Don’t Let AI Flatten Your Brand Voice

I read a company’s blog last week and couldn’t tell you a single thing about the people who run it. It was competent. Every sentence was fine. It had that smooth, faintly helpful, faintly weightless quality — the prose equivalent of hotel-lobby music. Pleasant, forgettable, and unmistakably written with a lot of help from a machine.

Then I read three more, from three different companies, and here’s the unsettling part: they all sounded like the same company. Same rhythm, same tidy transitions, same little summarizing flourish at the end of each section. Four businesses, four brands that presumably think of themselves as distinct, all converging on one voice. The voice of the average of everything.

The risk isn’t that AI will make your writing bad. It’s that it’ll make it the same — as everyone else’s.

There’s now research putting a fine point on this. Anil Doshi and Oliver Hauser ran a study, published in Science Advances, on people writing stories with and without AI help. The finding is a genuine paradox worth sitting with: AI made each individual writer more creative — the stories got better, especially for the less naturally gifted writers. But collectively, the stories written with AI were more similar to each other. Everyone’s work improved and converged at the same time. A separate line of research on AI and creative ideation — Anderson and colleagues, presented at the ACM Creativity & Cognition conference — found the same homogenizing pull: groups using the same model produced ideas that clustered together, less diverse than groups working without it.

And here’s why it’s so easy to sleep through. Each individual piece looks like an upgrade. The blog post is cleaner than what you’d have written tired on a Thursday. The email is tighter. Every local decision to let the machine smooth it out is defensible. It’s only in aggregate — across a year of content, seen next to your competitors doing the exact same thing — that you notice you’ve quietly sanded off the specific texture that made people feel they were dealing with you. The flattening never arrives as a decision. It accumulates as a series of small, reasonable conveniences.

I’m not anti-AI about this, for the record. I use it every day, including as a first draft for things like this. The point isn’t to write everything by hand like it’s 2019. The point is to know which surfaces are load-bearing for your brand and defend those on purpose. The odd turn of phrase you’d never find in a training set. The opinion stated a little too baldly. The joke that doesn’t quite land but is unmistakably a human reaching for one. The specific, slightly-wrong word that’s more true than the smooth correct one. Those are the fingerprints, and they’re exactly what a model, optimizing for the most probable next word, is built to file off.

So use the machine to raise your floor — to get past the blank page, to fix the clunky paragraph, to make the competent parts more competent. Just don’t let it near the ceiling. Keep a few things in every piece that only you would have written, that a competitor’s AI would never generate, that make a reader think oh, these are actual people. In a world where everyone’s marketing is getting smoother and more similar at the same time, sounding like a real, specific human is about to become the cheapest competitive advantage left.

What’s the one thing in your brand’s voice you’d never let a machine smooth over?

Sources

• Doshi, A., & Hauser, O. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances.
• Anderson, B. R., et al. (2024). Homogenization effects of large language models on human creative ideation. ACM Creativity & Cognition.