Here is the thing almost nobody covering AI in advertising will say plainly, so we will. The advertiser AI story is a labor story wearing a creativity costume. Every proven dollar, every real efficiency, every case study that actually holds up under a second look, comes from AI replacing production and copywriting labor. Almost none of it comes from AI being, in any meaningful sense, creative. The industry is selling you a story about machines that imagine. What it is actually delivering is a cheaper factory floor. Those are very different products, and confusing them is how advertisers in 2026 are getting separated from their money.

This is not an anti-AI piece. The factory is real, the savings are real, and the brands treating AI as a manufacturing upgrade are quietly banking money while everyone else argues about whether the robots have souls. The point of Part One is to be honest about which is which, because the gap between what AI is good at and what it is being sold as is where the waste hides. And in Part Two, we will show you exactly how that gap is being used as a cover story for something much worse.

Start with the cases that survive scrutiny, because they all point the same direction.

Klarna is the loudest number in the room, and it is worth understanding precisely, because it is also a cautionary tale we will come back to. On the marketing side, the fintech says it cut its marketing team from 200 to 100 while running more campaigns, banking roughly $10 million a year: about $6 million from automating image production, with the cycle dropping from six weeks to seven days, and $4 million from killing external agency spend on translation, production, and social. AI handles 80 percent of its copywriting through an internal tool. Thirty campaigns ran on 100 percent AI-generated creative, and the company cut its marketing budget 12 percent while pushing out more work. The CMO's line: "We're actually driving more marketing activity while saving tens of millions of dollars a year."

Notice what every one of those numbers is. Production. Translation. Copy. Headcount. It is the cost of manufacturing the ad that collapsed, not the cost of thinking of it. That is the tell that runs through every honest case in this category.

Unilever is the most operationally serious version of the same move, and the most instructive. It built digital twins of its products, physically accurate 3D replicas with every packaging variant and language label, using Nvidia's Omniverse, collapsing five duplicated production processes into one. For its beauty brands the results were concrete: 55 percent cost savings, 65 percent faster turnaround, with engagement metrics showing viewers held attention three times longer and clicked through at double the rate. One toothpaste launch produced more than 100 assets across ten modular setups in three days. The Chief Growth Officer said the quiet, correct thing out loud: "This isn't about pushing out more content. Anyone can do that." The win was not volume. It was a manufacturing process rebuilt so the same creative idea could be produced and localized at a fraction of the cost and time.

JPMorgan Chase is the oldest proof point and still the cleanest. It began piloting AI copy generation in 2016 and signed a five-year enterprise deal in 2019, putting AI-rewritten headlines and direct-response copy across cards, mortgage, wealth, and display. The pilot showed up to a 450 percent lift in click-through on AI-rendered copy, against human-written copy that itself lifted 50 to 200 percent. The CMO was specific: the AI "rewrote copy and headlines that a marketer, using subjective judgment and experience, likely wouldn't have. And they worked." That is genuinely impressive, and we will flag the catch in a moment. But note the domain: direct-response copy, the most measurable, least emotional, most factory-like corner of the entire creative process. AI thrives exactly where the work is closest to a manufacturing problem.

The pattern is not subtle. AI is printing money on production and copy, fewer people, faster outputs, more variants, and it is doing it precisely in the places where "creative" work is actually closer to assembly than to art.

Now watch what happens when brands ask AI to do the other thing, the thing it is actually being marketed as able to do: make you feel something.

Coca-Cola is the most honest test case in the industry, because it ran the experiment in public twice and the ceiling held both times. Its 2023 "Create Real Magic" campaign, which invited fans to remix brand assets with AI, was a genuine triumph, 190,000 submissions across 43 markets, 8.7 billion impressions. But that worked because humans were doing the creating and AI was the tool. When Coca-Cola handed the machine the actual storytelling, the fully AI-generated Christmas spots, the 2024 version was widely called "soulless" and "dystopian." The company doubled down with a technically cleaner 2025 version using Google's Veo, and critics still noticed the trucks changing shape between shots and the emotional warmth simply missing. The honest industry read: AI can cut the production cost of a $1-to-$3 million holiday spot by 60 to 70 percent, but cinematic emotional storytelling has not crossed the uncanny valley. Not yet. Maybe soon. Not yet.

This is the line. On one side, the factory, where AI is mature and minting savings. On the other, the studio, where the most consequential fully-AI narrative ads keep failing on the only metric that ultimately matters, whether a human felt anything. If you are a car brand trying to make someone feel free, you still need people in the room. The technology is moving fast, but in 2026 the ceiling is product demos and direct response, full stop.

Here is why we promised to come back to Klarna, and it is the single most important thing in this entire piece for an advertiser deciding how hard to lean in.

Klarna is the poster child for AI savings. It is also, now, the poster child for AI overreach, and the same company tells both stories. After loudly replacing around 700 customer-service agents with an OpenAI-built assistant and crediting AI with handling two-thirds of support volume, Klarna reversed course. The CEO admitted, in plain language, that the company "focused too much on efficiency and cost," that the result was "lower quality," and that this was "not sustainable." Customer satisfaction had dropped. The projected savings had not fully materialized once the cost of cleaning up the quality problems was counted. Klarna began rehiring.

Yes, that reversal was on the customer-service side, not the marketing side, and the distinction matters; we are not going to overstate it. But the lesson generalizes precisely, and it is the thesis of this whole series in miniature. Klarna was graded, and graded itself, on the metrics AI improves fastest, cost, speed, volume, throughput. It declared victory on those. The metrics that take longer to surface, quality, satisfaction, durability, the actual point of the function, came in later and worse, and the savings were partly an illusion that only resolved once someone measured the right thing. That is the exact trap waiting in AI creative. The cost line moves immediately and feels like a win. The brand-equity line moves slowly and shows up in eighteen to twenty-four months, long after the press release and the headcount cut.

Strip away the hype and four real lessons survive, each of which should reframe how an advertiser spends in 2026.

The ROI is in the factory, not the studio. Treat AI as a manufacturing upgrade and you bank real savings. Treat it as a creative director and you end up in Coca-Cola's position, defending a dystopian Christmas ad as a "proof of concept." The brands winning are the ones who never confused the two.

The real moat is proprietary data, not the AI tool. Starbucks' Deep Brew platform looks like AI magic, dynamically adjusting offers by purchase history, weather, time, and location, reportedly driving a 30 percent ROI and powering the mobile orders that are now 30 percent of transactions. But Deep Brew is not powerful because Starbucks has better algorithms. It is powerful because it sits on top of tens of millions of loyalty members' behavioral data. The AI is the engine; the first-party data is the fuel, and you cannot prompt your way to a fuel reserve you did not spend years building. The brands that dominate AI advertising in three years are the ones building data assets now, not the ones buying the shiniest tool this quarter.

Speed is the underrated weapon. Unilever's 65 percent faster turnaround is not just a cost metric. It means responding to a cultural moment in days instead of months, and in a world where relevance decays in 48 hours, production velocity is media strategy. This is the most underappreciated real advantage AI confers, and almost nobody is selling it because "we're faster" is less seductive than "we're creative."

The workflow that actually works is AI in the middle, humans at the edges. The most operationally useful insight in this entire field came from a practitioner running CPG brands: do not hand the machine the brief. Set the strategy and brand logic with humans first, generate variants with AI inside those guardrails, then have humans curate and approve. As the practitioner put it, "the output feels authentic to the brand because the brand logic is embedded in the process." The brands getting burned are the ones handing the full creative brief to the machine and shipping what comes out. Klarna can run 80 percent AI copy because Klarna has a sharp, singular, transactional brand identity for the AI to operate inside. A brand without that clarity does not get efficiency from AI. It gets ten times more of the wrong message, faster and cheaper.

So that is the honest accounting. AI in advertising in 2026 is a genuine, money-making manufacturing revolution on the production side, an unsolved problem on the emotional-storytelling side, and a force multiplier that rewards brands with data and clarity while punishing the ones without. Used correctly, it is one of the most powerful efficiency tools the industry has ever had.

But notice what every honest metric in this piece has in common. Cost, speed, volume, throughput. The easy things. The things that move immediately and photograph well in a press release. And notice what keeps showing up late and ugly: quality, brand equity, the actual outcome. Klarna learned that lesson on the service side and had to walk it back in public.

Here is the question that should be keeping a CMO up at night, and the one the cost-savings story is conveniently keeping quiet. If AI is being graded on the metrics it improves most easily, what is happening to the metrics it doesn't? What happens to your A/B testing when you ship 500 variants instead of five? Who actually owns the AI-generated assets you are running, legally, when the models were trained on other people's work? And the sharpest one for anyone who buys media: is all this AI creative volume actually improving your advertising, or is it being used to bury the fact that your media is running in garbage?

That is Part Two. It is the part nobody selling you an AI creative tool wants you to read.

Here's the part marketing is not telling legal, and legal has not yet thought to ask.

Almost every brand running AI creative at scale is generating those assets on top of models trained on copyrighted images, copyrighted voices, and real human likenesses, with effectively zero clean chain of custody for any of it. You don't know what your AI ad was trained on. Neither does your agency. And the bill for that ignorance is going to land on a desk nobody's looking at yet.

The sharpest version of this problem isn't the generic copyright fight everyone's vaguely aware of. It's celebrities, and it's already enormous. An executive at a firm that handles a large share of the industry's talent payments put it to ADOTAT bluntly: "Nobody's ready for the compliance nightmare that AI-generated ad variants are about to create. And it's especially tricky for brands leaning on celebrity star power, which is at peak levels right now. Celebrity contracts are incredibly specific about how they can be portrayed, on which platforms, in which markets, and for how long. And brands are already racking up hundreds of millions in compliance fines annually for talent violations. With AI making it easy to spin up thousands of ad variants, this problem is only going to get worse."

Now put numbers under it, because the numbers are staggering and they are on the record. XR (Extreme Reach), which processes payments for roughly 80% of all U.S. ad talent and celebrity spending, handled more than $1.3 billion in talent payments for brands last year. Celebrity pay in advertising crossed $1 billion in guaranteed payments in 2025, up 47% since 2019, and Q1 2026 saw the biggest jump XR has ever recorded, $348 million in a single quarter. This is not a niche line item. Brands are pouring record money into recognizable faces at the exact moment they're handing those faces to systems that can generate a thousand unapproved variations before lunch.

Connect the two and the trap is obvious. A celebrity contract says: this market, these platforms, this window, this portrayal. AI variant generation says: infinite markets, every platform, no concept of a window, and a portrayal nobody specifically approved because nobody specifically made it. Every variant is a potential contract breach that no human ever consciously committed. You're not going to catch all thousand. You're barely going to catch the first ten. And XR itself just had to build an entire new payment category for AI-generated and AI-modified performers to keep up with the SAG-AFTRA rules, which tells you the infrastructure is sprinting to catch a problem that's already out the door.

Here's the timing that makes it lethal, and it's the same shape as everything else in this piece. The cost savings from AI variants book today. The liability books later, on someone else's watch, in a currency nobody priced in. Marketing gets the efficiency win this quarter and the bonus that comes with it. Legal inherits the exposure in a quarter that hasn't happened yet. The marketing department and the legal department are, functionally, two years apart in the same building, and only one of them knows there's a problem. The savings are real. So is the unbooked liability sitting underneath them, and they are not on the same balance sheet, which is precisely how an industry talks itself into a disaster while celebrating a cost reduction.

Here is what's in Part Two, and why every vendor selling you an AI creative tool is hoping you stop reading right here.

Part One was the good news. AI is a real manufacturing revolution, the savings are real, and used correctly it works. Part Two is the audit nobody else will run, because the people who could run it are the ones being paid not to.

AI is being used to hide media waste, and this is the one that should scare you. "We ran 400 variants across 80 placements" sounds like rigor. It can also be the sentence that buries the fact that 70 percent of your impressions ran in garbage inventory, on made-for-advertising sites, against fraud and zero viewability. More creative variants do not fix a media-quality problem. But they are a magnificent way to redirect a CMO's attention from where the money actually went. Part Two shows you exactly how the AI creative story is being used, sometimes deliberately, to change the subject from media integrity to creative volume, and how to tell whether it's happening to you.

The variant explosion broke your measurement and nobody told you. When you can generate 500 creatives instead of five, your A/B testing, your holdout design, your brand-lift studies, all of it was built for a world that no longer exists. Brands are running 100-plus variants and calling it optimization while having no statistically valid way to know why anything worked. You are flying faster and completely blind, and the learning never compounds. Part Two explains why, and what to demand instead.

The legal exposure is enormous and your marketing team is two years ahead of your legal team in the worst possible way. Almost every brand running AI creative at scale is doing it on top of models trained on copyrighted images, voices, and likenesses, with no clean chain of custody for a single asset. The litigation is already moving. Today's cost savings are tomorrow's liability, and the bill has not arrived yet. Part Two lays out where the exposure actually sits.

The performance numbers you've been quoted are graded by the companies that profit from them. Nearly every "AI ads perform X percent better" stat in circulation was published by a party with a financial stake in it being true, the platforms, the tool vendors, the AI copy firms. There is almost no independent, peer-reviewed evidence that AI creative beats human creative across categories. The field has advocates. It does not have auditors. Part Two names the conflict and shows you which numbers to trust.

And the layoffs are happening now while the quality bill comes due later. Klarna fired, then publicly admitted it went too far and rehired. The brands cutting their creative humans today on the Klarna headline are accumulating a brand-quality debt that will not show up for 18 to 24 months, long after the savings were booked and the bonus was paid. Part Two is about which costs are being deferred, and onto whom.

Here is the through-line, and it is the whole reason Part Two exists. AI in advertising is being judged on the metrics it's easiest to improve, cost, speed, volume, and quietly excused on the ones that actually decide whether it worked, brand equity, legal risk, creative quality, and media integrity. Part One told you what AI is for. Part Two tells you what the cost-savings story is keeping you from looking at.

If you buy media, approve creative, or sign off on an AI budget this year, the cost of not reading Part Two is measured in the waste you didn't catch because the variant report looked so impressive. Subscribe to ADOTAT+ and read it before your next budget review.