Intro
AI beat humans at advertising
A recently published research paper shows that AI is a whopping 19% more effective at creating ads than human beings.
While it's scary to see AI beating human beings at yet another task, it’s not all that surprising. Humans are notoriously bad at predicting advertising effectiveness. Expert marketers can barely beat a coin toss when judging a winning creative, and experts actually do worse than the general public.
There’s a reason the most effective creative emerge from iteration and testing, not a designer. The sad truth about advertising is ugly ads are often more effective than pretty ads – advertising works by capturing attention, a pretty ad blends in — a phenomenon related to what scientists call processing fluency — and can therefore be overlooked. An overlooked ad is an ineffective ad.
Meanwhile, GenAI has no cognitive preconception bias; it’s trained on existing content and the existing content is the content that has historically “won.” It’s no surprise GenAI is getting better at creating better ads, we’ve trained it to be good at ugly.
Industry Buzz
AI x Commerce in 2025
I’ve come across a number of clickbait titles recently titled “agentic shopping will never work,” or some variant of “agentic shopping is already here.” The reality, of course, is much more subtle.
Juozas Kaziukenas does an excellent job at breaking down the current state — and in my opinion, a very well-argued future-state — of agentic commerce.
Among others, his thesis is that agentic commerce won’t emerge fully-fleshed in a 0-1 leap. Like self-driving cars, it’ll emerge through gradual capability upgrades. Today, AI offers basic shopping assistance — roughly 6% of current ChatGPT queries, akin to early cruise control. This will quickly iterate in stages to full-blown autonomy.
Check out his great primer with a temperate view on the market.
Walled Gardens
Podcast: Meta’s Business AI tools and the ad-product divide
If you want high-level advice on how to buy on the Meta (or really any scaled ad) platform, check out this very interesting conversation between Eric Seufert and Meta’s VP of the Global Business Group, Simon Whitcombe.
One of my favorite bits of advice: Stop iterating on your successful creative, expand creative diversity to increase the unique people your ads reach.
AI
tl;dr: state of ads in chatbots
Confused about the state of advertising in AI chatbots? Me too. I researched the latest, so you don’t have to:
- OpenAI: caught distracted by space data centers, got scared by Google, issued a code red, delays ad testing. Later explains: code reds? totally normal.
- Google Gemini: claims "no ads, no plans to", immediately follows with announcement: "ads coming in 2026".
- Amazon: testing sponsored prompts in chatbot Rufus, claims a (suspicious) 60% increase in purchase intent.
- Walmart: testing ads in Sparky, their AI shopping agent. Walmart announced a partnership with ChatGPT for Instant Checkout in October.
- Shopify: launches native AI integration for easy AI enablement for its clients.
- Perplexity: started ads last November, paused them last month.
- X: claims they plan to include ads, but are instead busy banning the EU Commission’s account after receiving a fine.
- Other players? There’s a Lumascape for that.
Amanda's Take
AI disclosure - where is the line?
AI-only ads are now outperforming both human-made and human/AI hybrid creative, according to a new research paper featured on the Mobile Dev Memo Podcast this week. But there’s a catch: The moment people know an ad is AI-generated, performance drops. It’s a paradox. AI performs best until people realize it’s AI.
This hit home for me because around the same time, I received a Christmas card of a glamorous winter vacation, captioned: “AI… because actual winter adventures are too much work.” Before I knew they were AI, I thought, "wow." Once I did, the reaction quietly shifted to…"oh."
The law hasn’t fully solved this for advertising. As AdExchanger puts it, AI disclosure is still Wild West: a few mandates and a mix of state laws and requirements.
The stakes obviously vary. Using AI to place a shampoo bottle on a tropical beach isn’t the same as inserting someone’s face into explicit content. That’s why laws rightly exist around deepfakes and non-consensual imagery. But in the vast middle ground the question becomes fuzzier.
Where does consent actually enter the equation?
I think it depends on what we’re asking AI to do. I’m fairly comfortable with AI helping organize ideas, synthesize information, and accelerate production. But I’m not comfortable when AI is a substitute for human expression. Especially when we’re talking about using AI for art, such as movies or songs that imitate real lived emotion.
But I even notice I scroll faster past ads of obviously fake humans endorsing a product. It’s not that the output is bad; it’s a completely simulated experience. And the research suggests I’m not alone.
This isn’t really about AI vs. humans. It’s about trust, context, and expectations. AI can outperform in many cases, but it shouldn’t replace human expression. So my vote: use AI freely to speed up iteration, explore creative directions, and run smarter experiments. But when AI portrays a human claiming a lived experience or emotion, disclose it—or don’t do it at all. That’s where trust breaks.
Podcast
Inside a Designer’s Mind: The MarTech Operator’s Playbook
In this episode of HIGT, hosts Adam and Amanda sit down with Lindiwe Stenberg, newly promoted Chief of Staff to the VP of Marketing at Venmo (formerly Growth Marketing Operations Lead), to explore her unconventional career journey from multimedia design to nursing school to fintech marketing, her philosophy on continuous learning, and practical strategies for building high-performing marketing stacks and cross-functional teams.