Who Controls the AI Answer? | Saroop Bharwani on Signal & Noise at Zeta Live 2026
What happens when the search result disappears—and the answer itself becomes the interface?
Recorded live at Zeta Live 2026 in New York City during Advertising Week, Brett House sits down with Saroop Bharwani, Co-Founder and CEO of Senso, for a deep dive into one of the biggest emerging questions in marketing: Who controls what AI says about your brand?
For decades, marketers optimized around search rankings, websites, paid media, and the familiar “10 blue links.” But consumers are increasingly turning to ChatGPT, Gemini, Perplexity, and other AI assistants for direct answers. As Saroop puts it, the answer is beginning to eat the entire funnel.
That creates an entirely new challenge for brands. It’s no longer enough to rank on Google. Brands need to understand whether they appear in AI-generated answers, which sources those models are citing, how they compare with competitors, whether the information being surfaced is accurate—and ultimately whether they can influence the context from which AI systems generate their responses.
Saroop explains how his background building technology inside highly regulated industries shaped his perspective on AI. Long before generative AI became mainstream, financial institutions were already wrestling with issues that have suddenly become critical for every enterprise: accuracy, governance, explainability, permissions, data sovereignty, and trust.
The conversation explores why context may become one of the most important forms of enterprise intellectual property in the AI era. As agents begin operating across websites, product catalogs, knowledge bases, CRMs, data platforms, and internal systems, companies need a trusted layer that tells those systems what information is correct, who can access it, and where it came from.
Saroop also discusses Senso’s Verified Sources approach, creating an audit trail around AI-generated content that can show which sources were used, which humans verified the information, which agents generated it, and how that information changed over time.
Brett and Saroop also get into:
Why AI assistants are rapidly changing consumer discovery
The difference between being mentioned, ranked, and cited in an AI answer
New metrics such as mention rate, citation rate, and citation share
Why enterprises should start by measuring how AI currently represents their brands
The growing importance of AEO and GEO
Why fragmented enterprise context is such a major AI problem
How brands can influence AI answers without simply flooding the web with “AI slop”
Why verified content and traceability could become essential to maintaining trust
The widening gap between frontier AI development and enterprise adoption
Why human judgment, taste, and governance still matter
What happens when AI systems move from answering questions to proactively recommending products
And the increasingly blurry line between an advertisement and an AI-generated product recommendation
The conversation comes at an important moment for Saroop and Senso. Zeta Global recently announced an agreement to acquire Senso, bringing its AI discovery and verified-content capabilities into a much larger marketing technology ecosystem.
If AI assistants and agents increasingly become the front door through which consumers discover brands, products, services, and information, the implications are enormous.
The next marketing battleground may not be the search results page. It may be the answer itself.
#SignalAndNoise #ZetaLive #AdvertisingWeek #ArtificialIntelligence #GenerativeAI #AI #AEO #GEO #AnswerEngineOptimization #MarketingTechnology #MarTech #AdTech #Senso #ZetaGlobal #ChatGPT #AgenticAI #FutureOfMarketing
Transcript:
Saroop (00:00.174)
Two names that no one's ever heard here, so it takes a few times.
Indian subcontinent. Generalized how long do we have?
Yeah. Yeah.
Brett (00:21.759)
If it's a problem, yeah. So and is there a hard stop? Just no because everything else swaps swapped around. So I think we're shuffling with this room, so I think you're good. Yeah, so yeah, th I think thirty is probably a good substantive. Yeah. yeah, so we were saying the region. The region that you're named.
My parents are from Mumbai. Okay. so and then they came to Canada. I was born and raised in Canada. spent most of my life there. I've been back to India a few times, but yeah. So and th there's a
And I know the name defend d depends on the well we should probably I think we're on. Are we on? Are we to turn it on? I think so. Yeah. Yeah, 'cause he just said I'll yeah, I guess we're on, but yeah, so I'll do I'll do like kind of formal introduction. We can have post production edit this stuff out. But yeah, so maybe maybe we talk a little bit yeah, we could we could say, Hey, introduce yourself. right? You don't have to go into the the full background. But but yeah, yeah, it's it's you know, I I've got a lot of Indian friends growing up and and and having
gone there. It's like the regionalisms. You know what eighteen eighteen hundred different dialects? Like the it's more languages in the Indian subcontinent than the rest of the
Brett (01:28.734)
So And someone's last name tells you can some can sometimes tell you the cast. Totally. And the region. Yeah, for sure. Right? That's
Yeah. Totally. And and some last names you just keep adding to it every time there's a new generation. So they get really long. Yeah. Yeah.
All right, so I'll do I'll do my interest and just gonna make sure I get it all all right. Okay. So yeah, so we'll start yeah, so I'll just do a kind of a general intro and then we'll and then we'll we'll start the conversation. For sure. Hey, this is Brett House from Signal and Noise, live at Ad Week advertising week, I should say, twenty twenty six. I've been saying ad week, but people are confusing you know, that and the actual publisher brand. and we are not in the Hollywood studio that we were producing in
In some of our previous episodes, but it's Zeta Live. we just saw Alex Karp speak, which was really interesting. And we have got Saroop Barwani. Barwani. Barwani. I knew I was gonna mess that up. Damn it. Saroop Barwani. We talked about this for five minutes and I still mess it up. And this is an ongoing joke, as I told you, for signal and noise. senso is a company that you founded and co-founded. Co-founded. Right. And ran. And and we were talking ahead of time of some of the the kind of origin story.
A very interesting company. Zeta's doing some some really interesting things with with Palantir and their partnership and the rollout of kind of an AI centric.
Brett (03:09.486)
tech model, which may not mean a lot of things to everybody, but you know, building kind of an AI and a sovereign AI and sovereign sovereign data strategy as an organization. Cause data kind of grew through a lot of acquisition live intent, right? To bring all this stuff together to ensure that you can deliver kind of cross-channel you know, marketing outcomes and advertising programs is is is important.
and you need like a continuous learning system around all of that, right? And that requires pretty advanced analytics, which is where sort of AI comes into play. But but that being said, and and these these I I'm feeling like I can hear my voice in these things. I'm feeling like we're alone in a in a little bubble out in space talking podcast world. so so introduce the the audience to tell tell them a little bit about your journey. I thought that story around you didn't come from a typical ad tech and martech background.
and kind of what and and you've worked in a lot of regulated industries and what kind of wh when did the light bulb turn on and and what kind of w how did you originate the company? What was the original idea and how did it evolve over time?
question. I've always been a builder from a young age. My dad dad taught me how to code when we actually needed to code. Yeah, yeah. and over the years I went did computer science at school and I was always fascinated by, you know, how the intersection of the human brain and like machines. I didn't really in school early two thousands was fascinated by it. But in two thousand and twelve in Toronto, it was the dawn of deep learning. guy named Jeff Hinton and Ilya Sutzkever, who's one of the co founders OpenAI,
They basically create they basically proved out that a model, a neural net, could tell the difference between cats and dogs. Yep. And that spawned off and I was inspired at that moment and I was like, I didn't think in my lifetime AI would actually be a real like commercial application, but in two thousand and twelve I was convinced. Took a few years and I had much experience working in regulated enterprise. So telecommunications, banking, insurance. I come from Toronto. If you've ever been to Toronto before, the skylines like
Saroop (05:16.544)
a big tall CN Tower and then all the big banks, right? Yeah. So I've been like working in banking for my entire career and in banking compliance is as important to a marketer than the actual like transaction and and then the brand.
And with that comes things like data governance, data protocols and security, so for privacy protection, right? Exactly. D data sovereignty in terms of keeping the data within a closed ecosystem within the brand.
And there's a lot of questions around that now. But in two thousand and eighteen, built a an amazing team, my co founder Tom Nelson, and a and a great very technical team that knows how to build models from scratch. this was before LLMs really popped. so we were building models for regulated enterprise and the and most of it was marketing and customer support use cases and we realized that factual accuracy
conflicts, really getting the right context to train the model so it had the right output to control that was a key capability that was needed in enterprise by default because compliance and explainability is a is a huge, huge topic that just naturally
Explainability, that's a really important topic, right?
Saroop (06:22.508)
Yeah. Alignment explained. Now all of these things are getting huge attention, but back then in a regulated enterprise, these were all defaults. I got a really early preview in how regulated enterprise, like large enterprise, thought about this. Yep. And then all of a sudden, you know, we saw the kind of like painting on the wall in in two thousand and in two thousand eighteen, two thousand nineteen, you know, Google launched a paper in two thousand and seventeen called Attention Is All You Need. That is the core paper that
Early days of the attention economy.
Transformer model. Yeah. And then you know, we got early access to GPT two and GPT three and then all of a sudden on November tw thirtieth, twenty twenty two, chat GPT launched and I'm like, this is the moment. This is when AI becomes useful. Yeah, yeah. And
This is when AI began, I think it might have been GPT two, but I might be wrong. We'll have to check check the record and we'll put the but when the it began to pass the Turing test. Right. And and we talk about that. Some people not everybody knows the Turing test, but it's basically like when you can when it's indistinguishable the output from a human output. When AI can create something that's completely indistinguishable and and it that that's when you pass the Turing test. And that's when you know you're starting to get to a level of we won't call it sentience, but we'll call intelligence and
in in predictive analytics that's at at a compute power level that is is comparable to what you would think of as a human solving the same problem. Prior to
Saroop (07:40.51)
Chat GPT, you know, we the deployment of chat bots in in the enterprise, it was very if then. Yeah. You had to kind of like it was it was basically that. So you could never make it answer everything no matter how much context you gave it. It was it was a struggle. I've I've been through those but once these models came out, yes, like inklings of the Turing test started to become real. Like I mean the first use case I was like
Bul bully it live.
Brett (07:56.116)
I've worked some bullion logic in my team.
Saroop (08:06.604)
Write me a Seinfeld episode about X, Y, and I was sending it to all my friends. That was the first thing I did.
Was that your
I asked a lot of history questions. I was like, you know, tell me the entire history of Stalin. I said the rise of Stalin. I don't know why I went dystopian, but I that was like I wanted, you know, just deep history questions to see how accurate its its outputs were. In in terms of my understanding, not being kind of a history buff, but like that was it was the it's funny, and you started with a complete
I started with a sign I sent it all my friend try this. And and and and you know, look a few years later a billion, you know, w weekly active users are using it now. I think they're a small two billion
Can it be funny?
Brett (08:44.994)
Yeah, it's it's one point two billion. Yeah, no I think I I quoted like nine hundred and fifty million. But weekly active users at one point two billion in like a couple of years.
In four years, yeah, that's where we're at right now. And and there's not just one platform, there's Gemini and Claude and Perplexity and many other models like forming. And I spent a lot of time in Silicon Valley. I spent a lot of time with the labs and you know Y Combinator actually pulled me into San Francisco in two thousand and twenty four. And so from Toronto. from
Why combinator, right?
Brett (09:13.966)
Told you from where? From Toronto, yeah, okay. So you were in Toronto, then you yeah.
My ma my career was always building models for regulated enterprise, but having the perspective on what's happening in the major labs and the ecosystem in Silicon Valley, it's like these are two different worlds that speak very different languages but are so connected. Yep. They're just so connected. Yeah. And you know I'm jumping ahead a little bit here, but when when I met, you know, David and Nij and the team at
Nij is great. We actually interviewed him at Possible in Miami.
High possible, yeah. I was like, this is an opportunity of a lifetime for me to do what I think is my purpose and that is connecting the most in like the most important brands on the planet. all Zetus customers, with frontier models. Yeah. The open source ecosystem. And that is just
And thematically connecting them to the the radical change in consumer behavior. You know, you see the one point two billion users of just ChatGPT, it's gone, it's completely global. And and the way that people search and find things, it's like the days of Google's, you know, blue links. Ten blue links. Yeah, ten blue links and and the first ten being or you know, paid, you know, and then you had to look below for organic. So when you're looking for a like a direct answer to your question, oftentimes
Brett (10:28.086)
you have to really go to even page two, page three, page four. And what I what I read recently is that G GPT results, thirty percent of the results are are answers that would have not appeared on page one of a Google search result. Which how many what's the percentage of of people that actually go to page two and page three? Very small, ten percent, less than ten percent? So it's incredible because it's going much deeper, much broader.
and then aggregating information and hopefully giving you, you know, a direct answer to the question versus, you know, we call it the shittification of the Google experience. You know, th then P Mac started flowing into the whole page search model, right? And so it's like, well, no, I had a question that I needed answered. I didn't want to buy a product. Right. So you're almost forcing or assuming intent. And what I've heard everything, all the research I've done around GPT specifically, a lot of the research is coming out of, is that it's the the intent of of
people that are doing product or informational searches around product services or solutions is significantly higher than any other sort of we'll call media exposure for lack of a better term. there is and I think there's facts around that. Totally. which which is kind of no surprise. You're looking for answers. And if you're looking for a product, you're gonna ask specifically, hey, rank the the top five, you know, home appliances of X type or whatever.
And ten blue links was, you know, it got you to a certain point. It di allowed you to discover, you know, a page to click on and then you had to kinda go on there and kinda select the products and check out and do all these the these manual things that that now machines do. Or maybe it was just research. But now the answer is eating the entire funnel. Yeah. And that's creating organizational design changes that are affecting the entire organization and and enterprises don't change easily.
Yeah yeah right yeah big brands definitely don't change.
Saroop (12:19.082)
So now you have the the SEO teams, the the teams that are focused on on CAC and really kind of actions and then the teams focused on payments, all having to kind of just look at this answer, yeah, this answer box and say that is our canvas. Yeah, yeah. Right. And that that's where everything is is the it's like a vacuum. And so it's become really interesting to see how different enterprises across various industries, and we work across like automotive, insurance, financial services, retail, so we get broad perspective on the patterns on how
And that cross industry and cross category insight and in information sharing is incredibly important because it just gives you a lot of wisdom in terms of consumer behavior and how it applies. Yeah. Right. you know, and I think it's it's to me that the back to your point about the Google model and and sort of you were kind of describing friction in the consumer process and the consumer experience, the CX, right? You go in, you do a search, let's say you do have high intent and then you've got to go to an insight, like a publisher site, and we talk a lot about how publisher traffic has dropped dropped
Not I
Brett (13:16.526)
Precipitously over the last 20 years. So the open web is suffering. But point being the consumer's got to take a few steps to get through to a some sort of of experience, right? Like if I'm it's a it's purchasing experience, an information gathering experience. And in classic sort of startup founder mode, somebody's like, there's friction in this process that can be removed. I mean, think Uber in the taxi industry, right? There's friction in this process. If we can remove that and create something that's better, that's you know.
It's it's it's going to win in market and I think that's what they've done. They've sort of c it's like cut out all that middle stuff and just get right to the
Point. And look, ty Bezos. I always remember what Bezos said in the early days of Amazon. What are like, no matter how much time passes, decades, what are what's what's always gonna remain the same? People are never gonna come back and say they want less convenient experiences. More complex spend more time on doing like searching for a travel experience. Or they're never gonna say, hey, we want to spend more money. Yeah, yeah, yeah. Right? They're always gonna say.
Exactly.
Saroop (14:20.672)
So these are things that are just timeless things that you as a builder you have to take into consideration.
You think you see the genius. I think we're kinda getting to the genius of this. And we and we talked at the very beginning about like this is the attention economy, right? And the biggest challenge is that, you know, and you look at the advertising stats and people's exposures to advertisements have grown by a order of magnitude over the last twenty years, in the last five years, right? In terms of the total number of ads we're exposed to on a daily, weekly basis. So every brand's fighting for attention, right? And how do you remove friction from that process and get quickly to
to the convenient answer. That's what that's what these L LMs I'm not and it's it's sort of like I'm not putting them up on a pedestal, but that's what they've solved for. Right? Like how do we cut through all this attention morass that's around us exact and get right to the point and and give utility to consumers so they can get directly to the point. Yeah.
Exactly. But but you know, the answer is the canvas. But now there's a whole s other slew of problems. Like, you know, is it sourcing the right information? Yeah. Is that information outdated? Does it conflict does a publisher site conflict with your own site? Yep. is there missing information, which, you know, I'm not displayed, but my competitors are, right? like there's all these like fun topics that come up and each of them are categories, right? They're very
Different like you can write a chapter and a book on each of those. Yeah. And and you can go deep in each of those. I mean maybe we'll save that for our
Brett (15:41.618)
Our our longer episode, we're gonna do a deep dive into AEOG, right? Yeah, yeah, you know, it you can write a a chapter and and part of the challenge I think with with brands and with publishers, so the buy and the sell side is sort of getting their data house in order. Right. We talk a lot about, you know, you know, we were at we had Telium on the podcast and they were a tag management player that became kind of a C D P. and they talk about just the mess of just legacy tags that people have for tracking purposes, firing pixels.
Tied to a thousand and one campaigns across the web that they have no governance over. And so there's just this collection of data, most of which is noise, and it muddies the the decisioning process, the analytics process. You know, if they're plugging in agents into kind of a a unified data layer, oftentimes that can be because there's just not a fundamental governance to all context wrong.
Context wrought. Context is centralizing context in a permission-friendly way so that the right agents and the right humans have access to it is a problem that hasn't been solved. We're in the early days of that. But we all know that context does need to be shared. It does need to be centralized. It does need to be accessible to the models. It needs to be governed by the humans to verify it's correct. Yeah. Because in a world where agents are
Literally under the hood doing everything for us and we get
And in sort of a headless sort of way. They can go they they can go a bunch of different let's say internal, you know, your your your three data lake houses, your ERP system, your CRM system. You can set them up, but they need a governance layer. They need a layer that says, hey, we've got this un now now we've got this unified data ecosystem within our brand or w as a publisher, right? How do we actually you know tell the agents how to go into this data, how to pull things, how to interpret the data, right? And that becomes your semantic layer and your context layer.
Brett (17:29.492)
And and would you agree that that context layer is sort of where brand IP, like the defensible like because everybody's got the data. First party data arguably for a brand is is unique or audience data for a publisher, right? It's certainly they share some of the same first party data with other brands within the same category, right? Or publishers, right? But would you agree would you argue that that that top layer, that context layer that that is that is governing in in all of your your IP as a brand, all of the pattern recognition.
all of the rules of engagement for how the agents interact with with the data. How critical is that to like brands from a differentiation sort of go forward perspective?
I mean your website has always been a public source of context on the web to humans. Now it's it's a public source of context for for agents primarily. but but internally in in any enterprise you have private context that's usually fragmented across the organization that is you know, these models all of a sudden give us the ability to kind of throw like natural language at it and all of a sudden it can find useful things. But the context being fragmented
It is the problem. Yeah. So so most organizations are gonna figure are currently figuring out how to bring that together in one unified context there where permissions are incredibly important. But when there's an evaluation on, hey, this is incorrect or correct, that's when a human needs to step in and verify it. And we've actually just introduced a product a few months ago called Verified Sources. Okay. Where you can publish a a P
It's you can publish a piece of content on the web. It has a full audit trail attached to it on which humans verified it, which agents generated, which sources in your ground truth across your product catalog, your knowledge base, everything actually produced that that generated that output. Yeah. And every time it's updated, it has a version control history so that when it's decided on Chat GPT, you can access the history. In case anything goes wrong, let's say an agent's making a transaction on a product, then some piece of information is incorrect and there's a chargeback.
Saroop (19:31.918)
You're gonna need to
Yeah, so you need the audit trail.
For sure. So you need the context trail that goes with it. That's incredibly important. And it's early days on that.
And I don't I don't think you can understate how important that is. Traceability. Yeah, the traceability in in the in then from a consumer lens perspective and even from a brand management or governance or publisher management or governance perspective, it's the it's the traceability, but it's also the trust. Correct. Like how do we trust, you know, it it it present it it it you know created this result, this whether it was a a commerce situation or it was an information situation. What went wrong and then how do we optimize that process going forward?
And and the implication the stakes are highest at the bottom of the funnel when a transaction occurs. At the top of a funnel, an inform piece of information can be incorrect and it could mislead someone. Yeah, yeah. But but as it goes to the bottom of the funnel, that that's when brands really gotta pay attention to the trust layer and ensuring their ground truth is consistent with what's being displayed in the model. So that constant cross-referencing and evaluation is something that you know we do on a typically a daily basis across thousands of brands right now, across hundreds of industries.
Saroop (20:35.852)
And we're just analyzing like who what's incorrect, what's correct, who's represented, who's not represented, what sources are representing those brands. It's an incredible amount of data that we're able to kind of look forward, look back on and really come up with insights that help brands like really gain perspective on on on how to really approach representing themselves on these new channels.
Yeah, and and how well prepared are some of the big brands that you worked with? Because we talked about their slow moving, they're largely you know, large brands, enterprise brands, matrixed organizations. Like how how mature are they in terms of their improving their ability not only to be found, but also to control the content of the answers. and maybe you can you can you can dovetail into i the the distinctions between AEO and GEO and how brands should be thinking about those things.
Percent. so as I said, I spent a lot of time with the the frontier labs, particularly the research teams, and I spent a lot of time with enterprises. And I can tell you there's an extremely large divide in the adoption curve between the enterprise side and the frontier labs. Yep. And the opportunity that exists for anyone like that's in our shoes is like how do you bridge that gap? Yeah. Enterprises
So you're saying what we're saying right now is it widely, you know, inherently intrinsically understood by like your average brand marketer? Or are we just really wonky, geeky nerds that need to explain this?
Well the the the problem is
Saroop (21:58.888)
It's moving so fast that humans are inherently linear thinkers that are not used to this like leap exponential type of change. Yeah. I can tell you in in in two thousand and twenty seven we are gonna have recursive self improvement. And recursive self improvement is is when the models day by day they can improve themselves and we're already seeing inklings of this. And if you could leap forward to when RSI as a catalyst to potential super intelligence exists, w w we're we're struggling
Recursive self improvement's a big time.
Saroop (22:28.695)
enterprises are struggling to keep up now.
Yeah, 'cause 'cause when it when you get to recursive self-improvement, oftentimes, well, even before that, even now with some of the most advanced models coming out of the Frontier Labs that haven't been released, they don't necessarily understand because it's it's it's developing, you know, maybe you don't say the word sentience, but it's certainly s there's things like self-preservation, self-control, self-direction. It's improving and optimizing it's it's creating collaborative functionality, right? Like where it's almost like humans collective.
You know, like we're gonna we're gonna we're gonna build other agents, you know, think of them as as sort of delegating responsibilities, right? All of and then yeah, i people don't know where that's necessarily gonna go, right?
Payton enterprises are still at the the point right now, they've adopted, you know, chat GPT enterprise Chat GPT licenses or cloud recent. They've rolled it out through the org. You know, some some are more ahead than others. They're they're in the mode of how do we centralize our context, but that there's a lot of implications around that. Who has access to those and you know, various things. And then there's just organizational friction. So what happens in your average enterprise, I explain it like the barbell. And this happens in every every enterprise. You have the top five percent that just
dive in, they're like us, they're like the model, it's inherent to them. They just get it. And they're using they're like token maxing to the extreme. Yep. And doing all sorts of
Brett (23:42.734)
Yeah, I talked to a guy yesterday that investment banker, an individual s like a solopreneur, maybe maybe not just an entrepreneur, maybe he's not solo, but that uses that pays ten thousand dollars a month on tokens.
Right. And but that's a very small piece of the organization. Then you have like a small piece on the other side, maybe it's like five, ten percent, that are resisting and they're just fearful of it. And then you have majority in the middle that are curious but they lack understanding and they need education. Yeah. Right. And and and those individuals have to shift over to the you know, the the the the adoption side where they're using it every day.
It's sort of a little bit of an a an existential crisis. A little bit of adapter die, a dr you know, don't you know, for one, don't let fear get in the way. and I I've had friends say to me, you know, I'd r I I might just retire. Like you know, like people in the in the twilight of their careers. I'm like, Do I have to really learn all this? And I'm like, Well you know, I I actually think I'm kind of if you call it, you know, the the apocalyptic thinking, I'm like a apocalyptimist. Right. Like I'm like very optimistic about this this actually transforming work and taking the things that I don't like to do.
away so that it can focus on more high value strategic work and and outsource the stuff that you know, building a PowerPoint, right? Things like that. You know, but not not giving away my human judgment. Totally.
sure. Judgment and taste are still very very that is extremely important. Agency, judgment, taste, those things are are still very important. but it's never been a better time to be a builder if you love it. Yeah. you know
Brett (24:59.768)
Craft advertising.
Brett (25:14.028)
Six million new companies year to date were f in the US right now. Right now. talk to Rashad Tabacawala from from and that he was he was naming that stat and and we're I'm advising a stealth startup that is going to get trying to get incorporated, escorped in Delaware. And there's like a two month wait. Two months they said they we have never seen this this type of wait time and this type of incr incredible surge. So to me, great time to be alive.
But back to that, let's finish that that that you know in terms of being a builder, if you're a builder or a founder, this is this is a a magical moment that's ne that's we've never seen in the may maybe since the invention of electricity and arguably. It's this is beyond the industrial revolution, right? I've heard people say invention of fire, right? Like that you know, this is this is a catalyst that's gonna change humanity, right? But but without the without the the futurist thinking, the future optimist thinking, so so yeah, organizations
still need education, they still need to understand where their data is and how to kind of up level that sort of the the the bar in between the bar bells, right? when it comes to to kind of building discoverable content online, right? you know, so that their brand can be discovered and represented correctly with a obviously a dominant consumer tool, which is LLMs, what what do you how are you coaching them on on
And and what are some of the details around how you set up specific next steps to be able to get started?
Yeah, you know, i I think that at the beginning, like this is so new that oftentimes, you know, just being so deep into this, sometimes I walk into like, you know, a new a a meeting with like a big airline or a big big brand and I just assume that they know some of the stuff that I gotta take a step back. That's basic. Like explain what the metric what is what is mention rate actually mean? Yeah. What does citation rate actually mean? What does citation share mean? Right? Like what are the like how do you actually just appro like that's explain it like I'm Vive.
Saroop (27:14.924)
Yeah. Right. So a lot of the time it is just starting the education on what are these new metrics, what can you actually measure, and what impact is it gonna have. And then you just start with measurement. Just start with by the telling them a story, using data on where they stand in the market, like relative to their products, topics their consumers care about, at each funnel stage, relative to their competitors. Yeah. or their peer.
groups. And this shows not only frequency of being found versus their competition or likelihood of being found, but also the actual content that's being bit that's being shown as the answer, let's say
So if you do a search on Chat GPT or Gemini, but you'll see these little cute little pills, these like these citations there. Right. And you can actually like measurably influence which citations appear in that. And there's a whole funnel of a an agent journey all the way from visiting your website and indexing you to actually citing you. Yeah, and and that could have a full audit trail and attribution tied to it. And that little citation influences the text.
that's in that answer. And and in measuring what text appears with what citations and doing that every day, all of a sudden having a historical record of that and having a really smart team that that just understands how to build these models and control the narrative is very powerful. And you can actually control the text in the answer and sort of tweak it the way that you want it to be represented. Obviously in an ethical way. Like, you know, we you have to tell brands like a lot there's a lot of organizations out there, n not organizations that we work with, that are just dumping a lot of slop on the web.
Yeah.
Brett (28:51.15)
We call that we call that the AI slop cannon. You know, and and people are starting consumers are starting to there might be a backlash because people start to see and know what is real and what is human what is not real. Right. And especially with the younger generations, there's sort of like, you know, the the the there there's a lack of trust oftentimes.
trust, right? Yeah. So there's slop. You know, like when you're on X or when you're on Instagram, you see a little badge check mark saying, hey, this is a verified identity. It's a it's a verified individual. I think that the same thing needs to exist for content, right? What is a verified source? You have that check mark. Yeah. Because there's going to be so much content on the web. S so much, so much slop that exists that you need a a some sort of standard that exists between
What's actually been vetted that tells the story of an individual, an organization, a publisher, whoever's putting content on the web, there needs to be a stamp of approval and that needs to be differentiated from the slop that will exist.
Yeah. and and do you sometimes when you walk into like a sales a sales meeting, right, do you show is there a little bit of FOMO like or or FUD, right? Like where you actually show them an example of how they're appearing within a particular Gemini or Perplexity or Chat GPT that's really bad, like they didn't control the quality and content within that answer and they're like shocked to see it. Is it is this something that you've seen and shown them and they're like, What can we do to prevent that from happening?
Because because this has been thrown at them and they don't control the narrative, yeah. There are more instances of that than there are not. Yeah. So it's very
Brett (30:23.246)
So the examples are abundant.
And and we we we we just it's this is automated for us surface stuff, but you have to get granular enough. Yeah. So you know, when you think about funnel stages, we we have four funnel stages. There's two non branded ones and there's two branded ones. The non branded ones are top of funnel discovery awareness when you know you wanna like your your pair of sneakers, like you need a new pair of sneakers. You don't even know you need running shoes.
I'm looking for a new pair of running shoes. This is the type of shoes or yeah, whatever. Yeah, yeah. Like this is what I'm using for. Yeah. Right. I need yeah.
versus the next layer down is you you know you need something. It's like what are the best, you know, sneakers, like new pair of sneakers out there, or what are the best life insurance products out there? Yeah, right.
I'm a cross country runner and I need, you know, like something to
Saroop (31:07.746)
Yeah. So that's a consideration level question and those typically display rank, brand ranks. Yep. And and when you display those and and you know, we measure this, i if if a brand is not ranked number one, that is like that hits them. Yeah.
Yeah, they're like, my God, and they see all their competitors. It's like Nike wants to be one and they see Adidas and New Balance and D Right. And the funny thing, I I so I did some research. It's it's funny because you do research like this, that's a and I should probably update this research, but it was about a year and a half ago, let's say two, maybe two, but a year and a half, which is ancient history at this point. But it was basically where GPT, Perplexity, Gemini, all the major frontier models are sourcing most of their information.
Exactly. Exactly.
Brett (31:46.078)
And and at that d when I did that search, I meant sixty seventy percent. I know this has changed recently with Reddit. Was Reddit? for I think for GPT, I think that's changed. I think they've they've certainly modulated that. Wikipedia. So so think about the sources. Where is it gathering this information, right? And do you want that type of source? Think about some of the the downward spiral conversations that happen on Reddit. Do you want that that sort of crowdsourced, uncontrollable information fueling an answer?
about your product, your brand, because it may not be what you want to see or you want consumers to see. So how do you control that? And so how so let me ask you this one question. Is so so you're controlling that or you're helping to manage that process. Where does it go from being a true answer? Right. And I think this is at least the positioning statements of the big frontier models is like we want to, you know, maintain the fidelity and the quality of the response so that we do not lose consumer trust. And if we start throwing advertising
People are gonna think this is this is pay to play. Yeah. So how do you actually prevent that pay to play sort of perception from becoming maybe reality? Mm-hmm.
Not all the model providers are are going towards ads. there s some some of them are staying away from ads inherently. OpenAI takes the stance that, hey, our ads are not gonna influence the answer, but the context of the answer will
enable a a you know a better ad, a more targeted ad. And it is it is bringing up the question, you know, what is an ad? If an ad is the perfectly targeted product for you that you need in that moment, and these models are getting more proactive. It's less about asking a question. It's more about it telling you what you need because it has all the context available. And as that gets better over time, it should just tell you, hey, you kind of need a new pair of sneakers because it's been a while or, you know, it I on on some other channel it's broken. You know
Saroop (33:35.246)
Athena everywhere, right? It should know that you're sneak you need a new pair of sneakers, right? And then it should recommend that on the spot. That's kind of the the the next phase that we're going into. And then is that pair of sneakers is it an ad? Or is it just a really great product recommendation? So I think there's blurred lines being
Yeah, yeah. And it's kind of a f a balancing act between d don't lose consumer truck, don't don't believe that this is just a mechanism z that's just shilling whoever's gamed the system well enough. Exactly. Right? Like yeah, and so so there's still relevance to the answer because at the end of the day that's where they're that's that's the the attention win that they developed along you know, a couple of years ago. Right, that we're gonna cut through all of this noise and all of this friction to get right to the answers that people are looking for. Exactly.
So I think that's it. You were we're get we're getting the warning that we're like we're like we gotta cut. We're gonna we could go on for hours about So we'll have you on a long version form a long form version and amazingly our episodes we have sometimes Joe Rogan length episodes, two hours and fifteen minutes was our one we had last week we did two that length. Fifty percent completion rates. So which is surprising. The people like long form content. It's the entire world isn't becoming TikTok. But hey, this was great, great talking to you.
For a couple of hours.
Brett (34:45.142)
congratulations one on on the whole Zeta acquisition in that hole. I'm sure it's a whole new phase of of the company's life, of your own life. Completely. and thanks everybody for joining Signal Noise at Zeta Live twenty twenty six advertising week in New York City. and we'll see you next time.
Brett (35:04.27)
All right. That was awesome, man.
Saroop (35:12.6)
You got a few more of these?
Yeah, we got we're doing Michael Cassan and then Ed C. Ed C yeah. Yeah, yeah. And and then if they can squeeze one other in, but like Michael Cassan, who I know, but it'll that'll be interesting. Yeah. I might not be able to be I can't I may not be able to get as geeky. I gotta keep it key.
C
Saroop (35:31.242)
I'm I like the I mean you ask great questions.
Like yeah, I I w yeah I want it to be less about like just just tell me about you know and then and then you sit there and talk about the announcement and about your it's more of like let's educate right that's what we're doing. We're like and if I don't know something you might know it or vice versa, we can share it that this that's what a conversation is. That was awesome, man. Cool. Are we now we gotta somehow get out of here before without look at that. There we go.
Hundred percent. No, it's great. Thank you so much.
Brett (36:02.294)
So you're in so you're in the West you're in San Francisco now or
Toronto
Comments