to record video. Hi, everybody. If you're on Spotify or Apple, you don't know what we're talking about. You just hear our wonderful voices. But if you want to see our faces and Jonathan Roberts face, when we get to the interview, we, we will be on YouTube with video. So give us feedback if you think that's a good idea. Bad idea if you want us to change our haircuts, whatever it is. All right, let's dive.
- [04:03]
B
Asking for this for so long. I'm so excited everybody gets to see you.
- [04:08]
A
Yes, it's definitely what people have been clamoring for. Not enough of my face. All right, I'm going to dive in. I'm doing the interview with Jonathan Roberts and then Eric will come back and we will do the refresh. News of the week. See you soon. All right, I'm here with Jonathan Roberts, the Chief Innovation officer of People. So people, huh? So celebrities that are just like us. Is that your beat now, Jonathan?
- [04:34]
C
Well, it's People Inc. So parent company, two people. A nice Inc. Callback to the Time Inc. Days. It's nice for this moment in time, particularly because there's a kind of unofficial tagline, buy people for people. It's a nods to the world we're heading into. We strongly believe in the future future of human created journalism. So this gives a good, well known, trusted brand to stand behind for it.
- [04:59]
A
It is. So for those of you who don't know what we're talking about, Jonathan, last time he was on the pod, he was representing Meredith ddm and they've changed their name. So what's the quick on People Inc. Or Meredith? What are your well known properties?
- [05:18]
C
So people obviously, but some huge properties. Better Homes and Garden, Travel and Leisure, Investopedia, Food and Wine allrecipes. Real simple, right? These have been around for, in some cases, over 100 years.
- [05:32]
A
Yeah. Great reads all. So last time you were here, you were talking a lot about some of the investment R and D that your company had done for targeting and not being cookie based. And you also spoke at Architecture Live and I thought that was really interesting. It's one of our better sessions. But today we want to talk about AI because there's so much AI stuff going on and you always have an interesting insight about this sort of stuff. How much of your time is being spent on AI stuff as the Chief Innovation Officer?
- [06:01]
C
Probably most of it at this point. And AI stuff is going to become part of everything. And I think for my life, it's already become part of everything. So I got to be a little out in front.
- [06:11]
A
Are you using it personally? Are you like looking up recipes and figuring out where to vacation and stuff?
- [06:17]
C
I think I did the thing everyone did, which is, you know, GPT5 comes out and you do a one prompt. Build yourself a little snake game to play before you get on a plane. No, I use it. We have a roof garden, so I use it to check on how to deal with the plants on the roof. You know, I've been spending a lot of time, like, building fun little travel apps. So I'm off to Sicily, literally this evening. So rather than planning the trip to Sicily, I built a Sicilian. AI Travel, apparently.
- [06:42]
A
All right. Does it like, does it curse at you and use talk with its hands?
- [06:48]
C
Not yet. Because these systems are a little too psychopathic and a little too well behaved.
- [06:51]
A
Yeah, right. I vibe coded a recipe app last night. I might release it. It's pretty cool.
- [06:57]
C
That's really fun.
- [06:59]
A
It is. Okay. But let's talk business here. So as the owner of all these sort of traditional content properties that have moved to the web, you're facing a lot of challenges with AI around the content creation, the audience understanding your audience, data licensing. So we have a lot to talk about. Let's start with the content and what you're doing with AI to help your content and also how you're potentially protecting it. From scraping and illegal usage or unauthorized.
- [07:33]
C
Usage, rather something I spend a lot of my time on. And I know then the teams I work with spend a lot of their time on is trying to figure out what things that were hard are now easy and what things that were hard are still hard. Because if you listen to the kind of AI hype crowd, everything that was hard is easy. All problems are solved, death is over, we're going to leave the solar system. Okay, that's not true. And then if you listen to the people who are pushing back the onset, they'd say, no, everything that's hard is still hard. This isn't a big deal. This actually hasn't solved any problems. And so that's your two extremes and we're somewhere in the middle. But when you're working through business transformation problems, a huge amount of the question is, which are the things that were hard and easy? What can you automate? What can you get help with? What really helps people move forward? And that obviously changes week by week. So obviously using coding assistance throughout the business. And sometimes they work great and sometimes they drive you off the reservation, right? And so you just. Things can look very easy and then you double checked. As a publisher, we are a large language business, right? We have about a billion, billion five words of content. We produce tens of thousands of pieces of new content every month across all these verticals. And we'll not have somebody write something with AI, right? The AI will not write our content. We truly do believe that you need people.
- [08:59]
A
Do you need people? Is that what you're saying?
- [09:01]
C
One might think you might want people. Yeah.
- [09:03]
A
No.
- [09:03]
C
But the output of an AI system is average definition. You can ask it a very unique question, you'll get quite a unique, average answer, but the answer is going to be average. And we didn't get into this business to write average content. That's a content farm. That's demand media. We shut all of that down as an industry 15 years ago. And if all we've managed to do is create AI content farms, that's not progress, that's backsliding. But there's lots and lots and lots of things that do. A fact is an average answer, right? So if you need to get access to lots of facts and you can build the systems to tap into authoritative sources of those facts, you can do a lot of research work and you can do a lot of the manual grunt labor of moving words around with those kind of systems. And the question is, where and how does that make sense so that we can do a Lot of the work quicker so that we can get to the point where a person can really write a great piece of journalism. So that's kind of the internal question. You can apply that to coding, you can apply that to video production. Right. All of these things. But there's quite a lot of steps even, you know, a lot of the stuff in, like the sales process of parsing an rfp.
- [10:06]
A
Yeah, I mean, parsing an RFP has got to be one of the most easy ad tech AI examples possible. I mean, the RFPs from the agencies are like, give us something innovative. It's never been done before. But also we want banners on your homepage. It's gotta be like a pretty easy thing to handle.
- [10:21]
C
Yeah, but a good RFP response looks at the RFP and says, oh, that's not actually what they want, that's what they think they want. Now let's think about what they actually want. Right. And that's the thing. You don't want to accidentally automate in like the bad RFP interpretation. You want to help people spend time thinking about the good RFP interpretation.
- [10:35]
A
Sure, I get it. On content creation, do you use it at all for summary snippets, SEO stuff on top of your human content?
- [10:44]
C
No, we've done some work generating alt text, the really basic manual labor for images stuff. But mostly it's really deep back office stuff of you want to go pull together big research briefs for a new topic that you're thinking about. AI systems as researchers are pretty helpful, particularly as assistants to researchers. AI systems as fiction writers I do not trust.
- [11:08]
A
Right, right. Okay. So you're creating this billion and a half words of content. You should start referring to them as tokens, I guess. Ask your writers how many billion tokens.
- [11:21]
C
If you want to do the math on that that we have sitting on.
- [11:24]
A
The web, pay your writers per token. That would be a. And now you have the AI folks who want to slurp all that up and train their models and whatnot. Where has people been on that spectrum of licensing, suing, being defensive, et cetera? Yeah.
- [11:47]
C
So we did the first OpenAI deal of any publisher in the US we had that deal in the spring of last year, which feels like a million years ago now. And the deal's actually interesting. It's got three parts to it. One is payment for training. So a standard kind of licensing structure. And it's a multi year deal. And so things that are trained within those multiple years are fair game. As soon as the multiple years ends, then no more training. Right.
- [12:15]
A
But They've already trained, so that just means.
- [12:19]
C
But they also retrain models every month, quarter year.
- [12:22]
A
Right.
- [12:24]
C
Then the second is that every time our content is used it requires meaningful attribution. So we actually got citation into the partnership. And this was before serious GPT existed.
- [12:34]
A
Meaningful meaning it has to be a story about Ben and JLO in order to talk about people. But it does.
- [12:45]
C
If you decide you want to understand what went on with Trade Desktop in the last six months, you might want to reference some Investopedia content understanding fundamental analysis and that will cite its source.
- [12:54]
A
Now I like my example better.
- [12:58]
C
And then the third piece was co development. They helped us build our new tech. So decipher that we talked about last time we were chatting now has OpenAI architecture under the hood both for our own sites and also as we understand the about 6, 700 premium domains on the open web. Right. We've stretched our targeting outside of our walls now and are running it as an open web play. And also we have been helping them as they've been going deep through new releases. So we were deeply involved as they were building out searchgpt. We helped kind of go through that path. And you know, if you saw operator rollout, operator had allrecipes is one of the sites that could take around it. And that's not an accident. Right. So we're trying to work with them as closely as possible as we kind of all figure out as a world, as an industry where this is all going.
- [13:44]
A
My vibe coded recipe site is going to totally kill all recipes.
- [13:47]
C
It's crushed.
- [13:48]
A
It's crushed. Okay, so we're going to use AI to automatically generate these backstories about being in the beach with my aunt and you know, all the meaningful background before I get to my recipe.
- [14:00]
C
Yeah, I love that AIs have the lived experience of knowing their grandmother in northern Italy who had a Sicilian olive growth. But no, the serious part of that is training is one part. Right. And training is kind of a one off thing. There's multiple one off things. They keep training new models, but fundamentally nobody has yet acknowledged or agreed or tried to untrain a model from a publisher. It's like once you've cut the deal, your stuff's in the model, right?
- [14:28]
A
Yeah, that was kind of my question about the term. It's like, let's say you don't come to a renewal agreement with OpenAI three years out, is their model going to degrade in performance or is there already so much learning baked in that it's hard to. I don't know, what's the expression?
- [14:49]
C
Unlearn? Untrained.
- [14:50]
A
Untrained. Yeah.
- [14:52]
C
Well, whatever we're up to by that point, GPT8 GPT8P or ATOR or whatever they end up naming would not be trained on our content. Now they could train it on the output of the previous model, like with lots of things to keep this rolling on. But I think a year ago if you talked to the AI companies and heard what people were saying, there's kind of a path of we just put infinite amounts of information in and then we'll never need any more content ever again. Right. This is kind of a one off deal. Now every single time you hit an AI system, it goes off and reads the Internet, not just for recent information, but also for evergreen information. And partly the reason for this is fairly straightforward. It's trained on all the information, but that also means it's trained on all the information, which includes Reddit and 4chan. If you want an accurate answer, you might not want to put Reddit, 4chan and premium publishing into a blender and see what comes out. Right. That's probably not a very like a fully reliable output. So the best way to get the output is to have all the reasoning, to have all logic, to have the parametric knowledge. Right. The brain of the LLM. But hand it something and say read this and use this as the basis for your answer. And that is very different because it means that suddenly these systems are not one and done. They actually require real time, high quality access to quality information for every answer. And that's a totally different relationship with publishing now than it was a year ago.
- [16:20]
A
It makes sense. And so this is a non exclusive arrangement with OpenAI. Right? Right. So what are you doing with the other folks out there in terms of if they're not licensed, keeping them away from your content. And related to this, I know we want to talk about cloud flares, recent announcements, or however you want to navigate these two things.
- [16:43]
C
Yeah. So if you do not have a deal with us right now and you want to access our content to ground your AI to use as reference, you're not allowed. Right. And historically that would have been robots. TXT was like sticking a sign on the front of a bar saying you're barred, but the person can just walk in. It was like an honesty system that didn't work. Right. Nobody was paying attention or very few people were paying attention to that. So now with ClientFlare, they get blocks at the actually at the CDN layer, at the bot layer. So anyone that comes in who is an AI bot just gets hard blocked.
- [17:18]
A
So you're using that product?
- [17:20]
C
We are. So we switched that on on the afternoon of June 30th, and in the first week we blocked ten and a half million unauthorized attempts to access our content. So any AI company that says, oh, we don't need publishers, like, well, you did ten and a half million times in a week.
- [17:36]
A
Yeah. So this is your message. If you want info about like Jennifer Aniston, don't go to Anthropic. They don't have the data.
- [17:44]
C
Very interesting that if you go to some of these systems now, they'll go to a Bing search. Right. So they'll find the headlines, because Bing can index our content, but if it then pulls in a story and fails, it'll make it up and cite us as the source. So it's actually putting the systems into a broken state where it will hallucinate and cite us, which is its own, like our own problem. That's clearly not okay. Right. But it actually puts some of these systems into quite a broken state. Whereas ChatGPT. Yeah, ChatGPT, we have the license, so it'll go to our content, pull it in and to ground insight and source it on our journalism.
- [18:19]
A
Maybe it's a little outside of your scope of expertise, but how do you think smaller publishers are going to be able to get any of that money? Do you think it'll be an intermediary, like a Cloudflare that'll do licensing of some kind?
- [18:31]
C
So this really is the topic of the summer, right? I think so. Firstly, there isn't a marketplace for information until you stop people getting it for free. Right. So step one is block. Right. And Cloudflare has their solution. They also have a free tier. So even if you're a small publisher and you maybe don't think you need a cdn, you can turn on Cloudflare for free. Fastly's also got a solution for this. I know Akamai has been talking about it. I'm not sure where they are and shipping it, but there's multiple ways now to effectively block people who are attempting to get your content for AI. The second piece is, this isn't like a Luddite. Let's burn it all down. We're not trying to wind the clock back to October 2022. This is obviously the future. AI is a huge transformative industry and information is at least as valuable as a fuel for this as chips and power. But if the information isn't paid for and motivated and incented for people to create more information, then there won't be any more information and all the AIs will get worse. We are going to need a lot more information in the future. Not just from big publishers, from small publishers, from individual creators. If you want great answers for a trip you're taking, then why would you not want to have it pull in information from local bloggers in a different language? It can do that now. That's amazing. But if those bloggers don't get anything for the creation of it, they won't create it. So there needs to be an economy for information that works. And so now the question is, who's going to jump and say they're actually going to build it? What does that economy look like? There's lots of different models. But I do think, in fact, I know that over the course of the summer and some of this was happening before Cloudflare, but Cloudflare definitely moves the story on a bit. We're now, I think at a point of it's not that information won't be paid for, it's that it will, but how and how much and where and who pays and what's the structure. And that's. I really want to make sure we get to the end of it and get to a place that makes sense. But I feel a lot more optimistic about where we are, where we're headed now than I did in say, May.
- [20:43]
A
Yeah, that's really interesting. I have a couple of like random Google related questions. I know you're a good partner of Google over there. These aren't negative Google questions. These are sort of neutral. So the first one is like, have you been following the AI angle to the Google Chrome search case where the judge the case started as search and search monopoly and they were found to be a monopolist. The DOJ case. But in the course of the trial, AI came up and now AI is pretty centered of the trial where the judge was considering forcing Google to respect robots Txt separately from AI Txt or whatever it's called. It seems very relevant to our conversation here in that you want to be in search results, but you don't necessarily want to license all your content for free.
- [21:31]
C
I'm aware of it. I know we're tracking and following along, I think to kind of pull back to the philosophical question. Right. Which I'm not going to give an opinion on the case itself. You've got companies that need access to all information on Earth and if you're running a search engine, you need to be able to index all information on earth to run a good search engine. But that doesn't, from our opinion, give you the right to read, store and reason on that content. For AI, that's a totally different use case to finding out what you have. Right? Like take an example of a future content marketplace for AI, you need a discovery step, but being able to discover the content is totally different from the process of using it. Right. And like those two functions are just fundamentally different economic relationships. And because we're starting from a point of oh, the search engines are also AI companies. So they start off as a massive index and they're kind of just using it because they've got it like, no, we have to draw a clean distinction between those two things. And once there's a clean distinction, then it's going to be a lot easier to kind of reason about who does what and pays who for which. But that requires the search companies to actually draw a clean distinction. We know the distinction.
- [22:45]
A
Well, I'll push back a little bit because historically the ability for Google to use snippets of content has been validated by the courts. You know, Yelp has sued Google multiple times in multiple ways and lost because Google basically took its reviews, snippetized them and offered a zero click result on certain, certain types of searches about local businesses. So why. So this, you're drawing a line, but the line is already pretty dotted when it comes to using the content.
- [23:20]
C
I mean, I think at that point all you're saying is AI overviews are not new. We've had snippets and FAQs and all these other things. So it's a matter of degree, not a matter of total change, I think, kind of. But AI mode is a fundamentally different thing to AI overviews. Right. And we were running Investopedia when snippets came out and one of the questions was Investopedia has a lot of financial definitions. Are we going to lose a lot of traffic? Because there's a short definition at the top of the page, right? Short financial definitions for a commodity. We actually end up with more traffic, not less. Because when somebody asking what is PE ratio? That's not the question. They want to know how to use it, they want some examples. So we actually did great out of that because we brought people from a simple question into the real question, which was 1500 word article with examples and charts. The problem with AI overviews is it does the summary and the deeper dive and takes you down the whole path. It's attempting to actually be the publisher, not quote and cite the publisher and the citations are tiny, irrelevant, hidden. Right. The explicit intent is to keep people on that search page, not have people leave. And you know, you could say we were walking in that direction anyway. And for some things we were. But it is an extreme change in degree, I think.
- [24:39]
A
Yeah, let's talk about the opposite use case where someone really wants to be in the AI results. So this is sometimes being called AEO or geo both GEO is the most confusing acronym in the world because GEO obviously means something and AEO is just almost impossible to say. But basically, yeah, we're talking about brands that want their info in the search results. Just this week and we'll talk about this in the news section with Eric Bluefish, who's a leader in the space, raised $20 million for this and they, they were at the Market Extra Live conference as well. So what is your hot take on this whole space? Is it real? Is it important?
- [25:21]
C
We've ended up in a very strange place because of the UX design of answer engines. This isn't an AI problem, this is a UX problem, right? We just like take a step back and say the job of a marketer is to affect choices, right? Like somebody who's going to make a choice, you do a great job marketing, that person's going to make a different choice, right? When it's people making choices, you can influence people. When the choice is being made by one single monolithic AI system, Gemini GPT4 or GPT5, whatever, then you can't affect the person because the choice got made already. So the only thing you can do is affect the choice being made by the AI system. That's not a great place to have got to. But that's where we are, right? That's what GEO is. It's attempting to change the choice made by an AI when it's finding sources or sourcing facts. If you kind of think about it a little bit further at the moment, AI systems chat as a form factor is many to one, right? Many people ask the question, they get the same answer. Search is many to many, right? It privileges user choice. It's actually pretty humble, right? If you search for something, you get 20 answers because search doesn't know which one of those answers you need set gives you one answer. So it's incredibly arrogant. It's like whatever context, whatever's going on in your head, I know this and the answer you need. And look, analogies of bars are pretty useful for this. I walk into a bar and I get a 300 page cocktail menu. I'm going to walk right out. That's not useful. I walk into a bar, I get five highly curated cocktails and I trust the bar and I can also go off menu and order a martini. Great. I love this bar. If I walk in, then I get handed an espresso martini because it's the average drink. Everyone else drinks in this bar and everyone's got an espresso martini. I'm walking out. Chat is that.
- [27:08]
A
Well, I'm not sure.
- [27:09]
C
The new X designed to give everyone the same answer.
- [27:11]
A
Hold on, hold on one second here. First of all, chat won't give you the same answer twice in a row if for the same human being. You know, it's all probabilistic and in my experience, like when I ask it a product oriented question, it gives me like, you know, this whole pros and cons and a little grid of choices and things like that. So why is that bad?
- [27:33]
C
So you are going to get the same, even though it's probabilistic, it's a probabilistic version of the same average answer. You're not going to get a dramatically different answer. Which means every single person on earth who's asking that question is getting pretty much the same answer, even though there are dozens or hundreds of possible answers. And so if I'm asking what's the best hotel brand, it's going to give me Marriott and Hilton and other major brands. But it's not going to put up any small niche, unique, unusual. It's putting the work on the user to self define what this is. And if everyone gets the majority answer, nobody gets the minority answer. We all have minority needs in various aspects of our life because we're actually all unique and very different. You're not getting those, you're only getting that bland majority, even if it gives you the top three or four. But that's a massive narrow ring of information from search which allows you into all of those routes with a choose your own adventure. So yeah, if I was a marketer it would scare the hell out of me.
- [28:36]
A
Yeah. So what would you do about it? Like, so I, I get what you're saying and I don't see that necessarily changing because the consumer wants an answer and they want, if the consumer wants a, a boutique hotel no one's heard of, they will type that into the chatbot. If they want, give me the major, give me the mainstream hotels, they'll just say give me a hotel. So to, to what extent, if you're the marketer who wants to stand out and Is not. Don't you need tools for doing that?
- [29:08]
C
Push back on the like. This is an inevitable outcome of the form factor, right? I think so. You're asking about your health, right? Which lots of people do. And you say, okay, you know, what's a good or a bad blood pressure, Right. That's not a yes or no answer, right? That's an answer which is dependent on lots of factors like your age, your demographics, your history of things, your family history, all the rest of it. If you walk in and ask a doctor that, the first thing they do is they hand you a form and make you fill in some information so they can give you a better answer. Right. That's not what any answer engine currently does. It just gives you an answer. Right. So you can build these systems that know what they do not know and therefore ask you for the extra information so they can give you a better answer for you. Nobody's doing that yet. But I have seen early prototypes of models and form factors that actually privilege you to choice and put it back in and they give you a much, much better, much more personalized thing very quickly. So it's not a failure of AI, it's just a failure of ux.
- [30:07]
A
So I think you're saying is like if I asked an open ended question like who's the sexiest man alive? I would not just get a single picture of Brad Pitt, I would get an array of gentleme who were attractive.
- [30:20]
C
Yeah, I mean look, we've done one every year for over 50 years, so there's at least 50 sexes alive. Right.
- [30:26]
A
But there's only one who gets on the COVID Right. So I think there's some irony to this conversation.
- [30:31]
C
I mean, I like your soft pitch for the next cover, Ari, so I'll pass it on to the team and take it under consideration.
- [30:36]
A
I'm going to hire some sort of technology vendor to submit my photo over and over again to see if potentially I can rise in that ranking. All right, I think that's a good place to stop. So we have a lot of great news this week. So Eric will be joining to talk about the latest AI news, the new MSNBC rebrand and a bunch of other stuff. Why Brian o' Kelly does not want to be a billionaire. So, Jonathan Roberts, thank you so much for joining. We appreciate it and congrats on the rename.
- [31:08]
C
Always a pleasure. Great to see you.
- [31:12]
A
Market Live is back and better than ever. Our first show was a sold out hit with 450 attendees and an action packed agenda. We even got an NPS score of 40, which means people would actively recommend it to their friends. We're back on October 27th in New York City and the theme is the Age of Outcomes. We're deep diving on everything that makes great advertising accountable and measurable with speakers like Eric Seufert, Dr. Mark Grether from PayPal, and Jenny Wall from Videoamp. Plus our popular AI Startup Showcase and a live recording of marketing. All that and no panels. Go to marketecturelive.com to register now. Early bird pricing ends September 2nd. That's marketlive. All1word.com.
- [31:58]
B
All right, we are back with the refresh. Everybody on the docket for this week. We got, as usual, a bunch of AI to go through, especially on the heels of this conversation Ari just had with Mr. Roberts. We've got a whistleblower allegation against Facebook meta about cheating on roas. This will be interesting to talk about some big legacy TV news, a potential ad tech IPO and maybe a couple other things if we have time. So take it from the top. Let's talk AI. Last week we talked about two of these companies that optimize search for LLMs with big fundraises. This week there is a third. So Bluefish raised 20 million for LLM search. This is interesting, I think for two reasons. Number one, I mean, this category is exploding. These are like big rounds led by notable investors. And then number two, Bluefish is actually the second company with deep ad tech roots that is participating in this space. So the first was Evertune, which we talked about last week. XTTD execs. Bluefish is the former CEO and founder of Promote iq, I think it was that sold to Microsoft. Alex Sherman.
- [33:17]
A
Yeah, Alex Sherman. He was at Architecture Live. He was early retail media sold to Microsoft and unfortunately, Microsoft didn't really run it that well and it's since been shut down.
- [33:27]
B
Yeah, well, he's onto something really big with Bluefish. Have you been tracking. So first of all, you told me that this would be a category before anybody and I really wish you would have started that company.
- [33:38]
A
And I really should have. I was going to start a company doing this like three years ago and right now we'd be, we'd have a product. But I didn't because I'm all talk. No, no game. The I'm an investor in a different company. So there's a third company that's ad tech deep. It's called Gumshoe. They were also a presenter at the Startup Showcase. It's Todd Suwicki, who formerly Samantha and Outbrain. And it's a very exciting space. And those who've been listening, John Roberts had some opinions about this as well. I think it's a very compelling space. I don't know how big a business it's going to be because like we've we said this last week, SEO turned out not to be a business. There's no big SEO companies. Is this SEO or is this SEM or is this something else? Is it HubSpot? I don't know.
- [34:25]
B
Yeah, I don't know either. But there is so much heat in this category, so it's fun to watch. Do you. Side note, do you listen to the Tim Ferriss podcast? Okay, so the Tim Ferriss podcast, he does a running regular one with Kevin Roast, who was the former founder of Digg and did a bunch of other awesome stuff. He's also a venture capitalist. They went pretty deep on.
- [34:50]
A
All of.
- [34:50]
B
These categories that can be popping up around just this transformation with AI. So Kevin bought Digg back from whoever owned it and is relaunching it with Alexis Ohanian, who was the former founder of Reddit, because he believes, and I believe as well, and I'm sure Jonathan said as well, that in this future of LLMs, human created content is just the data that feeds all of this stuff. And it's essential. And you know, we talk about how publishers are under so much pressure, but on the other side of it, content that's created by publishers, content that's created by humans is going to be like ever more important and needs to find a way to sustain. So they're talking about all sorts of interesting businesses that can pop up that aren't even on our radar. Like how do you validate that content is actually written by a human? Like what is the fingerprinting, what is the provenance? All sorts of interesting stuff. And it's sort of apropos what the IAB is talking about this week with this idea of creating collaboration standards with LLMs. So it's called the AI content monetization protocols Working Group comp. And it's got a three part framework, Block Bot Traffic, LLM Friendly Discovery, and LLM Ingest API. Basically the IAB wants to help publishers get to the other side of this.
- [36:11]
A
It's pretty interesting. We talked on the previous segment about the cloudflare initiative and how publishers are going to block the scraping, get paid for the scraping. Very dynamic area. I do have a question about the IEB being the spearheader here. I mean, it's great and I appreciate the Ambition and Tony's ambition running the tech lab app. It's a little outside of advertising. It's related. It's a publisher initiative. So it'll be really interesting to see what the backing of this is, how it relates to. As always with IAB centers, like how is the acceptance or relationship with the big tech companies going to evolve? There's definitely some innovation needed here.
- [36:53]
B
Yeah, I had the same reaction when I read it. It was like, why? Right. This is outside of the traditional scope of the iv. But then I went back to how we kind of introduced this segment where just data and content and human created data is going to be so important for LLMs that the IAB is actually in a pretty unique position since it's obviously it's expanded, but it's got its roots as a publisher centric organization. So somebody needs to be acting on behalf of, I think a bunch of publishers that are not the New York Times to figure this thing out. But I don't know if the LLMs care. I don't know if the LLMs are going to want to deal with the AI from.
- [37:34]
A
From a technical perspective, I think this initiative dovetails a lot with the server side ad insertion project. I forget the name of it that Tony was on the show two months ago talking about, because you're basically going from the old world where you have a bunch of tech that basically creates a webpage and it's just sitting there and anyone could access it. That's how the web was built. And that world is falling apart because of ad blocking and spidering and AI ing and all this stuff. So when you combine this content monetization protocol with server side ad insertion, you're sort of like putting a ring fence around your website. And there are tools and moats and ways to get to the website that the publisher takes more control over and they spit you back. Something that includes ads or includes permissions or is customized for whomever is watching it. And maybe that's part of the web becoming less open, but it gives the publisher at least a say at what they're exposing to the outside world.
- [38:44]
B
Yeah, there are folks that are taking the other side of this. So Trey Totone or Tatoni, how do I pronounce his name?
- [38:54]
A
I don't know. He works for me, but I don't know.
- [38:57]
B
You acquired his company.
- [38:59]
A
Okay.
- [39:00]
B
Our boy Trey thinks that, you know, the LLMs have such massive leverage and there's so many workarounds to, you know, sort of summarize badly a post that he wrote after this whole cloudflare thing that, you know, this, you know, things like this are effectively not going to work.
- [39:17]
A
But I just want to.
- [39:19]
B
Yeah, it just. The data is so important.
- [39:23]
A
Trey has a amazing post coming out this coming week about agentic AI and how ad tech companies are thinking about it and MCP servers and all this sort of stuff. It is, it's long, it's like 5,000 words. And it is like the definitive article on the subject. So keep your eyes out for ad tech explained. Probably come out Tuesday or Wednesday of next week.
- [39:43]
B
That's so awesome. But 5,000 words?
- [39:45]
A
Yeah, it's a little long. Get an LLM for you.
- [39:51]
C
All right.
- [39:53]
B
You found something interesting on X this week about someone basically alleging and providing some proof, you know, n equals 1, that Google search volume is going up as a side effect of search from ChatGPT and other providers.
- [40:09]
C
Walk us through this thing.
- [40:10]
A
Yeah, so what the general data point of view right now is that AI reduces the click out to publishers, but it's actually increasing the amount of searching so consumers are searching more. And Google in particular has said that to brunt any, to blunt any criticisms of its stock price. And this researcher who I cannot vouch for, I don't know this person really at all, but his interesting. Chris Camillo. Yeah, he did some research that showed that it appears that Google search volume is on the back of back end of AI searches such that I think we all know that if you search on various AIs, they do sort of spider and reach out to other websites for answers. In some cases, like with perplexity, it could be hundreds of search results for a single search. And so what this person did was they came up with a search term that didn't make sense that no one had ever used before. And it had zero on Google Trends. And then they searched on AI, I forget which AI for. And Google search trend data showed that strange phrase popping up. And someone else in my comments reproduced this. And it's not clear if what Google reports to the public about search volume is or is not aware of sort of headless browsers and spidering and all of that stuff. But it's a pretty interesting data point that Google may be getting hit pretty hard by the back end of some of these AIs.
- [41:36]
B
Yeah, the question is, does this make up anything meaningful in terms of scale such that it affects the numbers? And I question that. So what Chris did was he ran a micro test. He used the phrase organic Abercrombie socks which had zero search history. And then he put it into ChatGPT and then after 24 hours it popped up on trends. But I think the. It's unclear how much traffic it was. Like it went from 0 to 100 and I'm not sure if that meant 100 clicks or 100% increase off of 0.
- [42:14]
A
Yeah. I think we also have to keep in mind that Google search share, while it has been declining for the first time ever, it's been declining very little. So by, you know, minuscule, less than a percent. So what we're talking about here is potentially that the trends might be a little faster than. Than we thought. No one's talking about mountains of fake traffic or Google falling off a cliff. We're just trying to read the tea leaves on the trend, the first derivative of the search volume. And based on this research, it seems to be indicating that Google might be understating its decline in search volume.
- [42:51]
B
Yeah, I think that's right. Okay, all right, so last one on the AI front. So TikTok is mandating the use of its AI Custom Algorithm Inc. Black box product GMV Max for commerce advertisers. Before we talk about that, as the product marketing naming connoisseur, I want to know your opinion on GMV Max versus PMAX and Advantage Plus. Like, is it a better one or is it just another one?
- [43:22]
A
My first take is that GMV Max is a terrible name. I mean, it's hard to even. It hurts my brain just looking at it on our show notes. On the other hand, like it actually is saying what it does in a way that pmax doesn't. So it increases gmv. Okay, that seems like something I want. I would probably go with GMV plus because then you wouldn't have a second word in there or so funny.
- [43:46]
B
They have to use Max or plus.
- [43:48]
A
They have to use Max or plus. I don't know. Or you know. So. All right, we'll get Back to you TikTok on our recommendations. But I think this is interesting because basically this is just for commerce advertisers. So TikTok has a lot of different types of advertisers and effectively they're saying you can't control where your ads are going to show. You have to use our AI.
- [44:12]
B
Yeah, it's not just where your ads are going to show. It's literally what ad is used to drive products.
- [44:19]
C
So, all right.
- [44:20]
B
Right, so red team, Green team, Green team, on behalf of TikTok is it's the algo is so sophisticated, the reach is so broad, you need AI to do this. Like you Need AI to figure out exactly like what creative to what user, at what moment in what geo. Like all of that stuff because the scale is unprecedented. Red team, the brand side is, I then have no idea what ads are working, like what product from my catalog, you know, is being shown, how that affects my gross margins. There's like all sorts of stuff that, you know, the advertisers I think really need to know for commerce that they don't get. Which I understand both sides, but I.
- [45:01]
A
Think that TikTok's commerce product is affiliate based. Like I think they only get paid if they, if they sell something. This is just for commerce. Right. So someone wants to explain to me that TikTok commerce is effectively like a very thin shell affiliate business where they place orders in other people's e commerce systems, in which case they only get paid on a percentage of spend basis. In which case, like it makes sense. You have to use our optimization or else we're giving you free advertising.
- [45:28]
B
Yeah, I mean affiliates at its core, like real affiliate marketing, affiliates just pick their offer and in many cases like they go rogue because the offer in terms of creative and stuff like that, because the offer needs to be dialed up and you can benefit from affiliate marketing. So to your point, maybe it's just like the way this thing has to go, but I think mandating AI is like, it's interesting and it's a trend and it ain't going away.
- [45:54]
A
Yeah, I think that more and more it's going to be, it's going to be the case.
- [45:58]
B
Yeah, agreed. Especially for commerce, especially for CMB and mid market. Okay, let's move on. Adweek says meta cheated on ROAS by including tax and shipping. Also they use deterministic IDs even without ATT consent. So including tax and shipping, that sounds like it has to be an error. Like somebody just messed up on that.
- [46:21]
A
Like, well, there's a whistleblower here. So there's a person who used to work there, who got fired, who asked.
- [46:26]
B
For his job back.
- [46:27]
A
Yeah, right. And so, and he's, he's saying like he brought it to their attention and there were, it went on for quite some time. But basically roas would be revenue divided by spend. And if revenues increase with sales tax and shipping makes Ross look better by.
- [46:45]
B
A factor of 18 or 19% I think, according to the article.
- [46:48]
C
Which is meaningful.
- [46:49]
B
Yeah, which is definitely meaningful. This was also during the time of, you know, Facebook meta getting hit really hard by Apple. So they were scrambling to figure out some stuff. So I, I understand where the whistleblower claim might be, you know, somewhat valid here. You found something in the article towards the end that I thought was interesting around privacy.
- [47:09]
A
Yeah, well, you know, I think Jason Kent retweeted the article and making pointing out that this was a really big deal on the privacy front. So I did what I never do, which is read the articles at the very end. And so there was a second allegation in addition to the roas, which was that using Facebook's measurement system, off site measurement system, that they were using deterministic identifiers to re identify users on the web, things like IP addresses or email addresses. And that's not by any means like a illegal activity, at least in the US it isn't. But what was alleged here was that it was using that even after a user did not give ATT consent on the iPhone. And under Apple's policies, you need ATT consent to do anything with the user identity, even if it's not the idfa. So Apple's effectively claimed jurisdiction on all privacy within apps because of att. Even though Facebook never asked for an ATT acceptance, they opted out of that whole program. So I think it's kind of a messy issue here because why does Apple have the right to regulate your use of user data outside of att? I don't think they do. I think it's overreach. And then Facebook doesn't participate in ATT and does things which are not illegal currently in the US and this is potentially a hairy issue because they could. Who knows if they're doing it in Europe and who knows if Apple was willingly allowing them to do it in Europe because Apple could have liability if they had a policy that didn't enforce it. And this is exactly the sort of criticism that's also leveled on Applovin, that Applovin might be doing things that are against Apple's policies, even if. If Apple may not have a right to control those policies. So I've been talking a lot and it's confusing, but, you know, it kind of points out how many of these overlapping private jurisdictions there are in the privacy world.
- [49:13]
B
Yeah, what a headache.
- [49:17]
A
Yeah, I don't think it affects anyone on this podcast very much.
- [49:24]
C
Does it affect you? It doesn't affect me.
- [49:27]
B
Let's talk tv. So two interesting announcements this week on the TV front. Again, I'm going to ask for your professional opinion as the meme connoisseur. What do you think about MSNBC rebranding to Ms. Now, which stands for my source for news opinion and the World.
- [49:47]
A
Oh, my God. Oh, my God. Okay, let's start at the basics. Why is it called msnbc? Because there's a joint venture between Microsoft, MSN and NBC. Now, neither of those companies has anything to do with this in the 90s.
- [50:01]
C
Sidebar.
- [50:03]
A
Microsoft's been pieced out of MSNBC a long time ago, and now NBC is peacing out, and NBC, on its way out, kicking it out of the Comcast arena in Philadelphia is like, by the way, we have nothing to do with this liberal network. Mr. Trump, don't sue us. We're taking our peacock back. These guys, we don't know them. Rachel Maddow, Never met her. Don't know who she is. Nice haircut, like. So then, what does the M said for my. What does my mean in that sentence? Like, is it like a portal? We're back in the 90s. I'm, like, going to arrange the widgets on my homepage of the msnbc. Ms. Now. Sorry, homepage. Like, what does my have to do with anything? This is not a customizable product. This is a cable TV news network. Like, I don't understand the my at all. And why does it still have Ms. If Microsoft's not involved?
- [51:04]
B
I am just msn.
- [51:07]
A
Msn, right. Which doesn't exist either. I think it got rebranded live, like, 20 years ago. So MSN doesn't exist. Microsoft has nothing to do with this. And the M stands for my for no known reason. So I'm a little confused, I guess, where my. Net. Net on that name.
- [51:23]
B
Okay, I think we have your opinion on the rebrand. Let's move on. This is actually a big one. Netstar is going to buy Tegna. So, like, two big local TV providers effectively are teaming up for a $6.2 billion deal. And they believe that this is going to happen because of potential, like, changes in regulations. The first person that ever told me about this National Television Ownership Rule, I think it was Chris Vanderhoek, who kind of pointed to it as some examples of how there's just limits on scale in other industries outside of digital. So the National Television Ownership Rule caps a given company's reach at 39% of U.S. households. This would completely exceed that. And the bet here is that in the current administration, you have the opportunity to kind of push the boundaries with respect to regulation. I also think this might be dated just with, you know, where the world is with respect to digital and streaming and unbundling. Rebundling. You might need this for local TV to have a shot to survive.
- [52:34]
A
Yeah, I think the local broadcasters have been arguing this sort of thing for a long time. And there was, there are other rules about concentration. And in various cities that has been a battleground like with Sinclair pushing the envelope on owning multiple broadcasters in the same retro area. And I'm pretty sympathetic in that these companies are really competitive with streaming and Netflix and stuff like that and Ms. Now and all the other companies out there. So it's a shame that we don't have more competition in local. But at the same time the world's changed and I think holding them to these older rules maybe doesn't make sense anymore.
- [53:14]
B
I think that that's right. So this probably does go through one more on the TV front. So LG ads, which we've kind of covered over the course of the past year or so just with this ongoing long lawsuit with Alfonso, which is the startup that LG acquired to build LG ads, it's going to ipo. So we're going to have another CTV Pureplay public company perhaps this year.
- [53:39]
A
Yeah, it's exciting. It's a little bit of a game of chicken. So let me try to recap off the top of my ad here. So LG buys Alfonso. The Alfonso ownership maintains board seats in LG ads.
- [53:53]
B
But not control.
- [53:54]
A
But well, yeah, okay, not control. And whatever we say, we're going to get an angry email. So let's just, let's start there. I always, every time we've mentioned lg, we get an angry email from someone. So LG then replaces the Alfonso board directors in sort of a coup like thing. As it's. Describe it, LG folks, they sue in Delaware, they win. The LG owners win and get control back of LG ads. And as part of their original deal, they can force an ipo. So they are forcing an ipo. And now my opinion is that this is a little bit of a game of chicken where LG may not want a division to go public outside of its Korean control. And maybe a large enough check heads off this S1 in the past.
- [54:42]
B
Yeah, potentially. Just to clarify, it was Alfonso who sued Alfonso who won. Alfonso is now back on the board and pushing for this ipo.
- [54:51]
A
Alfonso I've known for a long time they're a big beeswax customer, I gotta say. Not a great product name. Another, another not so great name. Like what does that mean? It's like some dude with a, with a cigar, you know, telling you what to watch on tv.
- [55:03]
B
Pretty much.
- [55:04]
A
What was up with that name?
- [55:05]
B
I don't know. I don't know. It was a good product. The one thing I thought was interesting in this article, which we'll link to in the newsletter, is that it basically cited that LG Ads was a $750 million business last year, which is big for a CTV pure play. What I loved is that $750 million is 1 trillion won, which sounds way cooler. It's a trillion dollar business. I remember squid games when they were like this last season. Did you watch it?
- [55:38]
A
I didn't watch the recent season. No. Okay.
- [55:41]
B
Yeah. So the pot keeps growing and it's expressed in yuan. And the numbers were just so outrageous that I was constantly just kind of like, doing the currency conversion on myself. So 1 trillion won. I like that.
- [55:54]
A
6 comma club. So, I mean, once again, off the top of my head, and once again, please send in your corrections. 750 million sounds bigger than Vizio's ad business because we know Vizio was public and they're acquired by Walmart. I believe Vizio's, like, services and advertising business was smaller than that.
- [56:13]
B
I don't know what this number encompasses. Yeah, I don't know if it's. If it's like, for, like. But it's a big business.
- [56:19]
A
Video. Video also had, you know, the hardware business. So you have to separate it out. The whole business was bigger.
- [56:25]
B
Hey, like mountain, you know, just pure play.
- [56:29]
A
Eric, I have. I have an announcement to make. I don't know. I don't know how you feel about this, but if someone were to offer me a billion dollars, I wouldn't take it. I wouldn't take it. I would. I would give it away to charity. How do you feel? Would you also take the pledge to not become a billionaire?
- [56:44]
B
Oh, I see what you're doing here.
- [56:46]
C
You just.
- [56:47]
B
You had me thinking about LG and you wanted to slide this one.
- [56:51]
A
I was just moving on to the next topic, like, you know, because the other day someone offered a billion dollars from architecture and I just said, no, that's not. That's not my game, man.
- [57:00]
B
Okay, context for the listener. Friend of the pod. Former App Nexus founder and CEO. Current founder and CEO of Scope 3 portfolio company of Imperium. Brian O' Kelly was featured. I think he was in Fortune talking about how he basically kept, I think, $100 million after the app Nexus acquisition, gave away the rest. And I read that, and I basically interpreted, I think, a little bit differently Ari than you did. I was like, do you think I'm.
- [57:33]
A
Not giving him a fair shake here?
- [57:35]
B
I don't think so.
- [57:36]
C
No.
- [57:36]
B
Let me explain. Let me do this side of it. So the Giving Pledge. The Giving Pledge was started by Warren Buffett. Basically, most known billionaires have signed the Giving Pledge, saying they pledged to give away most of their wealth in their lifetime. When I read this, I said, okay. Bach has a different take on the Giving Pledge, which is just like, okay, I know how much I need pass this amount. I'm just going to give it away. So I don't know how this is different than the Giving Pledge outside of his statement that he doesn't think that anyone should need to be a billionaire in this lifetime.
- [58:12]
A
Okay, I'm poking fun here because it's funny. So if you're a billionaire and you say, I'm gonna give away a billion dollars, that's great. If you're not a billionaire and you say, if I were given a billion dollars, I would give it away, that's funny because. Because, like, you know, I don't have a billion dollars. I'm not giving away a billion dollars because I don't have it.
- [58:35]
B
I still. Maybe I'm thinking about this, like, too linearly, but I think it's just like, hey, more than anyone needs $100 million like you, you can live a great life. And then past that, I think there's great things that can be done in the world and just a fun life.
- [58:50]
A
Okay, I'm updating my pledge. If you were to give me $100 million, I'd give it away, too. Like the hundred million. I'm, you know, I will not accept a dollar over 100 million from architecture. So if anyone out there wants to give me more than $100 million from architecture, every dollar above 100 goes away. Is that a good pledge?
- [59:09]
B
Yeah. Yeah. I mean, as an architecture investor, I'm good with 100.
- [59:13]
A
Yeah. Yeah. I mean, obviously, we have to be humble. Only 100.
- [59:16]
C
Hit the bid.
- [59:17]
B
Hit the bid.
- [59:17]
A
Take on you. Okay. You know, I app.
- [59:23]
B
Look.
- [59:23]
A
I think he has his heart in the right place. It just comes off a little fun. It comes off a little funny.
- [59:27]
B
We should make this the topic of the Q and A with Brian at the next Architecture Live, if indeed you guys run it back.
- [59:34]
A
Yeah, we had a last one. This one. We're going to have Antonio Garcia as our podcast guest. So I have a lot of other things to ask him.
- [59:42]
B
Yes, for sure. All right, we call it here.
- [59:46]
A
Yep. All right. This is a great episode, Eric. Thanks so much. Hope everyone enjoyed our conversation with John. And we'll be back next week. Week.
- [59:54]
B
See you next week, everybody. Thank you for subscribing to marketecture.
- [59:58]
A
New interviews are added every week at marketecture TV and your favorite podcasting app. Thank you for listening to the marketecture podcast. New episodes come out every Friday and an insightful vendor interview is published each Monday. You can subscribe to our library of hundreds of executive interviews at marketecture tv. You can also sign up for free for our weekly newsletter with my original strategic insights on the week's news at News Market tv. And if you're feeling social, we operate a vibrant Slack community that you can apply to join@adtechgod.com.