AI has a care problem

“Its very existence has made the world a worse place.” Oof.

That’s how Pitchfork opened its review of Tyga’s new album, $TARFACE. It gave the album a 0.0, the first time it had given a new solo album that score in 20 years. Pretty savage.

Okay, first some context. Tyga is a rapper. He has a new album. And he has admitted to using AI pretty heavily to make it. He says he wrote the lyrics and performed the vocals himself. According to Tyga, AI helped generate some of the production elements he wanted quickly, including specific 1980s textures and guitar parts. So calling the entire thing AI-generated would be overstating what we actually know.

But this does not appear to have been AI in a cute “let’s create a sample and play with it” kinda way either. Pitchfork’s reviewer claimed he was able to make a convincing approximation of the album by opening an AI music generator and typing in some version of “Tyga, 1980s, synthpop, new wave,” with Tyga’s name repeated a few extra times for emphasis. Basically, prompt the hell out of it until Tyga comes out.

Predictably, the internet did not approve. Reddit caught fire. Publications torched the album. YouTube poured on more oil. Even Doja Cat decided to get involved, which is generally not how you know a quiet week is ahead.

The reaction was not only about whether the music sounded good. Tyga has released poorly reviewed music before. Most artists have, assuming they have released enough music. This felt different because the process became part of the verdict. People were not only judging what they heard. They were judging how much human intention they believed was behind it.

It is a nice little exhibit in the growing library of evidence that we marketers have lost the pulse on how culture feels about AI. Or maybe have deliberately decided to ignore it.

We are having two different conversations

From inside a marketing department, the appeal of AI is almost comically obvious. It is fast. It is cheap. It can generate 30 directions before the meeting where someone asks for 30 directions. It can resize, translate, summarize, version, storyboard and produce a fairly convincing woman smiling near a laptop in the time it takes to find the stock photo of a fairly convincing woman smiling near a laptop.

The industry has responded accordingly. In a 2026 study from IAB and Sonata Insights, 83% of advertising executives said their company had already deployed AI in the creative process, up from 60% in 2024. Cost efficiency had become the top perceived benefit, cited by 64% of executives. Creative innovation came second.

Then the study asked those executives how they thought younger consumers felt about AI-generated advertising. Eighty-two percent believed Gen Z and Millennial consumers felt positively about it. The actual number was 45%.

That is a 37-point gap between what the industry believes about its audience and what the audience is saying about itself. More importantly, the gap is growing. Gen Z, the group marketers generally assume will shrug and accept whatever technology arrives next, was more negative about AI-generated advertising than Millennials. Thirty-nine percent of Gen Z respondents felt negatively about it, compared with 20% of Millennials.

Other research points in the same direction. Ipsos found that roughly three in four Americans preferred humans to create news and entertainment, while two in three preferred humans to create marketing and art. In another Ipsos study, 79% said companies should have to disclose when they use AI.

This does not mean audiences are universally anti-AI. The IAB study also found that disclosure can improve trust, and many people are perfectly happy to use these tools themselves. The divide is less about whether AI should exist and more about where each group encounters it.

Marketers experience AI from the production side. We see the time saved, the options created and the line in the budget that got smaller. Audiences experience it from the receiving side. They see another synthetic actor, another interchangeable post, another customer service bot that cannot understand why they are upset.

We see a machine that helps us make more. They see the more.

The platforms are cleaning up after themselves

The companies running the largest content platforms appear to have noticed. Spotify has introduced credits that let artists disclose specific AI contributions, including lyrics, vocals and production. It also created a new “Verified by Spotify” badge designed to signal that there is an identifiable, authentic artist behind the profile. Profiles that primarily represented AI personas were not eligible when the program launched.

The company also says it removed more than 75 million spammy tracks in a 12-month period. That number is often repeated as “75 million AI songs,” which is not quite what Spotify said. The removals included mass uploads, duplicate tracks, search manipulation and other low-effort material. AI simply made all of those tactics much easier to perform at scale.

YouTube has made its synthetic-content labels more visible and introduced internal signals that can automatically identify some AI-generated media. It says the label alone does not affect whether a video is recommended or monetized. The company is giving viewers more context while separately reducing repetitive, low-quality content.

And then there is LinkedIn, which now lets users flag posts that “seem like AI slop.” Its chief product officer explained that slop is difficult to define, so user reports help the company tune its models and improve the feed. LinkedIn gets to say, “We hate this stuff too,” while turning every annoyed user into an unpaid slop sommelier.

These platforms are not turning against AI. Their objection begins when AI becomes a product liability. They want the AI that creates new tools, engagement and efficiency. They do not want the AI that makes discovery worse, erodes trust and fills every available surface with vaguely professional oatmeal.

The response is convenient for everyone involved. Users get more control and better signals about what they are seeing. Platforms get labeled data that helps them identify, rank and filter content. The same companies that helped increase the supply of synthetic media now get to position themselves as the people protecting us from it.

There is another, more technical reason provenance is becoming valuable. Researchers have found that indiscriminately training models on material produced by earlier models can cause degradation over time. The phenomenon is called model collapse. A widely cited paper in Nature concluded that access to real human-produced data becomes increasingly important as model-generated material fills the internet.

That does not mean Spotify built its badges to protect a future training set, or that every anti-slop button is secretly a data-cleaning scheme. But the need is real. Users want to know whether something came from a person. Platforms want to know what they are distributing. Future models need ways to distinguish the internet from an increasingly convincing photocopy of the internet.

Yadda yadda, the snake eats itself. Yadda yadda, dead internet theory.

The shortcut has a smell

The more interesting line is not between AI and no AI. It is between AI as assistance and AI as substitution.

In research Ipsos conducted for the BBC, people were relatively accepting of AI used for supportive tasks such as playlists, background music, animation and visual effects. Resistance increased when AI replaced human creativity, voice or editorial judgment. Seventy percent preferred human-driven movies. Seventy-eight percent preferred human-written online news.

People will be inconsistent about this. They will complain about AI and then use it to write a difficult email five minutes later. I have done some version of this. You probably have too. The tools will also get better, which means many uses will disappear quietly into the process and nobody will notice.

The average person will probably never know, or care, that AI helped with research, rough cuts, cleanup, translation or production. Those are parts of the process we rarely asked the audience to admire in the first place. The reaction changes when the shortcut becomes visible in the final experience.

You can feel it in the actor who has no life behind her eyes. The founder post that sounds exactly like the previous six founder posts. The customer support bot that keeps cheerfully misunderstanding the emergency. The song that resembles an artist without containing much evidence that the artist needed to make it.

The audience does not experience any of that as efficiency. It feels like an absence. Someone could have cared more here.

This is why “people will get used to it” is not much of a marketing strategy. Maybe they will. Auto-Tune was treated as cheating before artists turned it into an instrument with its own aesthetics. AI will produce its own forms too. Some of them will be excellent. But brands are making decisions now, and customers are forming impressions now. Eventual acceptance does not repair a present-tense loss of trust.

The gap is the brand problem

That gap between how a company experiences a decision and how its audience experiences the result is where the brand problem begins.

At Dawn, we work with tech companies at moments of change. The product matured. The audience shifted. The category moved. The ambition got bigger. Inside the company, everyone can feel that it is no longer the business it was three years ago. Outside, the story still describes the old one.

AI is creating a new version of that gap. Inside the company, it feels like leverage. The team can produce more campaigns, more landing pages, more posts and more versions for more audiences. The output graph moves up and to the right. Everyone puts it in the quarterly deck.

Outside the company, the customer experiences something else. The copy feels generic. The images feel strangely familiar. The support interaction goes nowhere. The company appears to be saying more while revealing less of itself. The internal story is efficiency. The external experience is distance.

By the time a company comes to Dawn, it rarely lacks things to say. It has decks, campaign lines, half-finished positioning documents, pages added by different teams and a homepage trying to hold all of it together with the emotional stability of a folding chair. More output usually compounds the problem. What is missing is a decision about what the company actually means now.

That work requires figuring out what changed, what remains true, what deserves to be said and what can be removed. Then the story and identity have to express those decisions with enough specificity that the company feels recognizable again, both to the people inside it and to the market outside it.

AI can absolutely help with parts of that process. It can organize research, explore language, prototype a visual direction or expose a cliché before it reaches a presentation. Pretending otherwise would be dishonest and, frankly, a little boring. The tools are useful.

But usefulness does not make them responsible for the judgment. A model can produce a hundred taglines. It cannot decide which truth a leadership team should be willing to stand behind for the next five years. It cannot notice the hesitation before a founder answers a question, or know that the product has outgrown the story everyone still finds comforting.

Those are attention problems. Taste problems. Care problems.

AI is very good at making the production of average things dramatically more efficient. That is a genuine achievement. It is also a strange thing to expect an audience to feel grateful for.

The shortcut may disappear into the work. Fine. But when it becomes the work, people can feel what has gone missing.

‍