Leah MessengerI trained my AI to sound like me — turns out I'm kinda annoying? But, there are other learnings too.
Like most content marketers using AI, I've been training my tools to write in my tone of voice.
Great news: it worked. Bad news: apparently I say "actually" approximately 400 times per doc. My AI has absorbed this completely. It now opens sentences with it more confidently than I do, which I didn't think was possible, and yet... here we are.
Also, food metaphors. Everywhere. Do I constantly write copy when I'm hungry? Because it bloody well seems like it.
Okay, so here's the one that really got me.
I love em dashes. I have always loved em dashes. They've been in my content guidelines at more than one job, and part of that is professional convention, but in all seriousness... part of it is that I think punctuation is beautiful and I like how it makes a sentence look.
In my totally humble (and perhaps a lil bit content-nerdy) opinion, the em dash is a very useful and very pretty piece of punctuation.
Alas, then the great em dash witch hunt began.
You know the one — LinkedIn tearing down any piece of content using em dashes. A dead giveaway for 100% AI-generated text, they said. And truthfully, I panicked. I went back through my writing, told tools to remove them, and started second-guessing every sentence that called for one.
Fear not, I have since calmed down (enough tears were shed). I use em dashes where they are needed; I don't just scatter them around for fun. When the sentence asks for one, you will absolutely see one there — in all its glory. ChatGPT, on the other hand: line up all the semicolons across three sentences, and you'll make a mile.
ChatGPT, as a side note: please discover the semicolon. It's stunning, and you are missing out.
The real irony of all of this is that AI learned the em dash from me, from you, and from other people who have content on the internet. It's got to get it from somewhere. My well-trained AI tool knew me better than I knew myself; me flagging "too many em dashes" was a simple case of caring too much about what other people think, and ultimately, battling against my own punctuation habits.
...and this brings me onto a bigger point.
Now we've all started externalizing our voices, we have to listen to it (and wow, sometimes the truth hurts).
I used Claude to dig through months of my own chat history to build out Personal Instructions in Optimizely Opal. By adding just one detailed instruction, I taught the platform my preferences, tone to a T, and a way of working so outputs actually sound like you from the get go (and yes, I did say actually again).
It was, at surface-level, just another work task. It was also a full personality audit.
And yes, I still read everything, and I still edit everything. Human-in-the-loop is not a disclaimer that I'm bolting on to make myself feel better about my career choices; it's how the work actually gets done. No draft goes on without me in it, which is how I know exactly how many times the word "actually" shows up, because I am the one removing 380 of the 400 instances before it gets anywhere near publishing.
But somewhere between documenting the metaphor habit and adding "do not use the word 'leverage' as a verb" to my instructions, I started thinking about why this matters way beyond just saving time on edits.
This is it, guys. The stakes are so much higher than your word count. We are in the middle of what I'd call a voice extinction event. Part of my 'personality audit' was that I was pretty dramatic though, so let's just say: things are bad and getting worse (if we're not careful).
Scan through LinkedIn and count how many times you see "It's not X. It's Y" — it's giving serious competition for "in this fast-paced digital landscape". Now, BRB while all creative writers in the world throw up.
But this is the thing. Language models are trained to produce the statistical average of everything ever written, so when enough companies route their content and communications through them, every company starts sounding like every other company. The shareholder letters converge, the press releases converge, the earnings calls converge.
State of Brand refer to this as the Great Flattening, and I think that's a super powerful way of putting it — because that's how fully AI content leaves me feeling. F l a t.
Content marketing is no different. Three-quarters of PR professionals now use AI on the job. An Ahrefs analysis of 900,000 new web pages found 74% contained AI-generated content. 91% of B2B marketers increased their content output in 2025. So we're talking: more volume, same few models, faster convergence (/total extinction) of the unique word.
I am not above this problem, which is why the "actually" audit matters.
Realistically, a lot of people training AI on their voice either don't know what 'good' writing looks like, or don't have a distinct voice worth preserving. And I say this knowing full well that I use enough food metaphors to constitute a cookbook.
Our own research in our recent Passion-Pressure Paradox report found that 38% of marketers say coordination has swallowed the creative work. Julianne wrote about this pretty recently — the grief of being a marketer who still wants to love marketing, but just can't find it under everything else going on. The report puts it plainly: the passion for the craft hasn't gone anywhere, but the access to it has.
What that means in practice is that a lot of us haven't actually written anything that's distinctively ours in a long time. We haven't had time to.
Instead, we've written by committee; we've written to pass legal; we've written to satisfy the brief, hit the word count, clear the stakeholder review. And then we wonder why, when we try to train our AI on our voice, it produces something that sounds like everyone else's AI.
The committee-driven approval process produces AI-like output even without AI. Legal reviews the draft. Comms softens the edges. The chief of staff strips out anything too sharp. By the time the letter goes out, it reads like every other letter that survived the same process. AI took a problem that already existed and industrialized it.
Remember those personal instructions I mentioned? They're not magic, they're a mirror. And that's what we've got to watch out for.
The work AI replaces follows a pretty distinctive pattern: research, content production, campaign reporting, media optimization — in other words, tasks with inputs, processes, and outputs that can be systemized.
The work that AI can't replace is entirely different: the ability to sit in front of a CFO and make the case for a brand investment that won't pay back for three years. The ability to say no to a bad idea when the algorithm says yes. This kind of human-only context and knowledge are not skills that can be prompted into existence.
This is what makes voice so strategically important (!!). Not in the 'gotta stick to brand guidelines' sense, but in the 'does this actually sound like a human who believes something' sense.
The companies doing this best still use AI. They use it to scale a point of view that already exists. Where the flattening comes in is when you ask AI to generate the point of view itself.
Psssst, check out our AI use case discovery template for exactly that.
That's the line. That's where personal instructions become so much more than a time-saving hack. When my AI drafts something in my voice, it's not generating my opinions — it's holding the container for them so I can fill it faster. And then I read it, edit it, and remove the "actually"s.
AI can scale your voice. It can't create one (no matter what your non-marketing pals might think).
Our content strategy is increasingly built around direct thought leadership from the people actually doing the work. Not polished, not sanitized, not written by committee. The kind of writing that finds someone in the middle of their ordinary life and says the thing they needed to hear. That requires a distinct voice. Specificity. Someone to actually have opinions and put them on the page.
And right now, that person is rare — and getting rarer.
The point isn't to make AI sound like me. It's to make sure there's still a "me" worth sounding like. Train it well, and it holds the scaffolding while you do the part only you can do.
Even if that part, apparently, involves a lot of food metaphors.
The "actually" thing I'm working on.