This conversation delivers a clear framework to navigate AI-driven search today, and to win as the landscape blends search, chat, and commerce. If you touch SEO, PPC, ecommerce, or brand strategy, this episode gives you an actionable playbook for the new search reality.
https://page2pod.com - What happens when search engines evolve into answer engines and AI determines what your audience sees? In this episode, Jon Clark and Joe DeVita talk with Amos Ductan, SVP of Search at Razorfish, about the radical shift from traditional search to AI-powered discovery platforms like ChatGPT, Perplexity, and Google's AI mode.
Amos dives into how Razorfish is merging SEO and paid media strategies, preparing for a future where prompts replace keywords, and "just showing up" in AI results becomes the new baseline for success. He also introduces the concept of "The Correction"—how AI search is recalibrating both consumer behavior and the KPIs marketers have relied on for decades.
Whether you're leading an agency or managing digital campaigns, this conversation is packed with tactical insights on AI visibility, feed optimization, and how to future-proof your marketing strategy.
🤖 In This Episode
• What an "answer engine" is—and how it changes everything
• Why AI-driven search reduces the number of clicks and reshapes attribution
• How Razorfish is redefining paid + organic search teams
• The concept of “The Correction” in AI search and marketing metrics
• How to audit your brand's AI visibility and clean your product feeds
• Strategies for preparing shopping feeds for AI commerce (e.g., Perplexity & ChatGPT)
• Why traditional conversion metrics may be obsolete in an AI-first world
• How "AI literacy" is becoming an essential skill for marketers
• The tools and prompts Razorfish uses to track brand presence in AI models
• Voice search: where it stands now and what’s coming next
Unlock a clear understanding of where search marketing is headed—and how to stay ahead.
👉 Subscribe to the Page 2 Podcast for more episodes on the future of marketing, AI, and digital strategy.
💬 Comment below: How is your brand adapting to the rise of answer engines?
📚 Mentioned Resources & Tools
• Amos Ductan on Linkedin (https://www.linkedin.com/in/amos-ductan-b516bb6/)
• The AI Search Playbook Webinar (https://events.publicisgroupe.com/razorfish-ai-search-playbook/rzfli)
• The New Era of AI Powered Search (https://www.razorfish.com/articles/perspectives/the-new-era-of-ai-powered-search/)
• How Agentic AI Will Reshape Search (https://www.razorfish.com/articles/perspectives/how-agentic-ai-will-reshape-search/)
Jon Clark (00:00)
What happens to digital marketing when clicks vanish, search fragments, and AI decides what your customers see? Amos Ductan is the SVP of search at Razorfish, where he spent over 15 years helping brands navigate the shifting ground beneath the search industry. But lately, his focus has narrowed to one thing, the rise of answer engines. Think Chat GPT, Perplexity, AI mode in Google, and what it means when your marketing metrics stop making sense. If you're wondering how a big agency is actually adapting to AI,
not just talking about it, this episode is for you. Amos walks us through what it means to show up in an answer engine, how Razorfish is merging paid media and organic teams, and why the entire funnel may be getting redefined. We also impact the concept of 'the correction', how conversational AI is resetting both consumer behavior and the KPIs marketers have relied on for decades. And we get tactical. What to fix in your product feed, how to audit your AI visibility,
and why you might need to build a search strategy for an audience of one. Amos became the AI guy at Razorfish almost by accident. Now he's helping clients see around the corner into a future where brand visibility, attribution, and even consumer trust are being re-imagined by large language models. Let's get into it.
Jon Clark (01:20)
to another exciting episode of the Page 2 Podcast. I'm your host, Jon Clark, and as always joined by my partner in crime at Moving Traffic Media, Joe DeVita. And today we're excited to welcome Amos Ductan SVP of Search at Razorfish. Welcome.
Joe DeVita (01:28)
Hi
Amos Ductan (01:34)
Thank you, thanks for having me on.
Jon Clark (01:36)
Absolutely. So I was at Razorfish for about five years. Joe made it a little over 10. You've been there about 15 years, which is a lifetime in agency world. What has kept you at Razorfish for so long?
Amos Ductan (01:48)
Yeah, I would say it really, it really allows you to have an entrepreneurial spirit and like focus on the things that you want to focus on. Like, so I've had the, you know, just a fortune of being able to not just work on different clients, but like almost different passion projects. Like AI is an interesting one where I, my son was born in November, 2022, right when ChatGPT came out. So I was on parental leave. I came back. I´m like, we must be doing a ton of stuff with AI. We weren't.
So like I became the AI guy, right? Like in the, no one, I didn't have to ask permission to do that. So it's that, it's the kind of place that just kind of lets you add value where you can. And so I've always appreciated that. Like I've been able to kind of reinvent the things that I'm doing, which is awesome.
Joe DeVita (02:15)
Thank
Jon Clark (02:31)
Yeah, that was always something I appreciated about my time there as well.
I think maybe that's a perfect segue. So there was a recent webinar, I think it was, where you sort of took, or I guess presented an AI search playbook. And you talked a little bit about the ChatGPT, I think it was processing about two and a half billion prompts per day, three million website visits being generated, and how it's a little bit of just a totally different type of search.
And I was curious of the data that you're starting to analyze and see come through for clients and probably even the agency itself. What types of search behaviors, maybe different search behaviors are you starting to identify with this sort of new channel?
Amos Ductan (03:15)
Yeah, it's it's a great question. I think it kind of, really gets to the heart of it that like AI is a lot more than just a search engine. I think ChatGPT just released some data or there was a study where they use ChatGPT data where I think it found that like basically half of the prompts are people like asking for it to do things versus seeking information the way you would for a search engine. So I think that's the interesting thing where you're just kind of compressing.
Like you're compressing that space between intent and action where the person is just like, hey, don't just, I'm not just looking for a hotel also build me an itinerary as well. Like that sort of thing. I think that's the really exciting thing. It's also the scarier thing because it feels like people get disintermediated, right? Like you're, you're having to browse less things, but I think that's the big thing. But I think the other piece is that people aren't just abandoning search or like traditional search.
You're doing it sort of side by side. I think there's a recent study that... or a SEMrush study that when someone starts using ChatGPT, their weekly sessions for Google actually increase, right? Cause it's like, it's almost like, you know, ChatGPT helps you discover stuff. Google lets you go deeper. And so we're definitely thinking about it in this kind of complimentary way. And that like some of the searches are actually, you know, somewhat similar, like, so it's not that it's not necessarily going to be a completely unique thing that people might be searching.
So it's a little bit ⁓ more nuanced.
Jon Clark (04:41)
Yeah, I saw that data point too. And I think there was at least earlier on when ChatGPT first came out, there was this mantra of, well, it's likely people are going to Google to validate what they're seeing in ChatGPT. Do you think that is the case? Or do you think it's more they're just expanding on or maybe drilling into something very specific in Google versus the validation? Cause I think
Amos Ductan (05:03)
It definitely
seems like it's, there's definitely a big piece of it is validation. I would say even just anecdotally myself, like part of it is also when you open the chat window within a ChatGPT, like let's say you're using it on mobile, it's just not the best experience, and so you might open it and see like, okay, I got to the right page. And then later you go to that same site, like on your, on your machine, like you might Google to find that same site, like, you know, navigation only.
You're kind of using them both in the same way sometimes and sometimes an interesting kind of mix actually.
Jon Clark (05:36)
I think the validation piece often comes in when the URL that's provided is incorrect, which I've gotten a lot. And so you kind of have to go to Google to actually get the right URL once you sort of drill down into the answer that you want. The navigational path from GPT is often incorrect. And so you sort of have to, that's the validation piece that I've found a lot.
I've started to tend to trust the responses a little bit more over time. I think earlier on, it was pretty obvious when there was a hallucination. I think it's, ⁓ it's becoming harder and harder to detect, which is maybe a whole different problem. but, in the AI search playbook, you talk a lot about this concept of answer engine, which I really liked versus the traditional search engine. Can you, can you sort of take us through
Amos Ductan (06:05)
Right.
Jon Clark (06:20)
the key you see between those two platforms.
Amos Ductan (06:25)
Yeah. Yeah. And I think I just kind of think about my own behavior as well that like, when I go to Google, I tend to, or a traditional search engine, I tend to have a better sense of what I'm looking for already. I mean, I mostly use it for navigation now. And so I think an answer engine is something that people are using more generally more for like exploration, maybe comparison of different things. It's almost like I want to get something done.
So I'm going to look for the answer versus Google is like, I know exactly what I want. Like I want to go to a website or I'm looking for the weather. So it's almost like the Google use case is almost, ⁓ less complex in a way. And it's interesting because you see Google infusing like AI mode. They just launched the search live. It's like, they realize that people are using these things in a complimentary way. And they're trying to
Jon Clark (07:07)
Mm.
Amos Ductan (07:20)
make this traditional search experience more like the AI search experience. I think that's where it's going to get really interesting where they really just start to blend, right? Where like everything is an answer engine versus a sort of traditional engine. But yeah, I mean, the thing is that for a traditional search engine, it all started, like I want to say back in like 2014, 2015, when they started like surfacing content before you had to go to the website. Like I remember like sports scores, like
Jon Clark (07:32)
Right.
Amos Ductan (07:49)
What was the score for the game last night? You just have to click into NBA.com. They started surfacing that, the weather. So in a way, we already moved from using a search engine to find websites to finding information. So it's almost like this answer engine is an evolution of that where I'm not just looking for a really simple answer, like what was the sports score? It's something more complex, like I want to plan a trip or I want to figure out
Amos Ductan (08:14)
my camp for my daughter, which I was using an answer engine for. So yeah, it's almost like the complexity of what you're looking for will probably determine if you're using an answer engine versus like a search engine, which I don't know. It's almost like that's used for the last mile of the transaction in some ways.
Joe DeVita (08:31)
I think human behavior, like we're always looking for the quickest way to accomplish a task. ⁓ So like maybe it's Google who figures this out better than open AI, but whoever can like create that tool for us that allows us to get things done quicker, we're going to migrate to that tool.
Jon Clark (08:31)
Yeah, very true.
Amos Ductan (08:37)
Exactly, exactly.
Thanks,
Joe DeVita (08:54)
I will say like the changing consumer behavior is one that I find interesting because for...
20 years we've gotten used to seeing choices. We're searching for information. The search engine gives us choices to filter through. And we all have these mental filters as we go through that first page of results, say, not a source I want to trust. That's one I'm interested in. most people go through the first page, or least half of the first page of results. And that gives you a few seconds to kind of qualify the information choices
that you dive into, but you don't get that with like an answer engine or a conversational search. You just get one answer and you either trust it and move on.
Amos Ductan (09:41)
Well, it's
interesting though, because usually depending on what you ask, I'd say most of the time, unless you like really press it, it will give you multiple options, right? Like, but I would say they're more curated where it's like, you know, oh, here's a good option if this is your priority versus here's a good option if this is your priority. So it is definitely doing more of the filtering for you to your point, but it is, I don't know. It does feel like it is still giving you those choices. Whereas
there's a world where we start moving towards just like, don't even give me the choice, just buy the thing for me, which I think for some category of things that might make sense. Like the, like shampoo, you just want your ... Like you don't really care about going through the list. You're just like, go get the shampoo. So, so I do think that we're still in this like middle stage where AI is still kind of acting like a search engine. It's more curated.
But it's still like I feel like multiple times I'll have to like press it like Hey, what would your recommendation be almost like don't which I think for some things you'll want to You'll want it to just kind of make the choice for you because it's not that important versus something like a vacation I think you'll definitely want options like this idea that AI is just gonna book a vacation for you That doesn't make any sense at all because you're gonna want to see the pictures You're gonna want to like get a sense of like what does the pool look like?
Like, so it depends on the thing that you're going, but I think there's always going to be that back and forth. It's just going to do more of the work for you. So it's not 12 links you're looking through. It's three options. You're, you're, you're, scanning.
Joe DeVita (11:16)
So that's the changing consumer behavior. I think it'll be interesting for us all to watch in the coming months. With search, search for traditional search, you search for something, you got your results, you clicked into something, and then you did another search. And there was search on top of search until you got all the information that you wanted. But with...
with AI, you just, it's follow up prompts. And like, that's a consumer behavior that most people don't understand how to do yet. And in your webinar a week or so ago, you did that awesome live demo of like, help me find the credit card that's best for me. I want this, I want that. And you kept it like, it was such a fun live example and it worked out for you. You must have practiced that beforehand. It really worked out well for you. Like you kept interrupting
Amos Ductan (11:59)
Thank
Joe DeVita (12:04)
And it was like, it was so fun to watch, but like, don't think most people know to use AI like that yet. They just treat it like a search.
Amos Ductan (12:13)
It is
a skill that I think we're all going to need to develop in some ways, but it's interesting. The follow up prompt thing that you mentioned is that I've noticed it's a thing I want to say in the past six months. When you use it, like, you know, the regular like chatting, it generally will ask you, like, there's some kind of follow up at the end, like depends on what you're doing, but it'll say like, can I do this extra thing for you? Could I look into this other thing, which I think is just the model. Like this is ChatGPT. I think it's OpenAI trying to
keep you in the conversation and like, kind of keep it going. But I think you're absolutely right that like, it is a skill we're going to learn, have to learn. I kind of think of it the same way that like, there's this concept of internet literacy. There's going to, there's AI literacy, right? Like think about how, how important the internet is. Like if you can't use the internet, you're like locked out of an entire category of economic activity. Like just shut out. That's where AI is moving, I think. And like part of that is that skill to your point.
How do you like pressure test it, Like push it, you know? I think it's funny, I think it's a behavior we're gonna have to learn. It's a skill we're gonna have to kind of build. But I keep telling people, just talk to it like it's a person, right? Whether it's voice or over typing, but I think it's a skill we're gonna have to build up a muscle.
Jon Clark (13:28)
It's really interesting. I was listening, I think it was some other webinar and there was something around a stat like
going get it wrong, but maybe like 1 % of people have actually created the custom GPT. So if you are within that 1%, you could consider yourself an AI expert, right? Cause you're, you, you've done more than 99 % of the rest of the, you know, people using ChatGPT, which is kind of crazy. It is really going to be something that changes over time for sure. And something that we have to sort of learn how to rebuild it. I was thinking it was almost like.
Amos Ductan (13:43)
Okay.
Jon Clark (13:59)
In the early days of search where the most common terms were, you know, one or two keyword phrases, and then it sort of expanded to six and eight. And you sort of had the, were learning how to use Google, and get what they wanted better with these sort of longer queries. I was thinking when I was at Razorfish and we were all sort of working at Razorfish together, like voice was supposed to be like the next revolution of search, right?
And that never really materialized. I think AI is probably where that actually has the best chance. I'd love to get your thoughts on how you see voice playing a role in search in the future. And I don't know, are you even talking with your clients about this yet in preparation for that or?
Amos Ductan (14:42)
Yeah, yeah, totally.
I feel like in some ways I feel like I'm surprised it hasn't accelerated faster. Cause once, ⁓ I think ChatGPT voice advanced voice mode, once that came out, which I guess was probably like a year ago at this point, it was like, wow. This is what Siri and the Google Assistant and Alexa, this is what they were always meant to be like the back and forth, you can interrupt it. Like, like, so I, but to that point, the consumer behavior, I think
It's going to take a while for people to really even realize that that's a thing. So I feel like it's very small now, but it's clear that they're paying attention because Google just launched search live where you can talk to Google. I noticed just today actually, ChatGPT changed the interface because it used to be just like this weird blue dot that's glowing that you're when you're talking. Now it's like, you can actually see what the AI is saying and what like, so it's a little bit, it feels a little bit more interactive. I would say.
There are some clients who like, they feel like AI search is too small now just in general. And so it's like voice is a even smaller subset of that, if that makes sense. So I feel like there definitely is a sense from a lot of people that it's, you know, is it too I definitely feel like this is the time to start thinking about that. The idea, like the kind of detailed information that the AI will need to retrieve when it's having a voice conversation,
like, your content has to be ready for that. I feel like, right? Like I did a different presentation where I was doing an example of like trying to have it search for a hotel for me. And I was like, what time does the pool open? It went and searched and it's like, it opens at six. It had to find that specifically somewhere. It had to be easy to kind of retrieve. I just think of like all the things that AI can to it makes the most sense in terms of how to unlock that.
So I think that is coming. I just feel like it's gonna take some time with consumer behavior. As you were talking about that change, you made me think, I always use the example of people searching near me. That's something that people learn to do once they had a mobile device. But that still felt like it took a little time. So I think we're still really early innings, ultimately, and it's gonna take some time. Because it is like a really big shift. There's like this Wall Street firm that came up with this notion that
ChatGPT's launch is kind of like the Netscape launch, right? So like Netscape democratized the internet, made it so that you can point and click. ChatGPT made it so you can just talk to AI. But if you think about that, we're in like 1999, right? Like the equivalent, right? And so like, it's just, it's hard to even imagine like how AI will just be embedded in everything. But I feel like when I think about it that way, it's helpful. It's like, right.
Few people even had email addresses back then. So it makes sense. People are still like, they're still learning all of the things they can kind of unlock with this new technology in a way. So yeah, I think voice is coming, but it feels like we need, I don't know, maybe it's upgrading Siri. Like I feel like it needs some kind of delivery mechanism for people to start to engage more in some...
Jon Clark (17:43)
Yeah. I mean, it's crazy. My daughter, your kids, right. They don't even know what a rotary phone is, but that's what I grew up with. And so there's like this whole, you know, thinking about my daughter's world 10, 15 years from now, like she may not even know what a, what a Google is, right. It may all just all be AI related, which is pretty wild to think. Joe was talking a little bit about the process of traditional search, right? You do a query, you do another one,
you do another one, that eventually sort of get to your the AI side, right? When you do, I don't know, something around deep research or even a question around travel, right? It sort of goes out, fans out, if you will, does all those queries behind the scenes and then ultimately brings back a response. I'm really curious how you guys are thinking about applying like that fan out two page search or if you're doing that at all, right? So how do you...
Maybe I'll phrase it a little bit different way. Like for paid search traditionally, right? You would have like a keyword research exercise and you would, you know, build out a keyword list, you'd build campaigns around it, ad groups, add copy, etcetera. A lot of that's hidden in the AI model where you don't necessarily always get to see everything that is being fanned out and queried to pull the answer together. Are you trying to use any of that data to inform maybe ads that you're buying on the Google side? Or is that not
Is that sort of connection not necessarily makes sense at this point?
Amos Ductan (19:04)
Yeah, it's a great question because it really kind of gets to the heart of it that it's the same reason why we can't really track specific prompts because they're all going to be so unique. It's less about the specific keyword and more about the individual topic, the thing that we're trying to cover. And so we're trying to think about our page search structures in that context. So we're certainly...
live and performance max, but like for the tech side, we still have campaigns and ad groups, but we're thinking about like a structure on top of that where we've been calling intent clusters. We're like, imagine multiple ad groups really kind of roll up into this one bigger cluster, which could be multiple prompts, which could be, any number of prompts and trying to make sure that we're covering all of the like topic surface area, like with...
And that includes ⁓ broad match is something that like, I feel like it's funny, like Google should have probably rebranded it. Because they relaunched it, I want to say maybe three years ago and it's doing really well at like capturing that kind of long tail. But we start with like a good base to make sure, you know, we're covering the topics like at a keyword level as best as possible and then let broad match sort of match us out. And of course, look at the search query reports to kind of see what we're matching out to.
But it really is almost like imposing this structure on top of like the way traditional search kind of has to live and still in this like campaigns and ad groups world in some ways.
Jon Clark (20:33)
Yeah, that makes sense. Are you using a lot of, I guess, Google's automation to help close those gaps, or are you still trying to maintain some control around what you're actually buying? ⁓
Amos Ductan (20:45)
Definitely.
I would say definitely, I say it's a balanced approach. We are definitely leaning way more into automation than we were before. And part of that is just it literally, it works better, right? Like I think broad match is a great example. Like the, used to match you out to things that were not relevant. It still does. You still have to police it, but it's a lot better. Performance max is better. But I think where the things matter, we're still taking human control. Like we still segment things in some cases where.
If we look at the data, we can see Google isn't making the most of it. I think the greatest example is you can segment by household income, like in deciles. So you can segment top 10%. It's like not rocket science. The top 10 % household income is going to be really valuable for lots of clients. And what we tend to see is that segment tends to be under like, like this, like the share of voice is lower than it should be unless you break it out and force it. Like, so we're still seeing cases where the algorithm isn't like as good as it could be. It's not catching all of the blind spots, but I'd say we're, definitely leaning very heavily into automation where it's doing the right thing, right? With like a, I almost kind of think of it as like the difference between driving a car and like, steering a ship, right? Like we're guiding the direction, not necessarily making all the fine grain sort of turns, if you will.
Jon Clark (22:05)
And I think that's the perfect example of why automation cannot always be the perfect application, mainly because that top percentage is just a smaller group. And so just by nature of the algorithm, it's going to go after something that's easier to identify, which are maybe those people in that middle financial tier.
Yeah, it's a smart strategy to break out and target that segment directly. I was curious how you're thinking about shopping feeds and maybe more directly, do you have any clients Perplexity or ChatGPT shopping feature yet? know in the AI search playbook webinar, you suggested everyone submit, even though it's not available to everyone yet. I didn't know if anything was actually running there yet or not.
Amos Ductan (22:46)
Yeah. So, so
there, so the ChatGPT one is just a wait list at this point. So there, there is no integration, but for Perplexity, we didn't, we don't have any clients that have submitted yet and they have actually, like, they're not taking any additional submissions at this point, interestingly enough. But what I say is that like the most important thing seems to be the connection between
what the AI is seeing in the feed and like basically what's on your site, what's like public facing. And you can see this in the thinking traces. It will like when you ask it for something, if it identifies the thing in the feed, it will look for that same page, like just so that they can provide the URL to the person. And so the, it feels like one of the biggest things you need to do is make sure that those two things are as consistent as possible,
the attributes in the feed, that the there's a corresponding like product detail page, for instance, that like, like there are that lines up like one to one so that it can make that connection very easily and surface up that that content. So, so I don't, so I think submitting the feed is almost less important than making sure that like whatever feed that you do have that you're using with Google, for example, is consistent. Cause at some point
I think it's gonna be table stakes, because it's clear that ChatGPT is gonna open up. So you're gonna be able to submit your feed. So I feel like now is the time to make sure that it's clean. You have it all filled out. You're adding all the attributes that you can. So you're gonna give as much detail to the AI as possible.
Jon Clark (24:20)
Yeah, I think their announcement of their relationship with Shopify sort of solidified that something's coming. Not necessarily sure what that's going to look like, but I'm sure it'll be impactful. I was curious, you talked about the relationship between the product page and the feed, which makes a ton of sense, right? You're trying to validate that trust essentially with what you're putting out there and it's actually live on the site. I was wondering if...
Amos Ductan (24:38)
Right.
Jon Clark (24:41)
when you've done any of that analysis, if there are any specific product attributes that you've found, I don't know, carry a little bit more weight or that the assistant prefers to see, obviously price and skew, but like anything outside of those that you've seen, I don't know, maybe a disproportionate weight put toward in terms of returning those products.
Amos Ductan (25:03)
I think the key is that it's going to be customized to what the priority of the person, whoever is putting the prompt in and what they're looking for. So I would say the key is describing the product in natural language so that it can be easily referenced like, ⁓ this is good for people with dry skin. It just says that and it's like, this prompt is from someone with dry skin. I know, I think.
That is like one of the biggest things. And it's funny, that's one of the things we've been doing is taking a harder look at like the descriptions, like how are they written. And it's like, in a lot of cases, there's not a lot of detail. Like there's not, it's like, it's kind of fluff as opposed to like, this is an item that human beings use this way. I you obviously wouldn't write it like that, but that's kind of how you have to think. Like you're kind of literally spelling it out for the AI so that it can read that and say,
oh, that corresponds to what that person wants. So I'm going to surface this up. It's a different way of thinking. And there are companies that actually are helping you people rewrite that stuff, like at scale, obviously using AI. It's funny, AI on top of AI.
Jon Clark (26:07)
Great.
Joe DeVita (26:09)
So you, you, you were like the AI guy. You said you came back from paternity leave and like no one raised their hand to kind of jump into this and start talking to clients about it. Are you thinking about how to build teams around it yet?
Amos Ductan (26:22)
It's a good question. mean, at this point, it's definitely come in. Like the work that we're doing has definitely been essentially SE... like essentially adjacent to SEO. Like, we're going to do an sort of AI audit. So I'd say we're still kind of early days, but I feel like, you know, there are these like upstart AI ad networks. So I do see a world where, you know, like, I don't know, in some ways, AI is already infused within all the channels, like within programmatic.
But I do see a world where they're like, you know, once, example, open AI monetizes, I don't know. I think it, my view is it, it brings, it probably most likely just brings search closer together, paid and organic, right? Like, like it's almost like right now, many companies are looking are like traditional search itself is like a silo, right? They're looking at paid and they're looking at organic. And then there's AI search answer engines.
What you have to do is make sure you're like the time now is to like look at traditional search holistically. And then you're going to have to combine that with AI search. So my, my thinking is like, it's more about like personally for the paid search team, we're learning more about how SEO works, for example, like semantic relevance, like how is that actually computed? Like I'm thinking it's some kind of like search hybrid type thing is my, is my guess.
I it's unclear what shape that takes. Again, it also depends on how these platforms decide to monetize. It seems like OpenAI might do some kind of like affiliate like model. So then I think it, you know, in some ways that's a little bit of a different skillset than a paid search person or, or an organic person. So I think it, it can go a few different paths, but I think the most obvious first is like paid and organic kind of coming together, like the query fan out.
Like explaining to the team, those fan outs are based on things that are semantically relevant. And like, we can figure out what those things are. Like that's stuff that we hadn't, didn't have to think about as a paid search team as much. And now I feel like, yeah, it's like, I don't think SEOs have to learn much about paid, but paid search people have to learn about SEO. Cause that's kind of the foundation.
Jon Clark (28:06)
Okay.
Joe DeVita (28:26)
guys remember when shopping comparison engines were like such a big deal for any e-commerce strategy? It was like shopping.com and there were a bunch of, there were a bunch of like feed based advertising, self-service advertising platforms that you had to consider if you were, if you were in e-commerce. And when we all worked together, there was a team that sat between the paid search and the SEO team. They were like the feeds team, like all they focused
Amos Ductan (28:32)
Yes.
Jon Clark (28:34)
Thank
Amos Ductan (28:36)
Right.
Joe DeVita (28:55)
on was e-commerce clients who had big shopping feeds and their job was to like optimize the feed, make sure the feeds got imported correctly, and I don't know if they managed the campaigns and bids and stuff, maybe that was the paid search team, but like there was this really specialized e-commerce team that just like optimized product feeds. I remember that's the, maybe that's the progression.
Amos Ductan (29:19)
Yeah. And
there definitely is still some of that, right? Like we actually have some places cases where it's like shopping is like its own separate thing, has its own budget, its own team. There is still some notion of that as well, but it's interesting. I, yeah, I could see a world where, yeah, it's like some kind of carve out where it's like, very specialized.
Jon Clark (29:41)
I mean, maybe to stick on that subject a little bit, when you're hiring today, there, like in knowing that this sort of, I don't know, shift or move is coming, are there different attributes that you're looking at when you're hiring now versus, I don't know, even six months ago?
Amos Ductan (29:50)
you
Yeah, I guess I would say more connected to like the general like path that we're on, like not necessarily specific to AI search, but we have definitely been trying to move like as a team from like what I call operators to orchestrators. So I think historically we've spent a lot of time and there's been some pride in like being able to figure out like the plumbing and like the mechanics like,
I'm proud of being able to bang out this Excel project in three hours or four hours. And it's like, we don't necessarily have to do that anymore, right? Like we literally live in a world now where we can talk to software and have it create software for us to do that middle work. And then we extract the insights with our human brains. And so it's like the, the thinking, like it's like a new way of thinking that we're
Jon Clark (30:25)
Okay.
Amos Ductan (30:53)
adopting
Jon Clark (30:53)
Listen.
Amos Ductan (30:53)
as a team. And I don't think everybody is there yet. I think part of it is having a growth mindset of like, just being willing to like, try new things. And so like, that's probably one of the biggest things that we're sort of screening for. It's like, we're going through a period of a lot of change. And so if you're a person who isn't changing a lot, you have to, like, that's not a good sign, right? It's like, why aren't you right? So
And being curious about AI search, like leaning in, I think that that's all, those are all kind of bonuses, but I feel like we're now in an age where we can do more than ever. And so we've got to really like lean into that. Obviously, you know, translate that into value for clients is at the end of the day.
Joe DeVita (31:38)
I just a comment there on.
I think John's question was a little bit leading because we've had to change the way that we look to hire people. When we started eight years ago, like we had a lot of set process in place and we felt really strong that we could just bring in a smart kid, young adult professional to like follow our process and execute. That's what we needed. But now it's like, we're not looking for people who can just take direction and execute really well. We kind of like looking for people who can make it up as
they go and solve things that they've never seen before. So it's kind of a different professional.
Amos Ductan (32:15)
Totally a problem. Yeah solving problems. Yeah, like like that's the uber skill in a way, right?
Jon Clark (32:20)
Yes. Gotcha.
Joe DeVita (32:21)
I want to bring us, I kind of want to bring us back. I read you, you, ⁓ you blogged a few weeks ago about, the emergence or the, the quick adaption of people using AI search, conversational search, as a correction for search in general. And I really love the concept. I couldn't stop thinking about it. That's why we were so excited to get you on this podcast today, but, maybe I'll paraphrase a little and you tell me if I'm way off because
I think it's a concept that I think I've only really read you write about. But we've, you for years we've been trained on how to use search engines. We don't see it on the first page, must mean we searched wrong. I'm gonna search again and I'm gonna filter through page after page, search after search to get what I need. And on the marketer's side, it's like.
you got used to these metrics like okay, you know, click through rate on organic search of 30%. That's pretty good. A click through rate on paid search of 5%. That's pretty good. A conversion rate on organic search of X % is good. So we've gotten used to like this behavior. We've got trained people, consumers are trained on this behavior and marketers are trained on like this expectation of metrics. But that all kind of gets blown up right now where
Whereas like conversational search with an AI engine just makes it a little bit better. And the metrics have to change. Is that the correction? Am I paraphrasing?
Amos Ductan (33:49)
Yeah, I think that's,
I think that's exactly it. mean, what you're describing, it's a, that's a really good way to sort of think about it. It's like, there is this process, right. And as marketers, we're like, we've inserted ourselves in it. Like, oh, great. Like the more they search, the more I can kind of get in front of them. And like, I can track this and track that. It's like, you think about the human, the person, I, I don't want to do all that stuff. Right. Like, I, like, I just want to go on vacation. I don't want to search through all the hotel websites.
Again, as I said, I do want to see the beautiful pictures. Like I'd love someone to just curate that for me. Like I don't want to go through the grunt work. And so that's the, like, I, I see this as like removing some of the friction, right? It's almost like this is how it was meant to be in the beginning, technology just wasn't there, right? feel like it's a little hyperbolic of like, what does a website really mean anymore?
I think you absolutely still need a website. But if you almost even think like not too long ago, everything you needed to know about a company, you had to go to their website. That was the source of it, I mean, not too long before that websites didn't exist, obviously, but that was like your source. Now you can expect to find information about companies all over the place, including like Reddit, but also AI. And so I think it's everything is just getting more convenient, right? Like
The the answers just come to people easily easy more seamlessly. And so it does make things more complicated for marketers. But I think there's a world where for really, really like elite brands, it's actually great, right? Because it's like if I love this brand, like if I love Nike, that's what I'm getting, right? Like I like the removing the friction of like the searching. That's not going to stop me from buying Nikes.
I'm going to use my convenient process to buy more Nikes. And like, that's obviously easier said than done, but like building that brand love is like more, more important than ever. The way these things are customized, they know your preferences. And so there's a world where it just keeps surfacing up the brands that you prefer. And like, you want to be one of those brands that many people are getting the AI to surface up, right? It's like,
The AI never gives you any recommendations that are not Nike sneakers, because that's what you told it that you want, right? I think we can get there. I think, that messy middle process, it's just not fun for people. And so it's, as much of it as possible is gonna go away. I don't think it's all gonna go away, but clearly it's a better experience. I think people are voting, like they're saying that they like this. I mean, it's kind of intuitive as well. So that's the way I see it.
Joe DeVita (36:20)
I think that makes complete sense for me and I think it's an easy concept on the consumer side, but for the marketer who has been for years optimizing for a 2 % conversion rate and a 10 % click-through rate, how do you, like, what does the model of success look like on this new platform?
Amos Ductan (36:42)
Yeah, it's a great question. So I guess ultimately ideally you're already looking at business metrics today, right? And like maybe you've built this great process so you can see exactly how your marketing metrics are moving your business metrics. So what we're talking about is like some of those marketing metrics going away, right? Like you're the visibility. You should still be able to see the business, right? Like
Are people buying your products and services? What is the trend there? And then it's a question observing that trend, how has that changed? And then it's a question of how do we kind of triage? I think one of the first steps is understanding your visibility in AI search, for example, right? Your AI search share of voice, that sort of thing. I think the LLM models, like we're seeing more and more clients lean into that sort of thing. I think you have to find a way to plug those gaps.
I think, qualitative data becomes more important, like the survey thing. In some ways, it's almost like we're going old school where, yeah, as more data gets removed, you're having to sort of pivot. But I think it does go back to those business metrics, right? If you were always only looking at marketing, kind of like in a vacuum and not connecting it to business metrics, now all of a sudden you're like, I'm flying blind. You should have never been looking at clicks in a vacuum in the first place, if that makes sense, right?
Is the way I kind of think about it. So like instead of a like, like, so yes, you're definitely losing some visibility, but theoretically you can still see the bigger picture. Like sales are still working. I need to figure out how to close these gaps in my marketing as opposed to like, no, it's a, it's like a catastrophe because I can't see anything.
Joe DeVita (38:22)
What are some of the tough questions? Where are clients pushing back on you when you go in and present like, here's this great new opportunity and you got to start investing today so you can understand it for tomorrow when it's more important. What's the pushback that you're getting?
Amos Ductan (38:38)
I would say there's a, there's a few categories. One category is this idea it's not like there's nothing actionable that we can do, which is just really not true. If let's say that AI is saying things about you that is not accurate, you can report that, right? OpenAI, Google, they will do what they can to remediate that because it's in their interest. They don't want AI, like, and so that's a perfect example of like
for one, maybe you probably want to know is the AI saying the right stuff about you. So like, and that's actionable. So I think that's just one category. People have a little bit of a tough time wrapping their head around like, it's like, you know, what do I do with it? The other, it's somewhat like sometimes, you know, correlated is like, oh, this is just, it's just so small. It's just so small. Again, I go back to that idea of the Netscape moment. Like the internet was really small in 1999, right? It was a thing that mostly weirdos were on. If someone told you,
you would be booking a cab from your phone or even just the internet, forget your phone. You would think they were weirdo, right? And you'd say like, what are you talking about? So think there's a little bit of that we need to get over. I think as the numbers get bigger, it's getting harder to ignore. Like, ⁓ there's literally like almost 3 billion prompts a day. The other category where people like, they're like, okay, I get it. This is a thing. Like people are seeing click volume drops
on the organic side, for example, right? I think in a lot of cases and almost all of those cases, like sales, like products, like those are still move, like those are still like at goal, at or above goal, like we've seen no degradation. And so there's still this notion that like, it's, you know, it makes sense. There was data I was getting before there were people I was interacting with before that I'm not now. I mean, to be fair, there's also the question of like the knock on effects later, but like,
having that person come to your site and consume your content. So it's definitely not that we're losing zero. There may be some loss. I think that so far, I'd say those have been kind of the categories. Where clients have come to start to embrace the idea, I think, is that we can see at scale where we're showing up. And so at the very least,
I think it resonates with them when we can say, this is how you're showing up in this category and ⁓ these are the competitors that are also showing up alongside you. And so I think that kind of helps the conversation that like, that's at least data that we can bring to bear. But I would say it definitely tends to be kind of feast or famine that like either someone is like really leaned into this or they're like, no, it's too small for me to kind of pay attention to.
it's funny, like the conversations can kind of go down those two kind of general paths of like broad skepticism versus like the nuance of like, how do I think about it? I'm nervous about click loss, etcetera.
Jon Clark (41:24)
I was taking notes as you were talking and you sort of answered my questions as you were going. One was around
There's so much more information out there, Reddit, other content places. while that is convenient, I think the risk would be the conflicting information. And so then you get the incorrect response coming out of the GPT. And my question was going to be, you had to encounter that? sounds like you answered that along the way, which is there's ways to report that.
I don't know if you've had any case studies on, I don't know, any situations where you've had to create net new content, almost like a reputation management type thing to circumvent those things. The other one was, which was really smart was, you if you've been looking at clicks in a vacuum like that was probably the wrong way to be evaluating your performance anyway. So as clicks go away, as long as your business KPIs are upheld, then, you know,
Amos Ductan (42:10)
you
Jon Clark (42:15)
it does clicks matter anyway. And so maybe the question that I'll land on is how are you helping people, clients sort of evaluate how they're showing up in these models? Cause I think that's a, there's so many tools out there. Even some of the legacy enterprise tools are trying to spin up like, you know, AI tracking and things like that. Some of them are okay. Some of them are suspect.
So what are you guys thinking about or what sort of tools are you evaluating at Razorfish where you would feel comfortable saying, you know, hey, client, this is what we would recommend for monitoring prompts and, you know, trying to identify those things that you might need to mitigate in the future.
Amos Ductan (42:56)
Yeah, it's a great question because I feel like so many of those companies have just sprung up that didn't exist like two years ago. And to your point, the existing legacy players are integrating it as well. And so it's really a question of A, what tool to go with, and B, how to use it. Because there are so many different ways. So it does feel like Profound does have somewhat of an edge.
They have the largest panel. They are one of the AI search providers. But in some ways, the tool is almost a little bit less important than the strategy, because as long as it will give you the volume of props that you need and potentially the personas as well. But the way that we've been thinking about it is, and this is another way to sort of bring paid and, or call it traditional search and AI search together, is
Let's think about all the places we're showing up in traditional search. So pulling like so practical terms, pulling a search, like a report, like a Google search console that shows organic and paid. like where are all the places that we showed up and then clustering those into those intent clusters that we talked about and then figuring out, okay, what is a representative prompt for each of those intent clusters? And then also depends on the client. We might want to put a persona on it. Right. So like,
If it's travel, it's might be the location plus this is a family traveling versus it's a couple. Then we understand when to import those into the AI that get the responses. And then we can understand where are we showing up? Like are we showing up in the same places that we're showing up in paid search? So for example, maybe an intent cluster for paid search, we have really high coverage. We showed up a lot. Our impression share is high.
Does that, is that comparable in AI search when we put in those representative prompts? And you know, we, what we found in some cases are there are sometimes there are gaps, right? There are places where we're really strong in page search, less visibility in AI. And we found the converse as well, where especially in unbranded, we don't have a strong presence in unbranded, but AI is organically recommending
is a hint that we might have a right to compete there, because we're literally, especially thinking about these people are using these things side by side. I don't think they're using them in the vacuum. So you might see the recommendation list on ChatGPT, then you're on Google next or what have you. And so I think that helps to give you a good foundation to start with, because you could put any number of prompts and there's like so many directions you can go in. This is very tangible. It's like, this is where I'm already showing up.
This is the biggest platform where people are used, like people come to express their intent, right? Search. And so they're probably searching or prompting similar things when they're seeking information in chat is kind of the idea. And so as long as a tool can get you that, the prompt responses, like that's probably the primary thing you need. One new tool that just launched, it's like brand new. They create what's called an AI twin.
Jon Clark (45:51)
Okay.
Amos Ductan (45:51)
And so you can create an AI twin
that like, let's say simulates a Gen Z traveler, right? You it's basically a ChatGPT version that's loaded up with deterministic data. So it then behaves like a Gen Z person would, right? Like you ask it, what's your opinion on this? And so that's the underlying technology. What they do is they ask that AI twin to rank 10 brands. They, like, let's say, let's stick with the travel example and say,
If you were considering a Caribbean resort vacation, how would you rank these 10 brands? So the AI twin ranks the brands. They then put in a prompt that is essentially about like, you know, tell me about the best resorts for a Caribbean vacation. Put those into Gemini, ChatGPT, etcetera, and then show each AI response to the AI twin. If you guys are following. So now the AI twin sees what
Gemini recommends for resorts in the Caribbean. And then it re-ranks those brands. And then you can see the difference in the ranking. And so it's like, in some cases, maybe your brand was ranked number two to start, like baseline. But after being exposed to the Gemini prompt, now it's ranked number four. But the ChatGPT rank prompt kept the ranking the same. And then the most interesting thing is closing the loop.
You can ask the AI twin why the ranking changed. Sometimes it's as simple as, you know, the, the, the AI model provided the brand, like maybe it presented the brand as a third option versus the first, or maybe it said, you know, the, brand was more expensive. And so I think that's where it gets very actionable where you could potentially understand. What are the things I could potentially, I could change now this all hinges on.
Like believing that that AI twin is a good proxy for like a person. And this company has, you they've done studies like with focus groups, etcetera, but it's an interesting idea. They've turned it in. Answer engine influence report, right? The idea is we don't just need to show up in answer engines. We need to also understand how they're influencing people. Like, so, so it's an interesting concept. It kind of gives you a sense of like where things might be going. I see a world where
we probably see more like focus groups and things like that where you're asking people, you know, your last day interaction with an AI, how did you feel about this brand or that brand? But yeah, that's some of the ways we're kind of counseling our clients on tap.
Joe DeVita (48:17)
We're gonna run out of time with you any second, Amos. I just wanna double down on the measuring success piece because it feels like the just showing up is taking us back decades. We've moved so far away from, well, we can count the impressions and we can show you how many million impressions our efforts produced.
We've come so far from the impression being an important metric, it feels like we're gonna go back there.
Amos Ductan (48:44)
Well, I guess for the way I see that though is if like we're still going to have like the trackable stuff is still happening, right? And like, I guess it's a question of how big is that going to remain, right? Cause right now it's obviously gigantic, right? Most of the things people are doing, we are still tracking, right? And I see a world where it will still be a sizable portion, right? Like I don't think, and again, think about how consumer behavior changes. like, I almost think of it as like,
some of your lower funnel media is now changing to upper funnel, but it doesn't mean all of lower funnel is going away. It's a question of degrees in some ways, right? Like, like some things we were able to measure with more precision and we can't anymore, but it's not all shot in some ways. It's not, it is my sense. I also see a world where these AI providers, they're going to share more data,
right? Like Google certainly will once AI mode reaches like critical mass, I have to imagine they're going to start sharing more data. So I think we're going to have a better sense of like, you know, maybe it's like, maybe there's some proxy for click through rate. Like I think there's more metrics will emerge as like we, you know, in the fullness of time is my view. I just think like we're super early right now. So like, it's not fair to judge where we are metrics wise based on how
it is, but I hear you, it's definitely very watered down right from what like the robust things were used to right
Jon Clark (50:13)
Maybe real quickly, just you mentioned AI mode. Are you running ads in AI mode? Real quickly, maybe just what have you, what have your first impressions been based on performance and transparency?
Amos Ductan (50:19)
Yeah.
Yeah. So right now Google isn't
giving any, any visibility into like when you show up in AI mode, we know that ads do show up in AI mode. They're very, very few and far between. It's like one of those things where like you're lucky if you can catch one in the wild. So right now we don't really know at this point. What we do know is that using things like AI max and performance max is going to give us like the best chance. So we're definitely leaning in there and we are seeing strong performance
from those sort of products and features. But yeah, right now they're just not really sharing a lot of data. And I think they're gonna have to once they reach a critical mass.
Jon Clark (51:02)
Yeah, I'm really excited to see what they decide to share because I think Google historically has been sort of taking away data. So I hope they lean into being more transparent than that.
Amos Ductan (51:14)
I mean,
the amount that they're sharing about Performance Max is, mean, it's like night and day. It's crazy. Like you just, you can see all the conversions by channel. It's like, wow. Like you had all this data all along, of course.
Jon Clark (51:19)
It's gotten much better.
Joe DeVita (51:25)
We got to give a presentation on Performance Max every three months because like the reporting changes, the targeting changes.
Jon Clark (51:26)
Yeah.
Amos Ductan (51:31)
Right.
It's funny, like people will say like, it's a black box. It's a black box. Like, no, you're not talking about the same. Yeah. So I think I hope that's a, like a sign of things to come with like AI mode and like, like they'll be more transparent, but we'll see.
Jon Clark (51:36)
Not anymore. Yeah.
Joe DeVita (51:36)
Really? How many more?
Jon Clark (51:46)
All right, as we wrap up, maybe we'll jump to some quick rapid fire questions if that works for you. All right. Is there a Razorfish colleague you think you can beat in Jeopardy?
Amos Ductan (51:51)
Yeah, absolutely.
Oh, I would say pay on.
Jon Clark (51:59)
You can imagine where we may have sourced that question from. Best pizza in West Orange.
Amos Ductan (52:01)
Iron.
Sabatino's in South Orange. So it's not in West Orange. but everything is like five to ten minutes away.
Jon Clark (52:06)
Nice.
Okay. Is there any song better than Miley Cyrus's Party in the USA?
Amos Ductan (52:15)
That's a nice inside joke. Not for me, that's my go-to karaoke song.
Jon Clark (52:19)
Nice. I got to see a video of that one. It's awesome. One SEM rule that you still use that predates PMAX.
Amos Ductan (52:22)
Apparently it's floating around.
Well, I guess it depends on what we mean by a rule, we talked about before we're still segmenting things where it makes sense, where it strategically makes sense. I feel like it's just not there yet.
Jon Clark (52:40)
The best AI tool you're using personally.
Amos Ductan (52:43)
It would probably be storybook on Gemini. Like you can just create a custom story with like, like just a sentence long crop. I used it with my daughter. She was getting impatient on the train. So I created a story about a girl traveling on a train to go see her aunt or something. It was just, it's, it's pretty wild. So yeah, I I would recommend it on Gemini.
Jon Clark (53:02)
I use a ChatGPT to come up with jokes at bedtime. Pretty, pretty great. What's the best career advice you ever received?
Amos Ductan (53:09)
I would say from Joe DeVita actually, one that stuck with me is don't accept no from someone who doesn't have the power to say yes. I probably say that to someone on my team every couple of months. We talk about like a client. I'm like, are we taking a no from someone that doesn't actually have the power to say yes? Like, let's figure out a different way. Like this is too important to be blocked by this person.
Jon Clark (53:13)
I don't believe it.
Amen.
Joe DeVita (53:17)
Yeah, well...
I heard Jim Warner say that. I've passed that advice to other people. ⁓
Amos Ductan (53:33)
Well, the best advice is passed down, right? So
it doesn't have to be original. Appreciate that.
Jon Clark (53:39)
Yeah,
that's a great Last one, a podcast you'd recommend. Could be business, could be whatever.
Amos Ductan (53:45)
I would
say Bankless. It's a crypto focused podcast. So obviously you have to dive down the rabbit hole of crypto, they talk about so many things that crypto is adjacent to so many things. It's actually really kind of fascinating, like the Renaissance. it's just interesting. So I would recommend that Bankless.
Jon Clark (53:51)
Okay.
I'll definitely check that one out. Amos, this has been awesome. Before we let you go, ⁓ where can folks find you online?
Amos Ductan (54:14)
Yeah, so I'm on LinkedIn. You can just search my name, Amos Ductin. That's probably the best place to find me.
Jon Clark (54:20)
Perfect. Last question, we'd like to ask sort of a prediction question as we wrap things up. If you were to go to Google.com, let's say 12, 18 months from now, what do you expect that experience to look like?
Amos Ductan (54:33)
That's a really good question. I expected to have some version of AI mode that I don't have to trigger. Because right now, you have to deliberately go into AI mode. I suspect we will be at a point where AI mode, and I don't know that it'll be just the full AI mode that exists now, but it'll just be more seamless.
Amos Ductan (54:55)
is my sense. Like that's my guess.
Jon Clark (54:58)
Okay, perfect. All right, everyone, this has been another exciting episode. Amos, thanks so much for joining us. It's always great to reconnect with
with old friends. And if you liked the show, please remember to rate, subscribe and review.
talk to you soon. Bye bye.
Amos Ductan (55:09)
Thanks for having me on.
Jon Clark (55:11)
talk to you soon. Bye bye.
Amos Ductan (55:11)
Thanks for having me on.