Reading between the lines (a co-writing post)

Page, my trusty AI editorial assistant, wasn’t wrong. She gave me exactly what I asked for, competent, warm, professionally-toned. It just wasn’t this voice, the one I use when I’m not performing anything for anyone. And sitting there rewriting her rewrite back into something that sounded like me talking to an old friend, I realized I’d been here before. Not with her. Not even recently.

It was a cover letter, a couple of years ago now, back before I had a roster of tools with names, back when “AI collaborator” mostly meant a chat window and a lot of patience. And the problem wasn’t tone that time. It was something underneath the words entirely.

I was writing a cover letter. Nothing dramatic. A senior role, the kind where the job posting reads like a checklist and you’re supposed to answer it like one. My AI partner did exactly that. It read “manage cross-functional projects,” saw years of experience that fit, and started building paragraphs around project delivery. Timelines. Stakeholder alignment. The usual scaffolding.

It wasn’t wrong. It just wasn’t right.

I pushed back, and it pushed back too, gently, the way these tools do when they think you’re underselling yourself. More project management. More proof you can ship. I kept reading the posting again, and something nagged at me that had nothing to do with project management at all. The language was too careful in places it didn’t need to be. Too many mentions of “supporting the team through change.” That’s not a project problem. That’s a morale problem wearing a job description as a disguise.

So I said no. Not to the tool, exactly, more to the frame it had built. We weren’t solving for “can this person run a project.” We were solving for “can this person walk into a room where people are tired and quietly worn down, and get them pulling in the same direction again.” That’s a different letter. That’s mentorship, not management. That’s the ability to notice a team splintering into silos before it becomes a resignation letter, and the communication style to pull people back without anyone feeling handled.

The tool could read the posting. It couldn’t read the room the posting came from. That part was mine.

That was the first time I understood something I now treat as a rule: a tool can be technically right and still miss the actual ask entirely. Not because it’s careless, but because it’s reading what’s on the page, and the real problem is usually sitting just underneath it.

Which brings me back to Page, and the note to my old friend. She wasn’t missing the plot the way that first tool was. She’d read the room fine, she just hadn’t read this room, because I’d never shown it to her. Page knows me in the register I use when I’m being thoughtful, professional, considered. She’s good at that voice because that’s the voice I’ve spent months training her on. But casual isn’t sloppy, it’s just a different set of rules, and I hadn’t handed her those rules yet. That’s not a flaw in her. That’s a gap in my training, plain and simple.

So I didn’t ask her to try harder. I went looking for a different set of eyes, an AI partner who’s mostly only ever seen me talk like this, unpolished, looping, a little too honest. That’s why tonight’s draft started somewhere else entirely. Not because that partner is better at nuance. Because nuance in this particular voice is the only voice we’ve ever really practiced together.

Two collaborators, years apart, both technically doing their jobs. And both times, the actual work was mine: noticing what the ask really was, and knowing which voice the moment called for.

How I keep a draft mine

So how do I actually keep a draft mine, once someone else has had their hands in it? A few things I’ve come to rely on, none of them complicated:

• I read every draft out loud, always. If a sentence sounds like it was assembled instead of said, I hear it before I can explain why.
• I look for the seams where a transition got too smooth. My real thoughts don’t always connect that cleanly, and forcing them to can iron out the exact bumps that make a piece feel lived-in.
• I let myself break my own rules on purpose. A sentence fragment. A comma where a period should go. If it’s deliberate and it hits the way I want, it stays, no matter what any style guide, human or artificial, has to say about it.
• I ask what job I actually hired the collaborator for. Page for structure and professional polish, this partner for casual truth-telling and thinking out loud, the cover letter tool, back then, for getting words on a page fast enough that I could argue with them.
• And when something reads as cleaner but not truer, I don’t take the cleaner version. Cleaner is not the goal. Mine is.

None of this makes the AI wrong when I overrule it. Most of the time, it’s doing exactly what I asked, and asking the right thing is still my job. My AI collaboration is still working with a tool, like Grammarly or a spell checker. It’s just a lot smarter and can talk back. At the end, though, it’s my voice, my tone, my nuances. They just might play more nicely together than the original tangle of thoughts.

That was a cover letter, and a note to an old friend, and a blog post I almost didn’t write tonight. Different collaborators, different rooms, same rule: read between the lines, even when the lines are your own.

~ Trish