The Quiet Risk Inside Your AI Marketing Stack

A friend of mine runs marketing at a mid-size brand. Last month she proudly told me her team now ships three times as much content as a year ago. I asked her how. She said, "We just stopped writing most of it. The machine does."

I nodded. Then I asked one more question.

"Who checks what the machine writes?"

She went quiet. And that silence is exactly the problem I want to talk about.

The Adoption Gap

Here's a number worth pausing on. Over 70% of companies are now putting generative AI to work in some business function. Campaigns, personalization, creative testing, strategy decks. It's everywhere.

And the rules governing it? Barely moved.

Think about what that means. Cars are multiplying on the road, but we're still writing the traffic laws in pencil. Companies adopted the tool faster than any institution could draw a fence around it.

That sounds scary. But here's the flip side. The lag in regulation is also a window. Right now, before the rules harden, you get to decide what kind of AI shop you actually want to be. You set the guardrails. You choose how the machine drives you forward instead of off a cliff.

Bias Is Not a Glitch. It's an Inheritance.

Let's talk about the first landmine. People assume AI is neutral because it's math. It's not. Every model is built by humans, trained on human output, and inherits human blind spots.

In 2022, a study found that when you prompted an AI image tool with the word "CEO," roughly 97% of the results showed white men. Ninety-seven. Not 60. Not 70. Almost all of them.

What happens when your campaign visuals come out of a tool like that, and your customer base doesn't look anything like the output? Your brand stops looking like your customers. And your customers notice.

It gets worse. A major HR software vendor is currently in legal trouble because a job seeker claims its AI screening tool filtered him out based on age, race, and disability. He never even got an interview.

Bias isn't theoretical. It shows up in your hiring data, your creative, and eventually your inbox. As a lawsuit.

The Bill Nobody Is Counting

Here's the part most leaders skip, because it doesn't show up on their dashboard.

Every day, over a billion prompts hit one chatbot alone. Just one. Add up the users, the image generators, the video tools. The electricity and water required to cool those servers is staggering. One cryptocurrency network already burns more power in a year than entire countries.

Nothing currently offsets that draw. And the demand curve only goes one direction. Up.

You can't fix the grid by yourself. But you should at least know the bill exists, and that your marketing efficiency has a hidden line item called "the planet."

Who Is Responsible When Nobody Checks?

This one is my favorite, because it's the most honest.

A McKinsey survey of 830 people using generative AI at work found something I can't stop thinking about. Respondents were about equally split between "I review everything the AI produces" and "I review nothing." Slightly more leaned toward reviewing nothing.

Let that sink in. Half the people pressing send on AI output have no idea what's actually in it.

And the law hasn't caught up to answer the obvious question. When something goes wrong, who owns it? The person who asked for it? The tool that made it? The vendor who trained it? Right now, that answer is a shrug.

If you run a marketing team, the answer needs to be you. Every time. A human reviews before it ships. Full stop.

The Problem With Owning Nothing

Now we hit the legal layer, and it's a strange one.

According to the U.S. Copyright Office, nothing created entirely by AI can be owned by a person or a company. Which means that gorgeous campaign visual your team generated last Tuesday? Your competitor can grab it tomorrow, and you have almost no recourse.

Some brands have already figured this out. They use AI to brainstorm and ideate. Then they go shoot real photos and real video, so the final asset is actually theirs. Smart.

There's a darker version of this problem too. Deepfakes. Someone's face, voice, or writing style lifted and used without consent. Recently a senior government official's voice and manner were cloned well enough to fool real people in real conversations. Once a deepfake is out there, you almost never get it back in the bottle. Sometimes it goes viral faster than you can draft a takedown notice.

Imagine a fake video of your CEO announcing a retirement. Stock moves. Investors panic. By the time anyone realizes it wasn't real, the damage is done.

So What Do You Actually Do?

I'm not going to pretend any of this is simple. But there are a few moves that separate teams that survive this era from teams that get surprised by it.

Run a bias and accuracy audit every quarter. Not once a year. Once a quarter. Check what the machine is producing against the world you actually operate in.

If you use AI anywhere near hiring, benchmark its decisions against your real hiring history. If the machine keeps filtering out the same kind of person, that's not a feature. That's a leak.

Track how your brand gets mentioned inside the major models. There are tools that show you the prompt-level context and where citations come from. If a model is quietly misrepresenting you, you want to know before your customers do.

Build a human review gate before anything AI-generated goes live. For medium and high-stakes content, get legal eyes on it too. Keep an audit trail. Make your AI vendors tell you what data they trained on and how they handle bias. Refresh all of this every year, because the rules will keep shifting.

And put a crisis playbook in the drawer before you need it. What do you do if a deepfake of your founder shows up on a Tuesday night? Who calls legal? Who decides on PR? Who has the authority to respond publicly? Figure that out now, at 2pm on a calm day, not at 2am during a fire.

One Last Thing

The ground under marketing is moving by the hour. The teams that win this stretch won't be the ones who adopted AI fastest. They'll be the ones who decided, early and clearly, how much risk they're willing to stomach, and then built a frame around it.

That frame has three legs. Accountability. Training. Guardrails.

Get those right, and you can use these tools with your eyes open. Skip them, and you're just hoping the machine behaves. Hope is not a strategy.

I don't know exactly what comes next in this space. Nobody does. But I'd rather walk into it with a map I drew myself than one the machine drew for me.