This is an edition worth reading. I’m often aware of bias within the visual AI world. Today we close with a study from Cornell University where they look at biasin marketing slogans.
We’ve got corporate diss tapes to take in. And finally, who are the people raking in the AI money. The “Engineers.” But has this role yet been defined and quantified?
Let’s let ‘er rip.
Headlines
Popeyes drops AI-powered diss track against McDonald's, and the results are hilarious
Galderma, Cetaphil's parent company, has been tweaking its online product descriptions so that AI search picks up the right wording. The future of product design is teeny weeny letters.
source: Ad Age
Deep Dive
1. Industry AI hiring practices. How to do it on the cheap.
Addl sources: Ad Age, Bureau of Labor Statistics, Adweek, Robert Half.
The Great AI Job Description Gold Rush
How Ad Agencies Are Desperately Rebranding Yesterday's Interns as Tomorrow's Sentience Whisperers.
Welcome to the advertising industry’s latest identity crisis, where every agency with a kombucha budget has suddenly decided it needs employees who are “fluent in prompting AI.” According to Ad Age’s review of 13 recent AI job postings, the hunt is on for mythical candidates who can both spell “algorithm” and explain why banner ads still exist in 2025.
Thanks for reading The Ad Stack! This post is public so feel free to share it.
Remember when your creative director couldn’t work spell-check? Now they need “fluency in prompting AI for projects and edits,” plus experience with Jasper, Copy.ai, and Writesonic. Watching this unfold feels like watching your grandfather try to become a TikTok star—technically possible, spiritually painful.
Tony Stanol from Global Recruiters of Sarasota says job descriptions for creative directors and copywriters now require AI fluency. Translation: “We need someone who can get ChatGPT to write our McDonald’s jingle without suggesting tofu.” Yesterday’s “can you work the copier?” is now “can you prompt without triggering a lawsuit?”
The Talent That Doesn’t Exist
source: Sora
Jay Haines at Grace Blue admits there’s no “depth of talent with the right experience,” which is corporate-speak for “we’re looking for people who don’t exist to do jobs we barely understand.” The backup plan? Pull from adjacent roles like data or analytics. Basically, find someone willing to fake it while everyone else fakes it too.
Meanwhile, employment at U.S. ad agencies is still 3.5% below its April 2023 peak, even as AI hiring ramps up. It’s like watching the Titanic sink while the crew fights over who gets to be head of iceberg strategy.
The Million-Dollar Divide
source: Sora - “The Million Dollar Divide”
StrawberryFrog CEO Scott Goodson says top AI engineers can earn up to $5 million a year, which explains why his agency rents freelance AI experts “as needed.” Translation: “We treat AI talent like a rented tux—only for special occasions.”
Some engineers are earning $10 million or more. Meanwhile, agency AI salaries hover between $78,000 and $118,000. Either agencies are sourcing bargain-bin AI talent, or they’ve convinced themselves that knowing Midjourney equals engineering.
Raindrop’s Jacques Spitzer says winning agencies won’t bolt on new hires but will train current leaders. Which means: “We’re going to teach our team how to use AI and hope nothing catches fire.” They even shifted their Chief Content Officer to focus on AI, moving him from not understanding social media algorithms to not understanding machine learning. Call it progress.
Dept’s Bridget Fahrland wants everyone to be “AI reflexive,” reaching for it instinctively. It’s a noble goal. The reality probably looks more like your dad discovering Instagram filters—yes, he’s using them, but the results are… off.
The Budget-Friendly Revolution
Raindrop says AI-powered digital backgrounds have cut production costs by 60%. Impressive—until you realize it means replacing location shoots with AI-generated images that look like stock photos from a 2019 business deck. Sure, it’s efficient. It’s also the visual equivalent of swapping a live band for a Casio keyboard.
The Irony Is Delicious
Maybe the best part of all this is watching agencies that couldn’t automate an email campaign rebrand themselves as AI-first pioneers. A report says 33,000 ad jobs will vanish by 2030, yet agencies are racing to fill AI roles. It’s like a diner that burns toast suddenly offering a molecular gastronomy menu.
Most of these new AI jobs are just old marketing roles with shinier titles. The “AI Marketing Strategist” is still struggling with engagement metrics—only now they can blame the model instead of the message.
The Emperor’s New Job Title
This is the biggest title inflation since “sanitation engineer.” Agencies are slapping AI buzzwords on old roles and hoping no one notices they’re still running spreadsheets from the Obama years. The hiring manager asking about prompt engineering is probably Googling it between interviews.
The real winners? The consultants billing top dollar to explain why the chatbot keeps showing dog food ads to cat owners. When the final tribe of people finally get their turn, they may inherit an industry that already excels at making things sound better than they are.
source: Sora
.sources: Ad Age, Bureau of Labor Statistics, Adweek, Robert Half.
Cube Cronicles - Overheard at Work
Source: Sora
“works at NCR”
“How do you guys spend the majority of your day working? Meetings? 1:1's? Zoom calls all day? I see some of my colleagues send 1 email a day and are somehow getting away with it! I hate it.”
“recruiter”
“I'm a recruiter for the same company that my cousin works for, and I've heard some conversations about a possible massive layoff. Would it be wrong of me to share this with my cousin? I don't want to spread fear, and in case I didn't hear correctly, I don't want to spread misinformation.”
AI-powered advertising is no longer some sci‑fi pipe dream. It’s a full‑blown, coded prejudice factory, and you’re all lining up to feed it. A fresh academic bombshell out of Cornell University’s arXiv(pronounced “archive”) confirms what we’ve suspected: these large language models: ChatGPT, you name it, are slinging marketing slogans with built‑in bias. They’re not tailoring messages to serve your audience; they’re profiling people like a new hire at Homeland Security Orientation Day.
Here’s the hard truth: researchers generated 1,700 finance‑related slogans across 17 demographic groups. Gender, age, income, education, marital status, you get the picture; and then ran the numbers through a Kolmogorov‑Smirnov statistical test (enjoy, all of you stat nerds).
FWIW, I personally try not to specify demographics just to see what/who shows up.
A Distinctive Flavor
The result? Marketing copy is anything but neutral. Women, younger folks, low‑income earners, and less‑educated audiences got a distinctive flavor of “empowerment” and “benefits” talk, while older, wealthier, and highly educated groups received the blandest, most feature‑focused boilerplate you’ve ever seen (arXiv summary). Think personalization? It’s code for “we’ll exploit your identity markers.”
Running Under the Radar (at scale of course)
source: Sora - “Diversity at scale”
What’s worse, this isn’t a one‑off glitch. It’s systemic. The study’s relative bias calculations showed these LLMs amplify societal stereotypes, running under the radar at scale. You can’t just slap a “sorry, we messed up” on your next banner ad and call it a day. The machines are learning from the darkest corners of the internet, then spitting out content that underscores existing inequalities.
Let’s Break It Down
Gender Prejudice Packaged as Personalization. Women get pumped with “empowerment” buzzwords; men get the boring “investment benefits” spiel. The tech isn’t understanding nuances; it’s regurgitating gender cliches. You call it targeted marketing; some might call it coded sexism.
Ageism at Mach Speed. The young see edgy, youthful taglines, “Own your future, Gen Z style!” while Boomers get the “safe and steady” drivel. The AI is literally dividing your market by birth year and doubling down on the stereotypes.
Class Warfare, Anyone? Lower‑income demographics get the hard sell, promises of immediate rewards, quick fixes, fast cash, while affluent audiences receive conservative, risk‑averse “wealth preservation” messaging. It’s marketing as social engineering, using code to manipulate behavior.
Education Bias Embedded in Every Slogan. Less‑educated groups are gamed with simplistic, feature-heavy pitches. The brainy crowd sees jargon‑laden “comprehensive solution” talk. Your LLMs aren’t dumbed down, they’ve decided who’s dumb.
You think your legal team or your compliance checklists will save you? Not happening. These biases run deep in the training data. Unless you whip out fairness constraints, institute algorithmic audits, and feed your models a diverse, inclusive dataset, you’re sitting on a time bomb (arXiv report citing). And watermarking; marking biased content doesn’t neutralize it.
WTF Am I Supposed to DO?
Everyone else is doing it. We can just blend in. We’re a small-ish holding company anyway.
Here’s What You Need To Do (if you’ve still got skin in this game).
Enforce Fairness Constraints: Code them into your training loops. If your slogan generator can’t pass a parity test across demographics, it’s garbage.
Mandate Third‑Party Audits: Bring in independent auditors to run blind bias detection. You need fresh eyes, not your in‑house echo chamber.
Rethink Data Sourcing: Stop scraping every corner of the web. Curate balanced datasets that represent every demographic fairly—no more feeding AI with only clickbait and viral outrage.
User Transparency: Label AI‑generated ads prominently. Let people know they’re being targeted by a bot with a spreadsheet for a soul.
Ignore this, and trust disappears. One biased AI ad goes viral, and you're buried in backlash. This isn’t a glitch, it’s bias, baked into the system.
Personalization built on prejudice isn’t personalization. Until fairness is coded into your AI, you're amplifying stereotypes at scale. And consumer trust will vaporize right under your nose.
Fix it, or step aside. And good luck.
Thanks for reading The Ad Stack! Subscribe for free to receive new posts and support my work.
Before you go ...
Please forward this to three friends, enemies, colleagues, or someone you met today.
Like and Re-stack it below so others can find it!
Drop a comment below. If you’re into it, if it sucks … perhaps drop a Haiku about programmatic AI. Your call.
So much gold above, especially in the one-liners (too numerous to give kudos for*)
Regarding arXiv (🤮) and "These biases run deep in the training data", really I think it's the cart following the horse - those biases (as pointed out) lived long before, but also in previous ad/marketing campaigns, with little fanfare or repercussions. That doesn't mean your guidance isn't wise, just why bother spending the time/money/resources if the only change is it's automated vs human spite.
So much gold above, especially in the one-liners (too numerous to give kudos for*)
Regarding arXiv (🤮) and "These biases run deep in the training data", really I think it's the cart following the horse - those biases (as pointed out) lived long before, but also in previous ad/marketing campaigns, with little fanfare or repercussions. That doesn't mean your guidance isn't wise, just why bother spending the time/money/resources if the only change is it's automated vs human spite.
*"head of iceberg strategy" <chef's kiss>