AI Transparency in Advertising What Gen Z Executives and Regulators Get Wrong

**Introduction to AI Insiders & Sam Khoury**
(0:05) Hello everyone, I'm David Berkowitz, welcome to another edition of AI Insiders by AI Marketers Guild from Marchitecture. We've an incredible session today with our IAB friends coming back for another round and Caroline Gilbert assembling another all-star session. Before I introduce one of my Marchitecture friends and colleagues, Sam Khoury, to share

**Sam Cury on Marchitecture's Upcoming Conference**
(0:31) a couple things about what's coming from Marchitecture next month. Sam, take it away. Thank you, David. Hi everyone, my name is Sam. I'm the chief strategy officer of Marchitecture. Focused on diverse content at the conference, heavily focused on its distribution and scale.

**Conference Details & Target Audience**
(0:53) distribute and scale it. To give you background on my focus, I wanted to present a couple of things today regarding the conference coming up on March 10th and 11th. This is a sales pitch, but take it as it is. Since everyone here is heavily focused on AI,

**AI's Impact on Marketing & Technology**
(1:14) we all in this room or on this call understand its impact on us as marketers, employees, technology providers. We know it will play a big role, and many of us are using it regularly, if not daily. We're addressing many concerns at the conference on March 10th, primarily how it plays a role in technology

**Conference Topics & Free Access**
(1:38) evolution and creation. Focusing on its impacts on publishing and what it means for better data utilization. I wanted to let you know it's happening March 10th and 11th. I'd love all of you to be there. Quick note, it's complimentary if you are a brand or an agency. David has some

**Discount Codes & Keynote Speakers**
(2:04) amazing codes he can provide to the community to discount it further. >> Great to have you here. >> Salesy, David. >> No, no, no. It's a great heads-up. I'll quickly ask you the direct question: Who are a couple of

**Notable Speakers at the Conference**
(2:25) the speakers you're most looking for? We have the commissioner of the FTC speaking. That's really interesting because it's not every day he sits on a stage and talks about policy and changes. That's huge. We have all the holding companies speaking as of right now. Every single holding company has representation at the conference. We

**Top Speaker Picks**
(2:49) have the chief marketing officer of Molson Coors, the CEO of People. For me, FTC is number one. The CMO of the NFL is probably number two, especially with the Super Bowl just ending last week. >> My number three is I really like Lipman. I don't know if you know him, David, but he's really

**Andrew Lipman's Unfiltered Session**
(3:11) unfiltered. I like that content. He has a session called 'Oversold, Overhyped, Overrated.' >> Okay. >> He's not holding back. We've seen what he talks about. There's no BS with this guy, and I love it. This is Andrew Lipman. >> Andrew Lipman. Yes.

**Diverse Conference Content**
(3:31) >> He's great. >> Fantastic. Every time he speaks I'm like, 'Oh my gosh, he said everything I've been thinking. Thank you.' >> I think overall it's pretty diverse. So if you're from an agency, brand, or publisher, it'll be valuable. This is very different than our last two conferences, which were

**Scaling the Conference**
(3:49) very ad tech heavy. >> I think we maxed out on that. We had 450 people, sold out a month in advance. This is a thousand people, two days. We've partnered with Ad Week for content tracks. We've partnered with TV Rev. >> Variety is also a partner.

**Broadening Reach Beyond AdTech**
(4:10) you'll be seeing an email from Variety coming out probably in the next two days. Generally, the content is diverse and intentional. The adtech core is there, but we want to reach beyond it and address everyone. >> Awesome. This is great. I'll add that the other thing I'm really

**Looking Forward to Catering & AI Ad Disclosure**
(4:32) looking forward to is Russ and Daughter's catering. >> Yes. >> We can get into all the details, but I don't want to short-change our panel because we have a lot to talk about regarding disclosure and AI ads, including one of my favorite newsletter writers on the subject, Debbie Williamson.

**Transition to IAB Panel**
(4:54) I'll head over now. Thank you, Sam. This is awesome. Glad to give people a taste of what's going on beyond our AMG walls and what's part of the bigger picture here. Hope to see many of you from this chat in person there. Caroline, who do we have then? We have IAB coming back. A reminder for everyone

**Community Conversation & Caroline Gigerick**
(5:19) or for those joining us for the first time, this is a community conversation. Everyone's expecting great questions and participation. I get to sit back and listen today as Caroline Gri brought together incredible thinkers and doers across the industry to talk about a really important issue. Caroline, the floor is

**Caroline Gigerick's Introduction & Call for Interaction**
(5:40) all yours. >> First of all, thank you, David, as always, for having us. It's great to be in these interactive conversations. As you just said, if you want to put comments or questions in chat, we'll try to go directly to them in thread, off mute, or whatever you are comfortable with. My name is Caroline

**Role of IAB in Advertising**
(6:01) Gigerick. I'm VP AI at IAB, the Interactive Advertising Bureau. If you're unfamiliar, I call it Switzerland among the advertising industry. We sit between publishers, agencies, brands, tech platforms, and we coordinate between many different goals and constituencies. I think we do a fantastic job. I'm going to bring up to the very

**AI Transparency & Disclosure Research**
(6:28) beginning a conversation where we show you a few slides from research with the talented Debbie Ao Williamson. She is founder and chief analyst of Sonata Insights, done in collaboration with IAB. This is on the topic of AI transparency and disclosure. We'll share a few insights, then show you a few insights from the framework we then

**Framework for AI Disclosure**
(6:53) cultivated to answer the questions of why, when, how, and who is responsible for disclosing AI use in advertising. Debbie, I'll ask you to unmute, and I'll share this presentation and let you take over. >> Sounds good. Thank you so much. I see a lot of familiar faces. I actually spoke, David, at one of the events probably

**Debbie Williamson's Background**
(7:22) almost two years ago. >> You were early on. >> Great. >> You and I worked together back in the day at eMarketer. My background is I was an analyst at eMarketer for 19 years, and now I run my own independent research firm focused on AI

**The AI Ad Gap Study**
(7:39) and consumer behavior. This study looked at the gap, which we call the AI ad gap. This is the second study. We did one in late 2024. We just published new results in early 2026 from a study we ran late last year among Gen Z and millennial consumers. We also included a comparison study among advertising executives with media budgets ranging from 1 million up to

**Study Premise: AI Exposure & Consumer Feelings**
(8:05) above 1 billion. Caroline, please move to the next slide. A few things to set the stage: We went into this study thinking that with more exposure to AI-generated advertising and more awareness of AI use, younger consumers, Gen Z and millennial consumers, would feel more positively

**Widening Perception Gap on AI Ads**
(8:30) more negative towards the concept of AI-generated advertising, and the gap between how ad executives thought they felt versus how young consumers actually felt widened. That sets the tone for why what you did, Caroline, with the framework is so important, because these people are feeling strongly about AI and advertising, and closing that gap is

**Gen Z's Negative Perception of AI Ads**
(8:57) important. What's interesting, if we can move to the next slide, is that even the youngest consumers are feeling more negative and less likely to feel positive, using negative attributes to describe companies that use AI to generate advertising, terms like inauthentic, disconnected, or unethical. We see younger Gen Z consumers

**Gen Z vs. Millennial AI Perception**
(9:20) more likely to choose those terms. We also see millennials, on the other hand, more likely to feel positive. We probably discussed and unpacked that more in the presentation, but it's something to note that these younger consumers, who are very likely to use generative AI in their daily lives, are feeling more

**Importance of AI Usage Disclosure**
(9:38) negative towards it when an advertiser uses it. One important thing we dug into in the study is the idea of disclosing AI usage. There's much to unpack, which we'll discuss with the panel, Graham and Ken, in a few minutes. Leaving you with a few data points, one thing that

**Disclosure Drives Ad Engagement**
(9:59) we found is that disclosing AI usage can drive engagement for advertising. When we asked what would cause people to pay attention to generative AI ads, high quality makes sense. Funny ads are always interesting, no matter where you see them. Think about all the Super Bowl ads you might remember. Most of them

**Disclosure as a Trust Builder**
(10:22) are probably funny. But then this idea of disclosure ranked number three. This is something we see young people saying: maybe we feel more positively about it if we know we're not being fooled or the wool isn't being pulled over our eyes. If we

**Disclosure Boosts Purchase Likelihood**
(10:43) move on to the next slide, this is the money slide literally, because we found that if an ad generated with AI had a disclosure, for 72% of respondents it had no impact or increased their likelihood to purchase from that brand. Only about a quarter said it would have the impact of being less likely

**More Upside with Disclosure**
(11:08) to consider purchasing the brand. Overall, we see more upside than downside when advertising to Gen Z and millennial audiences with disclosure. >> I'm going to ask you one question, Debbie, from Tamika. Thank you for this, Tamika. Are there plans to survey Gen Xers or even older users to find out their POVs on disclosure?

**Future Research on Older Generations**
(11:29) Yes, that would be great. We focused on this audience because we knew these were people more likely to use generative AI in their daily lives; we see that in consumer surveys. I thought this would be the best audience to survey

**Value of Full Population Study**
(11:47) because these are people already very familiar with the concept of generative AI. But, Tamika, you are absolutely right that having a full population study would be super interesting. So, Caroline, let's consider that for later this year. >> Absolutely, and Michael, we'll get to that question during the panel discussion. Debbie, moving

**Key Takeaways: Widening Gap**
(12:06) to the key takeaways. >> Quickly, the key takeaways: The perception gap is widening. That's something we need to close. We didn't discuss this in the data slides, but it's in the full study available on the IAB site. One thing advertisers are doing that might

**Efficiency Over Creative Quality**
(12:23) be contributing to this is prioritizing efficiency. When we asked advertisers why they use AI in their creative process, they ranked efficiency and cost savings above creative quality. In my opinion, AI should be used to make ads better and more appealing to consumers, improving your creative, and that should be more

**Transparency and Consumer Perception**
(12:43) important, and I think it would help consumer perception as well. Finally, transparency, which is what we're going to talk about for the rest of the time here, transparency around AI, what it means, and how it can potentially help consumer perception. >> Thank you so much, Debbie. I'm going to

**AI Transparency Framework Overview**
(13:02) take the baton and show you a little preview, if you will, on the AI transparency and disclosure framework. Alyssa, I saw you somewhere in the crowd. If you would put links in the chat to both the research and the framework, I would be so grateful. I'm going to start with the opportunity and the risk, lifting out of what Debbie

**Opportunity for AI: Adoption & Cost Reduction**
(13:29) was just talking about with the research. On the one hand, the opportunity for AI is quite high. There's been exponential adoption; we're here talking about it, and David does these AI Insiders every week, so I don't think we ever run out of things to discuss. Reduced production costs are real, especially for public companies who are

**AI Benefits: Personalization & Localization**
(13:51) constantly thinking about efficiencies. That pressure is high. Large-scale personalization. We at the IAB are even talking about all the benefits in terms of ROI upside you can get from personalization and rapid localization. On the risk side, we've got the trust gap Debbie just talked about, anything from

**Risks: Trust Gap & Label Fatigue**
(14:15) inauthentic, 'I don't really like it,' to 'this is creative, cool,' but as she just mentioned, that's a wide gap. Label fatigue: if we label everything AI, then all of us will just tune it out. We don't want to go there. Fragmented regulation, and I'm sure this comes up in this group to some degree, but we've got the federal

**Fragmented Regulation & Consumer Confusion**
(14:40) government that's somewhat light on AI regulation, and then we have a very state-level approach. We also have platforms such as TikTok and Instagram, who have their own labeling mechanisms. In the middle of all of this, with this last bullet, is the consumer, who's like, 'What? What is it? Stars? Does that mean AI? Is it this text here? What's going on?' Just like in a pinball machine, if

**Focus on Consumer Deception**
(15:04) you will. Moving on to our next slide, what we chose to focus on in our framework was this idea of consumer deception. If we're going up with a threshold for when we think something should be labeled, what is that threshold? It's: do we think that without this label, this consumer will be deceived? The example I always give, and I'm so sorry to everyone who's heard

**Deception Example: Synthetic Influencer**
(15:29) me use this over and over again, but it's my favorite. If I'm on TikTok and an ad comes up for some skin cream—I buy too many of them—and I think, 'Cool, it's going to solve all my wrinkle concerns.' I buy it. I use it for the recommended three weeks. Nothing happens. Then I find out it was a synthetic influencer this whole time.

**High-Risk AI Use Cases**
(15:51) Wasn't labeled. I thought it was a real person with perfect skin. That would be, in our view, deception, and we would want it labeled. This is for the record, underneath all the advertising industry regulation that already exists. This is just an example; there's more in the framework. On the left side, I'm going to give you an example: What do we think of as high-risk things

**Disclosure Not Needed for Routine AI**
(16:16) like synthetic humans, for instance? Prompt-to-video and image generation is something we talked about a lot; we can get into it in the panel. Voice clones, digital twins of deceased persons. We can get much more into all of this because these are just words on slides. However, we also have examples of AI in which we don't think there's deception, and therefore, no disclosure is needed. Things like routine

**Low-Risk AI Examples (No Disclosure)**
(16:42) retouching, color grading, upscaling, things like it's obviously stylized. If we were all in cartoon form here today, you would realize something had been done to all of us, unless you think the world is a cartoon world. I want to join you there because that sounds lovely. And generic AI background music. I'm going to tuck this away and bring up the

**Introducing Panelists**
(17:05) esteemed panelists to the virtual stage to get into this discussion now that you've seen what we're about to talk about. First, I want to introduce Ken Fischer, editor-in-chief of Ars Technica. If you don't know about Ars Technica, how? Because we're all lovers of technology. Secondly, Graham Wilkinson, EVP, chief innovation officer

**Graham Wilkinson & Super Bowl Ads**
(17:28) and global head of AI at Acxiom. I wanted to call him Crayon because he told us people mistake the way he says Graham for Crayon. But I won't call you Crayon, that would be disrespectful. Welcome to all of you. If you would unmute so I can officially welcome you to the virtual stage. We just came out of the

**When Does AI in Advertising Matter?**
(17:52) Super Bowl, and we saw multiple ads that had AI or were touting AI capabilities—Anthropic's ad, OpenAI. Let's start with the fundamental question: When does AI use in advertising actually matter, and when is it just white noise? >> First, thanks for inviting me. It's cool to be here and see this side of the

**Ken Fischer on AI Disclosure Meaning**
(18:18) business. I spend most of my life in the editorial world. It's a great question. I feel that, coming from an editorial background, my sensitivities change, but I think disclosure really matters when AI is changing the meaning of something. As we discussed, as you showed in that slide, it's just white noise when you're

**Transparency for Changed Meaning**
(18:42) talking about AI as a production or efficiency tool. But when you start doing things that change the meaning or create a perception that might not be real, it's always safest to be transparent about that. Consumers expect that now. Whether they do in five years is a totally different question, but today it's clear that users want to

**Trust Threshold & Perspective**
(19:04) see that if there's a material change. >> For me, I agree with Ken, and Caroline, your analogy sums it up well. I think it's a matter of the threshold for trust, and that's a matter of perspective too. That's why I suppose the

**Generational Trust Differences**
(19:33) research from Debbie is important: looking at that particular generation, but also looking at other generations. An older generation might say they are more trusting because they've grown up being bought into TikTok videos that may already be heavily touched up and are nowhere near reality. And

**Ad Aim & Trust Breakdown**
(20:00) therefore their threshold for trust is lower, in a relative sense, to somebody else. It all comes down to who the ad is aimed at and our understanding of trust in that, and obviously once it breaks that trust. It could be a celebrity they entirely trust.

**Defining Trust Across Generations**
(20:28) was authorized to use their voice, their image, whatever it was. For somebody else, that might not be the breaking point. So, defining and understanding what trust means to different generations and perspectives is a key part of it, and it's a moving target.

**Gen Z & AI Videos: Knowing It's AI**
(20:50) >> I would add, I have two Gen Z daughters. They go on social media and watch AI videos a lot. They love them, share them with me, saying, 'Oh my god, this is so cool! Look at this!' They buy into all the trends, but they know they're AI. I think that's where the challenge is and where the

**Trust vs. Deception in AI Use**
(21:08) difference is: if they're not being fooled, if they're bought into it, if they're part of a conversation, if they get it, if AI is used in a way they think, 'Oh, okay, that's cool,' they love it. But if they're being fooled, or if it looks fake or like 'slop,' that's where my daughters get really

**Honesty and Openness with AI**
(21:28) negative toward it. For me, it comes down to trust, but also being honest and open, bringing your audience in on the joke or game, whatever you're doing with AI. That's what I take away. >> That makes a lot of sense.

**The Movable Line of Perception**
(21:48) also picking up on something you said, Graham. Graham and Ken were part of the working groups for this framework for months. We talked about this for months because there are so many edge cases, and what Graham just talked about in terms of perception—there's going to be a movable line of perception—also how we all on this

**Synthetic Humans in Testimonials**
(22:11) call think about AI is going to change over time. We recognize that all of what we put here may shift in six months. I also wanted to pick up on some questions from the chat. Derek asks, 'Why would I use synthetic humans in a testimonial?' I see Miesa's response is interesting: it converts better. Sometimes, an example would

**Digital Twins for Celebrities**
(22:37) be, we've all seen Jennifer Aniston in the Smartwater commercial. Let's say Jennifer Aniston is in Bali, having the time of her life, but isn't able to come to the set in Chicago. What if she could send her digital twin? It's completely authorized. It's much cheaper than flying her from Indonesia to Chicago. That would be an example of it.

**AI Changing Meaning in Ads**
(22:58) would be a digital twin of her, completely synthetic. Maybe it's just the look and feel of the person that you wanted and how fast you wanted it. I also want to ask this question to Ken. Marshall asks, 'Ken, can you give us an example or examples of AI changing the meaning of what is being conveyed or advertised?' >> Sure. The example you

**Synthetic Influencers & Simulated Authority**
(23:21) gave, Caroline, very early at the outset, a synthetic influencer. I think that's a great example where if you're creating a person who is reaping the benefits of a given product, but that person isn't real, I think that's a good example of an area that is 'icky.' I think you can also do simulated authority. It's adjacent to how I think about

**Financial Industry & Fake Authority**
(23:46) influencers. If you watch an ad and the guy's saying '9 out of 10 dentists say whatever,' it'll say 'actor portrayal' or whatever. We're always very careful about that. So what if a financial industry wants to run an ad that has, say, the new digital EF Hutton? Remember those old EF Hutton ads? You

**Faking Product Realities**
(24:09) create the new digital EF Hutton. That would be an attempt to create what is essentially a highly credentialed but fake representation of authority. Another thing I think of is product realities being faked. For instance, if you show someone suddenly dunking when they couldn't before, or these

**Augmenting Human Capability**
(24:37) kinds of things. You can augment human capability, which also means their reaction to products in a way that's not realistic. I want to circle back. I think it's very interesting that Debbie's research is showing this gap because you have people who want really creative and interesting ads, and then you have people who are

**Trust is Dangerous Territory**
(24:59) hyperfocused on the trust aspect. Between those two possible trajectories, the trust aspect is the more dangerous territory right now. That's why I think we want to advise advertisers to stay away from those areas if there's any way it's going to come off as,

**AI as a Cost-Saving Tool**
(25:23) essentially faking it. >> Yes. >> There's real sensitivity around faking things. I was also very interested to see the advertising industry realizing some people are aware that AI is just a way to save money. On the journalism side, I can tell you that's the

**Consumer Perception: Cheapening Content**
(25:47) consumer perception everywhere, and why they hate it. They feel like you're cheapening this, making it less real, because all you care about is money. >> That sounds like a quality concern. If you're cheapening or reducing the money, are you reducing the quality of the content? Is that what I'm hearing?

**Upset by Cost-Saving Over Quality**
(26:08) >> No. On the edit side, even if the quality was just as good, I would say people would be very upset at the idea that we're using a tool simply to save money when there were traditional ways of doing it that, in their view, are just as good. >> Interesting. Can I ask a follow-up question? We talked about this in the prep call, but I

**Niche Audience vs. Broader Audience**
(26:29) think of your readers as a very niche audience because it's Ars Technica, emerging technology aficionados. Do you think that's representative of the broader audience, or is that a signal from your specific audience? >> It's very interesting because it seems to be in pockets of community. For instance, at Ars

**Community Pockets & AI Perception**
(26:55) Technica, broadly, most readers are incredibly suspicious of AI. But there are other brands in the building, not necessarily demographically different, whose readers embrace it. They're like, 'Oh, this is great! I want to see more AI stuff!' What we have learned is that you can't guess this. If

**Online Communities & Creativity**
(27:18) somebody had said, 'A GQ reader is going to think this about AI,' you'll be wrong. You'll be wrong when you collect the data because these pockets of communities, I think, take creativity differently, and they get something different from creativity. And

**AI and Creativity: Be Safe, Not Sorry**
(27:40) so, as much as these tools are touching on creativity, that's where I think you get different levels of response. To me, it's a minefield, and that's why I always say, 'Be safe, not sorry.' >> I also want to add that >> what Ken brought up about sensitivity is important. I also think people have been desensitized as well.

**Desensitization to Fake Content**
(28:07) There's kind of equal and opposite. There are groups of people that think it's okay for things to be entirely fake as long as it's funny and maybe they're not the butt of the joke. But the consumer doesn't always have to be the person being deceived. Think of a travel ad where a celebrity is being

**Deception Beyond the Consumer**
(28:32) superimposed on a location, and that location doesn't want to be associated with that celebrity. Maybe they're controversial, and they know that celebrity never even visited there. That's disingenuous. The subject of the deception is not just the consumer; it's the location itself. >> Right. I think that as

**Balance of Notification & Fatigue**
(28:54) much as there are groups who are sensitive, there are groups who are alarmingly desensitized to it. And that's why, again, we talked ad nauseam about the balance of giving people notifications on this type of information and the fatigue associated with it, but also a duty to the industry to stay honest in what we're

**Andrew Cuomo's AI Ad Example**
(29:21) doing. As I said, there's no silver bullet, but it's important to think about all its aspects. >> This also brings up an example we talked about a lot. How many people here are in the New York City area, David? For sure. I know. We had a mayoral race where Andrew Cuomo was one of the candidates, and he did an

**Digital Twin & Deception**
(29:47) ad in which it was his digital twin in a variety of activities he never did in real life. Of course, there was a label at the bottom of the ad. You'd say, 'Okay, that's his digital twin. He endorsed it because it was for his ad.' But wait a second, he's now in a variety of activities he didn't do. Doesn't that seem like deception? Because people could

**Consumer Deception in Political Ads**
(30:16) potentially change their vote if they're like, 'Oh, Andrew Cuomo.' I don't even remember the activities in the ad, so I can't speak to them. But there could be some consumer deception there, and we thought long and hard about that. >> I want to jump in because I watched that ad. I live in Seattle, so I have no interest whatsoever in the mayoral race in New

**AI Woven into Creative Strategy**
(30:37) York. Well, I do. The point I wanted to make was the way the AI was woven into the ad pulled the audience into the idea. He basically said something like, 'These are not things I would actually do, but this is what I can do for New York City.' He used these crazy,

**Labeled or Unlabeled AI?**
(31:02) AI-generated scenarios as part of the ad's creative to demonstrate something else. That, to me, goes back to: why wouldn't you bring your audience along on the idea? >> Let me ask you: Do you think it needs to be unlabeled or labeled? >> Oh, it does need to be

**Political Ads Need Labeling**
(31:23) labeled. When we did the research, we asked what types of industry ad categories should have labeling, and political is one of the top, and I believe that's very true. I would definitely vouch for it. I think it's smart that it was labeled, and the idea of the creative incorporating the AI part of it into the ad's conversation

**Most Surprising Discovery from Research**
(31:45) ad was what made it work for me. >> I'm going to turn back to you because I have another question. From all the research you worked on with us, what was the most surprising discovery? The fact that the gap widened, from the previous research in late '24 to the research in late '25, the fact that the

**Widening Gap Between Execs & Gen Z**
(32:08) gap between ad executives and how they thought younger consumers felt, and how younger consumers actually felt, was probably the biggest surprise for me. I thought with more awareness of AI-generated advertising, we would see more acceptance and positive feelings toward it. I think the Gen Z part was also really interesting: that Gen Z

**Gen Z's Negative View: Wanting AI on Their Terms**
(32:26) is even more negative. For me, I feel it's because they use AI so much and are so attuned to it that they want it on their own terms. They want to be able to use it on their own terms, but also appreciate other people's use of it on their terms, when it's as art, as creativity, as

**Gen Z's Skepticism of Advertising**
(32:52) something they really value. I also think, generally, Gen Z is more skeptical and discerning of advertising messages. We see that in other research about marketing effectiveness, so that could be another factor. >> Oh, I like Tamika's comment. Perhaps it reflects the broader sentiment of Gen Z

**Gen Z Anxiety & Desensitization**
(33:13) being overwhelmed and more anxious because of tech. We did talk about that, especially because entry-level jobs are being heavily disintermediated, and those would be their jobs. I'm going back to something you just said, Graham, about desensitization. That's a big word for me. I'm wondering if maybe when we do this study again, we ask: how do you feel, maybe

**The Awkward Time of AI Maturity**
(33:38) indifferent, because I think when we hit a plateau of the tech, that's when it's invisible, meaning we're not talking about all the AI in the ad anymore. It's just an ad, and we don't care how it was produced. What do you think? >> I think that's the awkward time we're in at the moment. I said this a lot: we're in this

**AI: Model T with No Roads**
(34:03) between where the tech is in everybody's hands, but honestly, it hasn't matured enough from how consumers are using it, how it's embedded in processes and businesses. The Model T has been invented, and there are no roads to drive on. So, people are driving all over the place, modifying their Model T so that it drives better off-road and

**Need for Faster AI Research**
(34:29) instead of focusing on building roads where more people can drive and get places faster. We're just in this weird in-between time. This type of crucial research that Debbie has done needs to be done faster. Debbie, do it faster and more often, because things are changing

**Speed of AI & Obsolescence**
(34:56) so rapidly. Even the window we get into it is already obsolete by the time we look at it. That's the speed of AI, the double-edged sword of this situation. >> I totally agree with that. But one of the things I heard from somebody else at Kai when we

**Loss of Trust is Immediate**
(35:20) were talking about this: AI is fast, but the loss of trust is immediate. >> It's instantaneous. That's one of the things we try to bake into our work. We may have to revisit our governance, revisit these things. It's going to be a cadence, but we should be very sober about it. We should resist,

**Trust Damage is Insane**
(35:47) the shiny new thing and running after it like fools because that trust loss will get you in minutes. It's not a developing thing if you get caught in it. One thing about the speed is I know it creates a lot of anxiety for FOMO, but the other side of that is the damage to trust is insane.

**Nuance Over Extremes**
(36:11) >> I like that a lot. It's the saying, 'A moment on the lips, a lifetime on the hips.' I think that's the learning: I get incredibly nervous around anyone who swings either overtly bullish or bearish on anything, because I think the nuance somewhere in the middle is the place to try to be always. I really like that perspective. Graham, I

**Transparency in AI Personalization**
(36:38) also want to ask you about something from your particular place at Acxiom. You are at this intersection of data and personalization. We talked about personalization quite a bit when we were trying to think about transparency. As you're innovating with AI for audience targeting, personalization, where does transparency fit there?

**Low-Resolution AI Solutions**
(37:03) of my time personally is spent with marketing departments doing this really boring exercise of mapping out agentic flows. The reason is that so many people have jumped into AI and built what I would call low-resolution solutions. There are many problems with low-resolution solutions; one is they go, 'I've

**Inability to Explain AI Processes**
(37:31) built a brief writer,' or 'I want to build a brief writer.' Then you say, 'What does that mean? What does a brief writer do? How many tasks are nested under that?' And people can't explain it. That's the problem, because if you can't explain it, you can't get to what we're talking about, which is supplying metadata associated not just with the output,

**Focusing on Decision Tracing**
(37:53) itself, but the decisions, the reasoning, the reason tracing that sits behind it. It also results in a generalized response. I think a big part of where we're focused is not necessarily, 'Hey, this is how you should use Acxiom data. This is how you should use audiences.' It's more, 'Hey, let's look at your entire

**Recursive Task Atomicism**
(38:22) process and break it down.' We carry out this thing called recursive task atomicism. Essentially, we break every task down to its atomic state. Its atomic state is defined as something that has one persona, uses only one tool or API connection, and its output can be judged by a pass/fail binary response.

**Mitigating Risk & Explaining Outcomes**
(38:49) Now, if you can break a description like a brief writer down into its constituent tasks, and every one of those tasks' output can be judged by a pass/fail output, then it means: one, you're going to mitigate risk associated with its use in the advertising process; two, if you have to explain why something was done, you can trace it back

**Business Process Articulation**
(39:17) very easily. And three, what I'm realizing most is that most businesses cannot articulate the processes they carry out on a day-to-day basis at that level of definition. That is a big problem. But for us, it also means that once we have that map, we can say, 'Hey, these are the points where you can inject data,' and that could

**Data Injection Points**
(39:45) be injected because it's low risk. It could be injected because it's going to have a multiplying effect on what you're doing. Maybe it does something that's not possible for humans to do if they undertake that process. For me, the focus is less on what Acxiom sells, and more on saying to brands, 'Just take

**Deep Dive into AI Solutions**
(40:07) a step back, let me show you that for you to get the ROIs you're under pressure to achieve for efficiency and things like that, you can't do it with these low-definition AI solutions. You have to go deep into it, and you need to. One of the byproducts of this is that it tends to bring to the surface the people

**System Thinking & Organizational Transformation**
(40:35) who are more leaned into system thinking and also highlights people who are more rigidly stuck in their ways—they generally cannot explain very well what they do, but they've done it for a long time. It's part of the organizational transformation too. That is big. It's pretty much where I

**Nuance & System Thinking**
(41:00) spend 99% of my time these days. >> I also think it connects to what Ken was just talking about. Yes, that was about breaking trust, but you're also talking about applying some nuance to how you're thinking about full context needed before optimal AI operating behavior, let's call it, which is the same sort of system thinking. Debbie, I wanted to

**Consumer Clarity on Transparency**
(41:26) turn it back to you because we did not show in the slides earlier the nuance around what the data was telling us in terms of what consumers wanted transparency around and what they didn't care about so much. Could you give us some clarity on that? >> For sure. I am going back up to get the study in front of me so I don't

**Strong Alignment on Disclosure Techniques**
(41:50) misstate... >> Okay. It's 9:45 in the morning here, and I still haven't had enough coffee, so I can't say the words I want to say this morning. When it comes to the types of ad creative techniques people want disclosed, there's a strong alignment

**AI-Generated Content Disclosure**
(42:11) between how the younger consumers in our study felt and how the ad executives felt. The largest percentage of people said, for example, if an ad is 100% AI generated, 57% of advertisers and 58% of consumers said that should be disclosed. Similar for AI-generated images. When it comes to video, there was a bit more discrepancy; 48% of advertisers said

**Video Disclosure Discrepancy**
(42:39) they felt that should be disclosed, but 57% of consumers wanted it disclosed. Almost the exact same percentage of consumers wanted disclosure if the ad was 100% AI-generated. Things less likely to be desired for disclosure were AI-generated copy or AI-generated avatars, which were somewhat less

**Obvious AI Disclosure Scenarios**
(43:04) likely to be desired for disclosure. I feel it's kind of obvious, right? If everything is AI, let's disclose it. If we're using a lot of AI images, yes, let's disclose it. Video is another thing. Those are the top takeaways, but what was really interesting to me again was that advertisers were pretty much like, 'Yeah, we

**Panel Alignment on Disclosure**
(43:25) agree,' similarly to what consumers feel about the disclosure. >> Interesting. I'm glad we're aligned. I also wanted to get deeper into this conversation for both the panelists and all of you. There are a lot of things I see coming through in the chat about the specific nuance. I saw Elena. Hopefully, it's not Elena. Is it Elena? Did I

**AI Photo Enhancements & Photoshop Scrutiny**
(43:51) get it right? I like how you brought up whether there's additional scrutiny around photoshopping, especially with AI photo enhancements. We talked about this because there's photoshopping, and there's Photoshop with enhanced AI. Is there any nuance for the consumer? We chose to think no in this case because if we hadn't been disclosing it

**Paris Hilton Ad Discussion**
(44:16) before, why would we disclose it now? What I wanted to do, because I thought it would be fun, is show us an ad. Plus, I really wanted a reason to bring a Paris Hilton video into this conversation, and the panelists agreed with me. So, I'm going to show you two ads. Does anyone here, by show of emojis or hands, remember the Paris Hilton ad from 2005 or 2006 in

**Carl's Jr. Paris Hilton Ad**
(44:42) which she's eating a Carl's Jr. hamburger on a car and spraying herself with water? Anybody? Derek, Ken, Peter, Michael, Debbie. Okay. Enough people. I'm going to show that one first to remind us all of what we're about to talk about. >> My bad. It's all good. Look.

**Digital Twin in Paris Hilton Ad**
(45:09) Famous Starwatch. What is this place? Good boys. >> Buy one, get one for $1. >> We're going again. >> Okay, a couple things to mention. First of all, the versions of Paris in the bathing suit—they're trying to draw your attention to the fact that that's the digital twin of

**Panel Question: Disclosure Needed?**
(45:50) her because of the blue eyes. That's her digital twin, not real Paris. The real Paris was her in the pink jumpsuit at the end when she comes out to them in the car wash. My question for our panelists, and I'd love to know from everyone, is a global question. Just put your answer into

**Audience Poll on Ad Disclosure**
(46:14) the chat here: Do you think any part of that ad needs to be disclosed as AI usage? Just say yes or no. Nos. Yes. No. I don't know. Marshall, very honest. Honestly, it feels very split here so far. No. Thanks, Nate. I want to get a little deeper into the panel. Ken, let's go to you

**Ken's View: Obvious AI Doesn't Need Disclosure**
(46:45) first. Your answer was no, I believe. >> Yes. I think there are going to be situations where it's obviously not real. I don't say obvious, as saying 'obviously AI' might assume too much on the viewers' awareness of technology trends. But I think even my grandmother would be able to watch that

**Obvious Manipulation Threshold**
(47:12) ad and know, 'Okay, they didn't clone this Paris Hilton person and give them laser eyes.' It's in the same way that we're not requiring cartoon avatars of famous people to be disclosed. I think it passes an obviousness threshold for being manipulated with technology. >> Fair. Graham, what are your thoughts? >> I said yes, and I think two

**Graham's View: Intermingling Real & AI**
(47:41) reasons. One is when you are intermingling real and AI generated, you are kind of inferring or at least giving the impression to people that maybe it's one or the other. I think too, in a world where vanity, appearance, and all these things have become heightened, and at the same time we all

**Unreal Expectations & Vanity**
(48:11) are apparently way more anxious and everything else. It's this idea of, 'Are you setting unreal expectations for people that view this?' Paris Hilton is a lot older these days than she was back then. >> Ouch. Come on. >> No, she's a beautiful lady, but it is

**AI Exacerbating Bad Actors**
(48:36) there is a reality aspect of it. As much as, maybe as a man, you can sit there and go, 'It makes no difference to me, I couldn't care less,' as a young girl, does that make a difference to you? I think it's not necessarily an AI problem. AI is just the tool being used, but there's a problem

**Societal Pressure & Unrealistic Expectations**
(49:00) with perception and how you set expectations that are unrealistic for people. Certainly, in a world where you can log on for 5 minutes and get yourself a GLP1, not necessarily because you're morbidly obese, but you just want to lose a little weight because you feel a ton of pressure from society to look better. Look better being a subjective thing, so I think,

**AI and Bad Actors**
(49:25) it's less that AI can be named as the bad guy in this, but I also think AI will just exacerbate bad actors, which we know already. >> Ken vigorously nodding. >> Yes, Ken agrees with you. I also think it's incredibly interesting because something that came up in my mind on the organic content side of things, and this is because I was at

**ABBA Experience & Voice Clones**
(49:48) Warner Music Group before and thought about this non-stop: the ABBA experience, how they brought them back, obviously as younger versions of themselves, and used voice clones of their younger versions. I would hope everyone would realize that ABBA is not perpetually youthful, but I haven't thought about the larger societal implications like the

**Debbie's View: Yes to Disclosure**
(50:14) Kiss experience; they want to do the exact same thing, but them as a younger band. I haven't actually thought through. Graham, this is why I love you so much because you constantly ping it into new parts of my brain. Debbie, let's get to yours. Was it a yes or a no? >> I was a yes. I think it's because I feel that right now I

**Current Lack of Obvious AI**
(50:35) wouldn't say yes forever, but right now, it may not be obvious to everybody that there's AI. We saw in the research that disclosing didn't have a huge impact on purchase likelihood, and it was one of its drivers of attention, in addition to high-quality visuals and funny content. I would agree this is funny content, and it's

**Upside of Disclosure, True Test of Ad Performance**
(50:59) pretty high-quality visuals. I feel there's not a downside to putting a little disclosure on this ad. There's probably more of an upside. But the true test, I think, is, 'Okay, we're selling hamburgers. Did Carl's Jr. actually sell more hamburgers? How did this buy-one-get-one offer do?' That's the true test of did

**Framework: No Disclosure Needed**
(51:20) this ad perform outside of whether they used AI? I'll come back to our framework, which, just for the record, would have said no disclosure necessary. The reason it would have said no disclosure necessary is because there's no real deception in using a digital twin that Paris Hilton has authorized. And whether people realize multiple Parises, she

**Nuance & Avoiding Label Fatigue**
(51:45) didn't just somehow have a bunch of twins she never disclosed in her lifetime. We would not recommend disclosure there, and we would also say so because we're being nuanced about what we want labeled, as we don't want it to lead to label fatigue. What I think this conversation should elucidate is that this is a constantly moving target of perceptions and trying

**Closing Question: AI Transparency Guidance**
(52:11) to figure out, and it's not an easy discussion. It took us many months to even get to some semblance of what we think for now. I know we're coming up on time, so I will ask a closing question for all the panelists. If you could give everyone in the room one piece of guidance on navigating AI transparency right now, what would your advice be?

**Prioritize Creativity with AI**
(52:37) >> I'll start. I believe that using AI to improve your creativity should be the top reason you do it. If you believe AI will actually be cheaper, yes. But if you go into it with the idea that you're going to be able to produce lots of ads very cheaply, they're going to look that way. I don't think over time consumers are going to react

**Transparency as Long-Term Brand Strategy**
(53:01) positively. Always have that creative north star in your mind when you're using AI. >> I would say, even though I said no with regards to this ad, think about and treat transparency as a long-term brand investment strategy instead of a short-term performance variable for

**Transparency as a Safety Valve**
(53:25) your campaigns. Because of the rate of change and how much the space will change over the next two years, something we think today is no problem could very well be a problem in a year. You never know what scandal might change public perception, etc. So, I think transparency is the safety valve for experimentation with

**Understanding AI Deployment Processes**
(53:50) AI. >> I would harp on my point about better understanding the processes you deploy AI into. Forget consumers or anyone else, but you're deceiving yourself if you can't fundamentally understand how the output was generated. I don't mean you have to understand

**Precision in AI Outputs**
(54:17) how neural networks work, but if you give a generalized task to a generalized agent, it's going to give you something very difficult to unpick and explain. At the end of the day, I think we should be focused on being more precise in our outputs. It doesn't restrict creativity; to Debbie's point, it enhances it, I believe. But I

**Value Human Skills with AI**
(54:44) think, put much more thought and effort into, 'Why does this process work? How do I do this task as a human being?' Don't devalue your natural skills as a human being because we take them for granted that we can do all these multifaceted things. Try to figure out what those things are and what it means to be human

**AI and What It Means to Be Human**
(55:08) because that'll make you super awesome at using AI. >> I'd love to do an AI Insiders, David, on what it means to be human. >> After going on some AI dates last week at a real-world cafe. >> Wait, did you go to the popup? >> I was quoted in the New York Times for going to the popup.

**AI Dating Popup**
(55:29) >> Put the link so we all can read. >> Does everybody know about the popup where you can date AI that's happening in New York? Nate, we should go. That would be a fun video for us to test this out. >> David, I want to read your comments immediately. >> Yes, that was my dinner date with

**Upcoming AI Dating Session**
(55:54) AI. I'm putting this in the chat right now. This is >> very top of mind >> and we will have a session on this. We had to schedule it a little bit later because we have amazing sessions like this already scheduled. Robin Gelfan, this incredible comedian who did this with me back-to-back, we're going to have a

**He Said, She Said AI Dating**
(56:17) >> When is that session? Tell us all. >> It's not until March 25th, but we'll get into it. We'll do a he-said, she-said version, and it's going to be a lot because these issues will be bubbling up for a long time. >> Thank you for having us, and thank you to Debbie, Graham, and Ken. I put in the chat, I've

**Session Wrap-up & Future Invitation**
(56:36) never seen a session go by this quickly. I thought we had tons more time. Thank you all. If you ever want to do a longer version of this, you've got the invite here. >> Thank you. >> Thanks everyone for coming. Incredible questions. Caroline, I'm happy to share the questions with you too in case you missed any of the

**Final Thanks & Upcoming Events**
(56:54) comments and things you want to follow up on. Thank you all. This is great. Hope to see some of you, as Sam mentioned, at Marchitecture live in a few weeks, and see you next week for a session on vibe coding for good.