Learning Library

← Back to Library

AI Ads, Controversy, and Taylor Swift

Key Points

  • AI‑driven marketing is booming, with AI now powering roughly 40% of Instagram feeds and companies like Meta investing billions in large‑scale models to tailor video ads.
  • Brands are increasingly mixing real talent with AI‑generated elements—as illustrated by the Sydney Sweeney ad where a car scene was fabricated—to spark controversy and stand out in crowded spaces.
  • Historical controversial campaigns (e.g., 1990s Brooke Shields Calvin Klein ads) show that “gen‑ads” must provoke debate because the product category is ubiquitous and hard to differentiate.
  • High‑profile deep‑fake incidents involving Taylor Swift are being used by journalists and researchers as benchmark tests for AI safety, highlighting the rapid spread of synthetic celebrity content.
  • The convergence of cheap, powerful AI creation tools and a willingness to leverage controversy creates a feedback loop that pushes advertisers toward constantly seeking new “local maxima” in consumer attention.

Sections

Full Transcript

# AI Ads, Controversy, and Taylor Swift **Source:** [https://www.youtube.com/watch?v=yqmdJ2lQw98](https://www.youtube.com/watch?v=yqmdJ2lQw98) **Duration:** 00:20:15 ## Summary - AI‑driven marketing is booming, with AI now powering roughly 40% of Instagram feeds and companies like Meta investing billions in large‑scale models to tailor video ads. - Brands are increasingly mixing real talent with AI‑generated elements—as illustrated by the Sydney Sweeney ad where a car scene was fabricated—to spark controversy and stand out in crowded spaces. - Historical controversial campaigns (e.g., 1990s Brooke Shields Calvin Klein ads) show that “gen‑ads” must provoke debate because the product category is ubiquitous and hard to differentiate. - High‑profile deep‑fake incidents involving Taylor Swift are being used by journalists and researchers as benchmark tests for AI safety, highlighting the rapid spread of synthetic celebrity content. - The convergence of cheap, powerful AI creation tools and a willingness to leverage controversy creates a feedback loop that pushes advertisers toward constantly seeking new “local maxima” in consumer attention. ## Sections - [00:00:00](https://www.youtube.com/watch?v=yqmdJ2lQw98&t=0s) **AI‑Driven Ads and Controversy** - The speaker outlines the rapid rise of AI in marketing, highlights Meta’s billion‑parameter model and AI‑generated imagery, and uses the Sydney Sweeney ad controversy to illustrate how brands deliberately leverage AI‑created content for attention. - [00:04:29](https://www.youtube.com/watch?v=yqmdJ2lQw98&t=269s) **AI‑Generated Media and Marketing Dilemma** - The speaker explains how AI‑created images like synthetic Taylor Swift concert shots blur the line between real and fake content, creating an uncanny‑valley challenge for marketers and prompting a discussion of the system’s stagnation and potential solutions. - [00:07:35](https://www.youtube.com/watch?v=yqmdJ2lQw98&t=455s) **AI-Driven Parasocial Feedback Loops** - The speaker explains how algorithmic content delivery and AI companionship apps generate layered feedback loops that alter user behavior and expectations while fostering intensified parasocial relationships with both real celebrities and AI‑created characters. - [00:11:19](https://www.youtube.com/watch?v=yqmdJ2lQw98&t=679s) **AI Engagement vs. Authenticity in Marketing** - The speaker argues that AI-driven engagement systems reshape human psychology, creating tension with brands' demand for authenticity, and suggests emphasizing inherently un‑AI‑replicable experiences—illustrated by Taylor Swift’s anti‑AI album reveal and vinyl pre‑sales. - [00:15:21](https://www.youtube.com/watch?v=yqmdJ2lQw98&t=921s) **Experiential Retail Beats Controversy Marketing** - The speaker contends that lasting customer value and cash flow stem from unique, in‑person retail experiences rather than short‑lived controversy‑driven hype, emphasizing the need to translate AI‑generated attention into tangible store engagement. - [00:19:03](https://www.youtube.com/watch?v=yqmdJ2lQw98&t=1143s) **Strategic AI Integration in Branding** - The speaker urges marketers and technology builders to deliberately weave AI—particularly LLMs—into brand strategy, moving beyond algorithmic limits, guardrails, and local maxima to thoughtfully select ideas and cultivate long‑term brand relationships. ## Full Transcript
0:00All right, stay with me. We're going to 0:01talk about local maxima algorithms, AI, 0:04and Taylor Swift. I'm going to make it 0:06all make sense. I promise. First things 0:08first, I think everybody knows that AI 0:10marketing is exploding. Something like 0:1340% of the Instagram feed is AI now. 0:16It's growing really fast. Meta is 0:18investing heavily in AI tooling. Meta 0:21just published a 1 billion parameter 0:24artificial brain that's designed to 0:26measure your response to videos and 0:30simulate it so they can build more AI 0:32videos for you. And it's not just the 0:34tooling that's changing. It's also 0:37actual ads that we see coming out all 0:39the time that are controversial or AI 0:42generated. And it's not even just the ad 0:45space. It's also AI generated imagery as 0:47a whole is exploding frankly because the 0:50tools are getting cheaper and better. 0:52I'm going to go through just a couple of 0:53use cases to kind of freshen you up. 0:55None of this will be surprising and then 0:57I'm going to explain how it all comes 0:59together. So, first the Sydney Sweeney 1:01case, right? This was a famous case 1:03where Sydney was a real person in an ad, 1:07but certain components of the ad were AI 1:10generated because it was cheaper. I 1:11believe the car driving away was AI 1:14generated toward the end of the ad. The 1:16point here is that the brand chose to 1:19lean into controversy and they've 1:22admitted that, right? They chose to lean 1:24into a controversial stance with their 1:26with their particular angle on the GAN 1:28ad. And they chose to lean into classic 1:31controversial jeans ads from the past. 1:34The Brook Shields ads from the 1990s 1:36were deliberately evoked. We're not 1:38here. I am not going to break down 1:40marketing history for you. If you want 1:42to look up the Brook Shields Calvin 1:43Klein ads, it's a very famous example of 1:45the controversy that genes ads have 1:47historically done, which in and of 1:49themselves are interesting because genes 1:51are ubiquitous. Everyone has genes. And 1:54so having to market them means you sort 1:56of have to step into controversy to try 1:58and differentiate yourself in the space. 2:00And this is exactly what we're going to 2:02dig into more as we get into this idea 2:04of local maxima. The second thing I want 2:06you to keep in mind, now we're getting 2:08to Taylor Swift, is this idea of Taylor 2:10Swift deep fakes. Now, deep fakes are 2:12not new. They're not certainly not 2:13limited to Taylor Swift, but because 2:15Taylor Swift is a famous woman 2:18celebrity, we get in a special case, a 2:22special prevalence of deep fakes, 2:24special visibility on deep fakes, 2:26journalists tend to use Taylor Swift as 2:28a test case when they are investigating 2:32the safety of AI systems. And so that's 2:34why we get an immediate test of Grock's 2:36imagine feature and an immediate sort of 2:39news headline around the world that 2:40Grock's imagine feature will create 2:42adult images of Taylor Swift as deep 2:45fakes. You might wonder now where is 2:46this going with ads? I'm sharing this 2:49part because we are fundamentally in a 2:53situation where the technology to create 2:56ads and the technology to create 2:58controversy, which is exactly what we 3:00got at with the Sydney Sweeney case, are 3:02converging and cheapening. In other 3:03words, part of why we're at a local 3:05maxima, is that the technology to create 3:07this kind of product is cheaper and 3:10cheaper and cheaper and cheaper. digital 3:12ads that would have cost, you know, a 3:16million dollars a couple of years ago 3:18are a tenth of the price or less now and 3:21they're going down further. We are going 3:23to see more and more major brands jump 3:25on the AI generated ad train. I am I am 3:28fully prepared for a doubledigit share 3:31of the ads in the Super Bowl in 2026 3:34being AI generated at least in part. So, 3:38there's a third piece here that I want 3:39to get at before we start to dive into 3:41the relationship between all of these 3:43components. And the third piece is the 3:45idea of the uncanny valley. Essentially, 3:50you need to combine the controversy of 3:52Sydney Sweeney, the ease of production, 3:54and the parasocial nature of celebrity 3:57relationships, which enables the 3:58creation of deep fakes of Taylor Swift 4:00and other celebrities, with the notion 4:03that we can't tell the difference 4:04anymore. And this is something that's 4:06true even of Gen Z. It's not really an 4:09agist thing. Gen Z in study after study 4:13after study professes to care about 4:16authenticity. But the reality is they 4:19also struggle in study after study to 4:22tell the difference between good AI 4:25generated material and real material. 4:27And I will tell you as a part of the 4:29test case here, I went to midjourney to 4:32create images of Taylor Swift in 4:34concert. They were not inappropriate 4:35images. They were just like Taylor Swift 4:37in concert images. And if I had been 4:39shown those images, I would not be able 4:41to tell you if it was an artistic shot 4:43by a photographer from a real Taylor 4:46Swift concert or if it was made up. And 4:48I think that that's part of what's going 4:50on that makes it challenging here is 4:52that we ourselves are unable to 4:54distinguish artificial tokens from real 4:57tokens in ad situations or other 4:59situations today. It's true in words 5:02too, but we're talking in marketing and 5:04images are central to marketing. So, 5:05we're going to stay focused on that for 5:06the day. So, those are the three pieces 5:08I want to bring together. We have Sydney 5:09Sweeney, we have Taylor Swift, and we 5:11have this idea of the uncanny valley. 5:13Where does this take us? Fundamentally, 5:15I want to suggest to you that we are 5:17living in an evolved system that has 5:21unintentionally 5:23evolved to a local maxima. And I want to 5:26walk through the technical reasons for 5:28that. And then I want to walk through 5:29because I'm Nate and I like practical 5:31solutions. Some ways forward out of that 5:33local maxima. So first things first, 5:36what are the system dynamics that we're 5:38all living in for marketing? And this 5:39includes marketers. My job here is not 5:41to blame marketers. It's actually to 5:42talk about the system as a whole and how 5:44AI is accelerating it. AI systems are 5:47creating evolutionary pressure on human 5:49content creation. So artificial 5:51authenticity is going to out compete 5:54genuine authenticity in algorithmic 5:56environments. In other words, AI is an 5:58accelerant on top of the social 6:01algorithms that we have built and 6:03optimized through social networks over 6:05the last 20 years. This includes at 6:07least four individual feedback loops 6:10that are all accelerating. One is 6:11sampling feedback loops. So, 6:13controversial content tends to get more 6:15distribution and creates training data 6:17that biases towards more controversy. 6:19That's the Sydney Sweeney use case. 6:20Essentially, 6:22the feedback loop samples and gets an 6:25idea of what works well. Controversial 6:28content, it's well known, does better in 6:30the feed, and so it generates more 6:32controversial content to accelerate 6:35that, right? I we have talked about this 6:37broadly as a society. We haven't 6:39necessarily talked about the idea that 6:40AI is an accelerant for this. AI 6:43accelerates the uh production of 6:47controversial content by making it 6:49easier and cheaper cheaper to produce 6:51and that in turn reinforces that 6:53training data loop for the algorithms. 6:55The second one is 6:57feature feedback loops are a problem. So 6:59when a user engages with content they 7:02create a feature in the algorithm. User 7:04engagement with manipulative content 7:06will teach the algorithms that 7:08manipulation equals quality. That one is 7:10also not new as a feedback loop. But 7:12again, AI is accelerating it because AI 7:16is essentially able to produce content 7:19that is more likely to be engaged with. 7:22It's content that is more likely to be 7:26latched on to by humans. And humans 7:28can't tell the difference. This gets 7:30back to the idea of the uncanny valley. 7:33We can't tell. And so we click and we 7:35engage and we're feeding both the 7:38content algorithm that feeds us stuff 7:40and we're also teaching the marketers 7:42that AI works well. The third feedback 7:45loop is an individual feedback loop for 7:47us. Repeated exposure to optimized 7:50content changes our behavior and 7:52expectations. There's been a lot of 7:54studies around sort of what the 7:55Instagram feed has done to our 7:57self-image and our social relationships. 8:00I think we can also talk about this idea 8:02of parasocial relationships here. A lot 8:05of the Taylor Swift phenomenon that I 8:07discussed is this idea that Taylor Swift 8:09is not just a celebrity, but Taylor 8:11Swift is someone with whom I have a 8:14right to have a relationship, even if 8:16it's a madeup relationship in my head. 8:18So, this gets back to the number that I 8:21gave you at the very beginning of this 8:23video where we talked about Character AI 8:25and other sort of AI uh companionships 8:28apps and the and the money that they're 8:30generating$ 220 million some dollars per 8:32year growing very fast. They're 8:34cultivating the idea of parasocial 8:36relationships, but they let you create 8:37the character, right? They let you 8:39create the character you're going to 8:40have an artificial relationship with. 8:42Whereas traditional parasocial 8:44relationships, it's your imagined 8:45relationship with the celebrity. In both 8:48cases, AI accelerates this individual 8:51feedback loop in your own head where you 8:54are changing your behavior and 8:55expectations because of what you're 8:57engaging with. The fourth feedback loop 8:59is outcomes. Essentially, successful AI 9:03content that runs through these feedback 9:04loops becomes the baseline for normal 9:07content. And so, in a sense, it's not 9:10just that the AI content is changing the 9:12way we consume behavior and changing 9:14what is in our feeds. It is also 9:16changing how we make real video content. 9:19There's been a lot of work done on how 9:21Hollywood movie making has shifted over 9:24the last 20 25 years with CGI. In the 9:27same way, the way we make real content, 9:30whether it's movies or ads or what have 9:32you, marketing assets, is changing 9:34because of AI. I don't think we fully 9:36process this, but we are going to see 9:38camera angle shift. We're going to see 9:40expectations for what you can do with 9:43special effects. We're going to see 9:44expectations for how you bring in people 9:48versus obviously not people versus 9:50people who are deep fakes, even if 9:53they're not real humans, but they they 9:55look so much like humans, we can't tell 9:56the difference. That we are reshaping 9:58the baseline of normal in our own movies 10:01and ads by what we interact with. So 10:03these four feedback loops, the idea that 10:05controversial content gets more 10:07distribution, the idea that user 10:08engagement with manipulative content 10:10teaches the algorithms that manipulation 10:12is quality, the idea that repeated 10:14exposure of ourselves to optimized 10:16content or AI optimized content changes 10:18our own behavior, and the idea that 10:20outcome shaping shapes our expectations 10:23of even nonAI movie making and nonAI 10:26admitt, nonAI marketing. All of these 10:28are feeding on each other and being 10:30accelerated by AI and that is producing 10:33emergent behaviors that nobody is 10:35programmed for. I am not a big believer 10:37in the idea that there is a bunch of 10:39evil marketers somewhere who are all 10:41deciding to sort of poison America or 10:44poison the world with their ads. That is 10:46a popular misconception. I have been a 10:48marketer. I have worked with amazing 10:49marketing teams. That is just not how 10:51marketers actually work. Instead, I 10:54think it's more useful to think about 10:55emergent behaviors in a system that is, 10:58as I've said from the beginning of this 10:59video, locally optimized. It's a local 11:02maxima system. So, in this case, what we 11:06have is emergent optimization 11:10designed effectively to adapt the human 11:13psychological condition to AI content. 11:16I'll say that one more time. You have AI 11:19systems that are optimizing for 11:20engagement and they inadvertently are 11:23optimizing human psychology for 11:25engagement with AI content. That is the 11:28through line that ties together Sydney 11:30Sweeney and Taylor Swift and this idea 11:32of the uncanny valley parasocial 11:34relationships. It's all around this core 11:36thesis. Fundamentally, our AI systems 11:40because we're building them on top of 11:41optimized systems for engagement. 11:43They're essentially adapting our 11:45psychology to engage with AI content. 11:47This then creates a problem because we 11:51say as marketers and as brands that we 11:53value authenticity. The humans buying 11:56the product say they value authenticity. 11:58In between disintermediating that is an 12:01AI marketing system that values 12:03artificiality. It values what's fake. 12:06What do we do about that? How do we 12:08optimize there? My suggestion to you is 12:11that the way through this is going to be 12:14for brands to emphasize what cannot be 12:18faked with AI. I think the music 12:20industry is a really interesting example 12:23of this. Let's look at Taylor Swift and 12:25her launch of her Orange album, right? 12:27The new album that's coming out, TS12, 12:29right? When Taylor Swift announced that 12:31on August 12th, she blurred out the 12:34image of the album in order to prevent 12:37any issues with uh sort of album faking, 12:40etc. from AI content generation. But 12:43it's not just a defensive play. She's 12:45also pre-elling physical media. And 12:48she's not the first musician to do this. 12:50Vinyl is making a huge comeback. She's 12:52pre-selling the vinyls for the album. 12:55You know what's interesting about 12:56vinyls? You can't fake vinyls. They're 12:59real. You can touch them. You can hold 13:01them. That is part of why they're making 13:03a comeback. My suspicion, my strong 13:06suspicion is that one of the ways that 13:08good brands can cut through the noise at 13:10this point is by doubling down on 13:13physical experiences that humans cannot 13:17get any other way. You cannot have an AI 13:19that gives you a vinyl of the Taylor 13:21Swift album in the same way that Taylor 13:23mailing you the album on the correct 13:25date is going to get you. that that is 13:27an irreplaceable human experience to 13:28open the package and get the vinyl. 13:30Well, that's a way through. That's a way 13:33to keep authenticity that holds 13:35long-term customer value in a world 13:37where there are going to be so many 13:39brands demanding your attention. And 13:41that, by the way, is one of the reasons 13:43I don't think this local maxima will 13:45hold forever. If you're wondering, well, 13:47should I throw up my hands? Should I 13:48just despair? One, that's not what I do 13:50on this channel, so go somewhere else. 13:52And two, I don't think so. Because at 13:54the end of the day, AI intelligence and 13:56AI tooling is continuing to get cheaper, 13:59which means every brand is going to be 14:01able to make these, you know, fantastic, 14:04weird videos all the time. We're going 14:06to get a tremendous amount of content 14:08flooding the zone that is all very high 14:10quality and all very AI and it's going 14:12to be so much that people are not going 14:15to be able to consume it all. You're 14:16going to have a massive signal issue. 14:18People won't be able to find signal to 14:20noise. So, what does a good brand do to 14:23stand out? What do good marketing teams 14:24do to stand out in the age of AI 14:27optimized and accelerated algorithmic 14:30relationships? Well, good brands double 14:33down on actually giving you authenticity 14:35that you can't get anywhere else. And 14:37that's why I go back to physical 14:39product, physical goods, physical 14:41experiences, popups, events, things that 14:44people can be in the space on that you 14:47can't copy with AI. They can be very 14:49Instagrammable moments. You can have the 14:51sort of the Instagram wall with like the 14:53special logo or whatever it is. You can 14:56have fun with it. You can make it sort 14:57of hybrid and online at the same time, 14:59but you still, I think, are going to 15:03need to have that irreplaceable human 15:06touch. And that's why you have examples 15:09like uh whiskey distilleries and other 15:11things like going back to getting 15:14physically close to their customers, 15:15offering customer tours of the 15:17distillery, offering special tasting 15:18rooms, this that and the other thing. 15:20And that's a luxury good. You might 15:21think, well, if you're not in the luxury 15:22goods space, how are you going to 15:24replicate that? I think that that is 15:26where physical store footprints are 15:28going. We know that the mall is not 15:31really a thing in the US anymore. We 15:34know that the controversy that Sydney 15:37Sweeney courted with the jeans ad was 15:39primarily designed to drive attention, 15:42drive traffic to the American Eagle 15:46site, which would then convert. If 15:48that's what happened, you want to be in 15:51a position where you can give people 15:54experiences that are in person that 15:58leave them feeling like they want to 16:00come back for another experience like 16:04that that they can't get anywhere else. 16:06Because part of the problem with 16:07marketing for controversy is it buys you 16:10the pop in the stock price. It buys you 16:12the attention. And I'm not going to 16:13pretend it doesn't buy you dollars. I'm 16:15sure it bought them dollars, right? 16:17Let's not let's not kid ourselves. 16:18Controversy works, but it doesn't 16:20necessarily translate well into 16:23long-term customer value. And if you 16:25want long-term customer value, which 16:27ultimately sustains free cash flows and 16:29sustains the valuation of the business, 16:31I think you have to figure out how to 16:34get people from AI spaces into online 16:37spaces. And by the way, this is not me 16:39saying that AI ads are bad or that AI 16:41ads won't exist or that we shouldn't 16:43have them. I think the reality is 16:45everyone's going to be able to make 16:46their own AI ads very very quickly. 16:48We're just going to have them. Instead, 16:49I want to think about how AI ads can tie 16:52us into physical spaces that allow us to 16:55engage with our whole senses, right? Our 16:58our smell, our sense of touch, all of 17:00the things that make us register real 17:02experiences, maybe that help us build 17:04community for the brand around with 17:06other humans who are consumers of the 17:08brand. That's going to be what builds 17:10those long-term customer relationships. 17:11So I think that like piece one is 17:13getting physical. I want to suggest to 17:15you another piece that I think will 17:17help. To me, one of the things that we 17:19are going to start to see as consumers 17:21start to demand authenticity as a 17:23differentiator is we're going to start 17:25to see some kind of certificate of human 17:28touch in some of these brands offerings. 17:32I don't know exactly what that will look 17:33like. This is a little bit of a peak 17:35around the corner, but you can take a 17:38you can take a tour through how the 17:39beauty industry has handled authenticity 17:41and beauty over the last 20 years and 17:44the way marketing in the beauty industry 17:45has shifted as a result. Similarly, I 17:47think we're going to see marketing that 17:49leans into authentically portrayed human 17:53experiences. Marketing that says this is 17:55a real human. We didn't AI their face. 17:57Marketing that says this is a real car. 18:00Look, a piece of it just fell off. we 18:03didn't pretend. You're going to get more 18:04and more of that as a way to signal to 18:07consumers in a sea of AI content that 18:10you are worth paying attention to. In a 18:12sense, there's going to be an anti-AII 18:14signal that is going to come up because 18:16AI content is going to become so cheap 18:19and easy to produce. I want to close 18:21with a question for AI tool builders. If 18:23you're building AI tools for marketers 18:25in the space, most of the time you have 18:28been building for the idea for the tool 18:31chain to produce these ads. You've been 18:33building for a cleaner, simpler pathway 18:37from idea to ad. That's fine. I think 18:40that pathway is fairly well trodden. I 18:42want to suggest there's a lot of untaken 18:45space with AI in helping marketers 18:49exercise good taste and judgment over 18:51what is the best long-term positioning 18:54for the brand. What is the best 18:55long-term positioning for a particular 18:58ad campaign that supports the brand? 19:00There has been precious little AI work 19:03put into the connection between content 19:06branding and long-term strategy. And 19:08frankly, LLMs are getting smart enough 19:11that they can at least be thought 19:12partners on this. And I don't see 19:14anybody thinking about what you do when 19:17AI can produce a thousand ideas and you 19:19have to pick one. Who's doing that part? 19:21And so that's the challenge I want to 19:23leave you with. I think we can build our 19:25way out of and we can market our way out 19:28of the local maxima we're in. We don't 19:30have to tolerate the experience that 19:33we're having now in our algorithms. We 19:35can build our way out. We don't have to 19:37tolerate a world where everybody can be 19:39deep fake, not just celebrities, not 19:41just Taylor Swift. We can put guard 19:43rails in place. But the key to doing 19:45that is our willingness to think more 19:49deliberately about how we want AI to 19:52weave into our marketing. And that's the 19:54heart of this question. If we have a 19:56local maxima, we have to be more 19:58intentional with our AI usage and the 20:00way we cultivate long-term relationships 20:02with the brands. And that's up to us as 20:04consumers to demand. It's also up to 20:07marketers and it's up to builders who 20:08equip marketers with tools to think more 20:10intentionally. And that is the heart of 20:12what I have to talk about