I write fast and I write sparse. 'This is wrong, please redo.' I genuinely believed that was a kindness — clear direction, no wasted words, respect for everyone's time. Then my teammate Sarah told me, gently, over coffee, that she'd spent the better part of an afternoon convinced I was furious with her over a button color. I was not furious. I was eating a sandwich. The message took me four seconds to type and cost her four hours of low-grade stomach-acid dread.
The breaking point came the day one of my one-line Slack messages somehow metastasized into a meeting on my calendar titled 'quick sync re: communication.' That is the corporate euphemism for 'you have upset someone and now we will discuss it in a small room.' I decided that efficiency which reliably produces emergency meetings is not, in any meaningful sense, efficient.
Tone is a feature, not a luxury
Here's the thing I had honestly not understood: my messages weren't wrong. The content was fine. The button really was the wrong color and really did need swapping. What was missing was the context the reader didn't have — that I was rushed, not angry, and that 'please redo' meant 'no big deal, just swap it when you get a sec' rather than 'this is a catastrophe you personally caused.' Text strips out tone of voice, facial expression, and the small but crucial fact that I actually like the people I work with. Blunt text fills that vacuum with menace, and the reader's anxiety supplies the soundtrack.
So I built the Rewrite a Too-Blunt Message So It Lands Right prompt and run it through GPT-4o. I chose GPT-4o specifically because it's fast and cheap. This is a thing I do ten or fifteen times a day, often mid-conversation with the cursor blinking in the message box, and I cannot have a slow, ponderous model in that loop — I'd abandon it inside a week. I paste my draft, set the relationship ('a designer I respect and work with daily') and the tone I'm actually going for, which is almost always 'warm but direct.'
Two versions means I'm still the author
The reason this prompt works where a generic 'make this nicer' fails is the two-version output. Version A is basically my message with softer verbs and a 'thanks' dropped in where it counts — minimal, fast, still unmistakably me. Version B is a small restructure that sometimes leads with one line of context before the critique, which I'll admit is usually the better email but occasionally too much throat-clearing for a quick Slack ping. The point is I get to choose. I'm not handing my voice to a robot; I'm getting two drafts of my own intention and picking the one that fits the moment.
The [NOTE: ...] rule has saved me too, in a way I didn't anticipate. Once, my blunt original asked a teammate to 're-export the file,' and GPT-4o flagged [NOTE: which file? You didn't name one]. I had, in my haste, asked for a thing without saying what the thing was. The prompt caught a clarity bug, not just a tone bug, before I shipped the confusion downstream.
- Stops me from hitting send on things I'll need an apology meeting to undo
- Two-version output keeps me the author — I'm choosing, not auto-piloting
- Takes about twenty seconds, faster than re-reading my own message five anxious times
- The [NOTE: ...] flag once caught that I'd asked for a file without naming it
- 'Warm but direct' is the tone that fixed ninety percent of my problems
The side effect: my actual writing improved
The genuinely funny part is that my unassisted writing got better over time. After a few hundred runs I started seeing the softer phrasings before the model did. 'This is wrong, redo' became, in my own head, before I even typed it, 'Hey — think this one's the wrong color, can you swap it when you get a chance? No rush.' Same four seconds. Zero emergency meetings. The prompt trained the muscle and now the muscle works on its own most of the time.
I pair this with my Draft a Hard Email You've Been Putting Off prompt for the heavier stuff — the messages too consequential for a quick Slack rewrite, where the relationship itself is on the line. But for the everyday 'oops, that came out colder than I meant' moments, this is the one I reach for reflexively. Grab the Rewrite a Too-Blunt Message So It Lands Right prompt on Prompt Dock and run it on the last thing you sent that got a strangely cold reply. You might discover, like I did, that it was never about them.