When to use AI: How we decide what's off limits

For the last two years, the AI conversation has been deafening. It's taken over the tech space, and for those of us in B2B marketing, good luck having a single conversation that doesn't find its way back to the topic. How are you using AI? How should you be using it? Where do you rank on this shiny new “tokenmaxxing” leaderboard? It’s an ongoing discussion but it’s not always one with much nuance.

There’s not a lot of room in the discourse for “it depends.”

At Iron Horse, we're AI-forward. We've built AI into our workflows, we help our clients do the same, and we genuinely believe it's making our work better. But we're not AI-everything. (My recently adopted chiweenie office assistant, Dr. Susanne Bronner, was named without Claude’s help, thankyouverymuch.) 

That distinction matters more than most people give it credit for.

The process we use for deciding when not to use AI is just as deliberate as the process for deciding when we do. That decision tree factors into pretty much everything these days, even if it’s not always conscious. Here's a look at how we think about it.

How we actually decide

Let's get one thing out of the way: In 2026, it would be lying to say that any marketing project is 100% AI-free. The tools we use every day (project management platforms, design software, search engines we use for research) have AI baked in at some level. We're not pretending otherwise, and neither should anyone else.

Having said that, there's a meaningful difference between ambient AI and intentional AI. When we're scoping a project or breaking down a specific task, we ask ourselves three questions. They keep us honest about where we're adding value and where we're just adding automation.

Can AI make the process better?

A lot of the time the answer is yes. We've made real strides in project and resource management by intelligently incorporating AI into our workflows, and it's made us a sleeker, more efficient organization. The less-than-glamorous stuff — scheduling, allocation, status tracking, capacity planning, note-taking in meetings — runs smoother when you take the manual overhead out. Not to speak too much for our project managers, but I don’t think anyone misses updating the resourcing spreadsheet by hand.

We’re also helping our clients figure out where AI fits into their own operations. We’re seeing a pattern there, too. The wins tend to be immediate, measurable, and deeply unsexy. That's fine. Not everything needs to be a headline.

Can AI make the decisions better?

One of the biggest critiques of AI is that we lose something when we "let the robots make the decisions." For the record, we agree. But there's a version of this that people consistently overlook: letting the robots do the research so the humans can make much better decisions.

AI-led research gives us a wider view of a situation than any single person — even a deep subject matter expert — could hold in their head. It surfaces edge cases the experts might have missed. It synthesizes information from scattered, sometimes contradictory sources and gives us a fuller picture to work from. It lets us unlock a level of research and competitive analysis that we always want to do, but that can wind up getting cut short due to time constraints. The decision still belongs to a human. But the foundation that decision sits on? It's a lot sturdier than it used to be.

Can AI make the final product better?

For writing, design, and anything else where the details really matter, we’re still Team Human. The judgment calls, the creative instincts, the vibe check to know when something feels right, we’re backing good ol’ fashioned Human Brains every time.

But there are real ways AI makes the end result stronger without taking the wheel. Style guide compliance is a great example. Our teams have built custom AI agents and frameworks trained on our best practices, house style, and specific workflows. Those get run as a QA pass before a piece ever hits copy editing. They catch the obvious stuff (inconsistent formatting, missed punctuation, the things that slip through the cracks at 4 p.m. on a Thursday) so our editors can focus on the gray areas where human judgment actually matters.

(On an entirely selfish note, this means there has been a 600% decrease in Oxford comma-based edits on my first drafts. Couldn’t be more grateful.)

AI has also served as a powerful equalizer for our team. As a writer, rather than going into a brainstorming session with my ideas in my head, I can come armed with a deck that makes it easy for the rest of the team to visualize how my "Big Thoughts" translate into an end deliverable. Meanwhile, Uzair, being a meticulous researcher, might bring an array of data and expert citations I hadn't even encountered. AI allows us both to maintain our unique workflows, but it has significantly lowered the entry barrier for anyone to contribute to a project. 

This accessibility ensures the best ideas still rise to the top — it's simply become much more straightforward for everyone to participate in that exchange … and speak each other’s languages.

The lines we won't cross

So where do we pull AI out of the equation entirely? There are a few places where we've drawn firm lines. Sure, AI could do the work, but the cost isn't worth the convenience.

Final copy is written by humans

We’re far enough into this blog that I can be honest — calling AI writing slop can be generous. Bad AI content is so bad. And wherever you find your personal comfort level for AI usage, I think most of us are developing a sense for when AI makes its presence a little too obvious.

The text might be correct, but there’s no life behind it. It’s too perfect, but it somehow manages to say nothing.

There isn’t anything to be gained from copy that sounds like everyone else's. When you let AI be your company’s voice, you’re really creating two problems. With the people who can spot AI-generated writing and care about it, you lose credibility. For the people who don't particularly care, you lose the impact of originality.

Either way, you lose.

AI might help us research the topic, organize our thinking, or catch errors in the final QA pass. But the writing itself? That comes from a person who has a point of view and knows how to put it on the page in a way that drives our (and our clients’) audiences to action.

Big strategic calls are made by people

The strategies that actually change the trajectory of a campaign or a company come from humans connecting dots in ways that don't fit neatly into the patterns that AI is so good at creating. As more people rely on AI models for strategic thinking, the more AI gets trained on content that’s already been influenced. The feedback loop starts, flattening big ideas a little more each time they get passed through, and giving results that all start to look the same.

Iron Horse isn't in the business of offering rinse-and-repeat solutions. 

Applying our frameworks to our clients' unique situations takes human thought, every time. The ideas and strategies that work come from experience, the intuition built over years of watching what works, and the willingness to make a bet on something the data doesn't fully support yet. 

If it's coming from a person, it's from that person

Marketing already has enough trust problems without adding "was this written by a human?" to the list. If an Iron Horse email, Slack message, or LinkedIn post has a human name on it, a human wrote it. Full stop.

The moment your audience starts wondering whether they're talking to a person or a bot, you've already lost something you can't easily get back. Trust is hard to build and easy to erode, and "I think that was AI" is one of the fastest ways to screw it up.

That matters a little bit for something like a blog. It matters a lot more for thought leaders, strategists, and anyone who’s saying some version of “I would like you to give my company money so I can make your company better.”

“This might be AI” can easily be replaced with “this is a little more boring than I expected it to be” or “this seems like the most obvious idea possible.” Whichever variation you choose, not exactly how you want to be feeling when you’re considering making an investment. 

And honestly, avoiding that slight smell of AI is becoming increasingly tricky, even if you’re writing every word on your own. They’ve taken em dashes and the “not this, but that” structure. That’s, like, 95% of the way I was writing four years ago. 

When Claude figures out how to add references to my pup, and unnecessary, single-sentence paragraphs, I’m cooked.

But until AI literally takes our jobs, we’re making sure we keep our voices.

Drawing your own lines

This is our process today. Every company, every team, every marketer is going to draw their lines in different places, but these are ours.

Companies that haven't thought through their relationship with AI probably haven't thought through their process, either. The question of "where does AI fit?" is really a proxy for deeper questions: What do you value? What are you willing to trade off? What feels like a step too far? 

You know, the light stuff.

We know the lines we've drawn aren't universal. They come from our values, our experience with clients across industries, and our (sometimes hard-won) understanding of where AI adds genuine value versus where it just adds more.

We also know they aren’t permanent. Part of being AI-forward is being aware this space is changing rapidly. We’re always testing, always evolving, and always re-evaluating what feels right for us and for our processes. (Plus, that experimenting can be pretty fun.)

But if nothing else, we hope this gets you thinking about where your own lines are. Because the companies that are going to do this well aren't the ones that use AI the most, or the least. They're the ones that use it on purpose.

‍