Hey there

I'm musing (ranting?) a few thoughts on AI content over here. It's become integral into what we do at Prologue, and yet I hate policing it. You know why.

You've started skimming past certain sentences.

You're expecting less out of the copy.

You're reading more to get less.

You may not have named it yet, but you know the feeling. A paragraph goes by, it's grammatically perfect, it seems to say something reasonable, and you retain nothing.

Not a dang thing.

The weird part is it's not exactly bad writing. It's average writing. Like, literally just the sum of all the averages.

Large language models are trained on the generic whole of the written internet. So left alone, they write like an average of it. That's the whole mechanism that makes it possible. Patterns in, pattern out. (Garbage in, garbage out?) Which means the model's default voice is also your competitor's default voice, and your vendor's, and the guy who just started a newsletter about the same thing you write about.

It can sound like anyone. Or, said another way, it sounds like no one.

Over the last year, readers have learned to spot it:

Antithesis constructions. "It's not just a tool, it's a philosophy." This one hurts a bit. Juxtaposition is a great tool at times, and can be used to sharpen your points. Now it's an instant red flag. The funny thing is most of these applications read like a prophetic-level insight without containing anything noteworthy. It's just a pedestrian comparison in costume.

Em-dash chains. The em dash is a fine piece of punctuation and I'm not giving mine up, but you know this one is favored by the machines. And they over use it.

Hedging openers. "In today's fast-paced digital landscape." This was bad copy before AI existed but AI uses it to clear its throat.

Inflated and frilled vocabulary. You know the drill. When someone sounds like they wrote a letter in the 1700s... "Lest we retreat." LOL. I like to keep David Ogilvy in mind. He said write like people talk. Nobody talks like this.

Symmetrical closers. Every piece lands on a tidy three-part summary and an invitation to imagine the future. AI loves a holy trinity as much as the pope.

None of these are or ever were wrong on their own. I've used every one of them and I'll use them again. The problem is the frequency. When a brand publishes AI-assisted content with no controls on it, the tells stack up, and the writing converges on the same place everyone else's writing is converging on. All of the sudden you have AI trope soup with your logo stamped on it (gross).

Here's the part that should bother you:

The content can be completely accurate and still fail. Correct, clear, professional sounds great, but anymore that means it's also interchangeable with your top competitors. You put time, real estate and payroll into publishing something nobody can meaningfully attribute to you.

The test that matters

  • Take a piece you published last month

  • Strip the logo, the name, the product references

  • Do the same for a competitor

  • Hand both to someone on your team

  • Can they tell it's yours?

Are you too afraid to do this test?

Do you already know the answer?

The best voice work is so specific to a brand that you can remove the name and still know who said it. Most content doesn't clear it, and AI didn't cause that. Brands have been making bad content themselves for years, but AI has helped proliferate this junk.

I've noticed a few things when seeing companies scaling AI copy.

Do any of these hit home or give you concern:

  • Do different people on your team produce AI output that sounds like different companies?

  • Are your actual opinions missing? Is the content agreeable about literally everything?

  • Is your team spending more time rewriting AI drafts than they'd spend writing from scratch?

That last one is a productivity killer. Teams adopt AI to move faster, then spend every hour they saved sanding the default voice off every draft. You've automated the typing, ramped up how much you can produce, then realize someone has to (or at least should) be proofing this stuff. It's a monster bottleneck right now for marketers.

Brand master Debbie Millman in the NYT recently, speaking to this exact challenge.

Believe it or not, the fix isn't choose to write less with AI or to only hire humans to write. Your team can and should be using AI to create at the very least content frameworks. But leadership needs to decide, in advance and in writing, what your brand does and doesn't sound like and what your core messages should be. It's not just "sound confident." You need rules an editor could actually check, such as:

  • never open with a question that lets the user answer and exit with a quick "no"

  • never more than one em dash per page

  • never the "it's not X, it's Y" move

  • here are the eleven words we don't use

  • no tentative language, it makes us look weak

Once you have the rules, you can teach them to AI (and your humans).

Your voice is whatever you can define clearly enough to enforce.

Everything else is the average.

Brad
Founder
Prologue

PS — We wrote about this at length and you can find it here. Our team used Premise and humans to condense this down for a newsletter.

Recommended for you

View all
caret-right