Somewhere in your drafts is a message you have rewritten four times. It is to someone whose work came back wrong, or late, or almost right, and every version is softer than the last. By the fourth pass it is so padded that the person reading it may not be sure anything is wrong at all. And yet it will still sting, because people can hear a verdict through any amount of cushioning. You will have spent forty minutes building a message that is somehow both painful and useless.
That is not a character flaw. It is what writing does to feedback, and once you can see the mechanism, you can stop fighting it with more adjectives.
Feedback makes people worse more often than you think
In 1996, psychologists Avraham Kluger and Angelo DeNisi published a meta-analysis of decades of feedback research, pooling hundreds of studies in which someone was told how they were doing and then measured again. The headline finding was unsettling. On average feedback helped, but in more than a third of cases it made performance worse. Not neutral. Worse.
Their explanation, which they called Feedback Intervention Theory, is the single most useful idea you can carry into any message you are dreading. Feedback directs attention, and attention can land at three levels. It can land on the task itself: what the work was supposed to do and where it fell short. It can land on how the person is approaching the task: their method, their effort, their plan. Or it can land on the self: who they are, what they are like, what you think of them.
The further feedback pulls attention toward the self, the less it helps. Once someone is thinking about whether they are competent, whether you are disappointed, whether this means something about their standing, almost none of their mind is left for the actual problem. Feedback that stays on the task tends to improve things. Feedback that touches the self tends to produce defensiveness, rumination, or a quiet decision to stop trying.
Here is the uncomfortable part. "You need to be more careful with this" is self-level feedback, even though it sounds mild. It describes a trait. "The totals in column C don't match the invoice; can you reconcile them before Thursday?" is task-level, even though it names a mistake more bluntly. The second one is both kinder and more useful, and it is not the one most of us write.
Why writing pushes you toward the worst version
There is a second finding that explains the four drafts. In 1970, Sidney Rosen and Abraham Tesser described what they named the MUM effect, for Minimize Unpleasant Messages. People are reliably reluctant to deliver bad news, and when they must, they delay it, dilute it, and hedge it. The reluctance has nothing to do with the recipient's needs. It is about the deliverer's discomfort.
Speech has a built-in corrective for this. When you tell someone across a table that a section is confusing, you watch their face. If they flinch, you add context. If they nod, you move on. You calibrate in real time, and the result is usually specific and short, because the other person's presence keeps dragging you back to the thing itself.
Writing removes the face. With no one to calibrate against, you hedge in advance, for every possible reaction at once. You add "just" and "a bit" and "I might be wrong." You open with praise that you did not plan to give. You bury the point in the third paragraph. Each cushion lowers your discomfort a little and raises the reader's uncertainty a lot.
And uncertainty is not experienced as neutral. People reading an ambiguous message about their work tend to resolve the ambiguity toward the threatening reading, a tendency researchers call negativity bias. So a note that was softened to protect someone's feelings arrives as "they are unhappy with me and won't say why." That is the self level. You wrote fog, and the fog got read as a judgment of character.
Written words also sit still. A spoken "this part is off" lasts a second and is gone. The same sentence in an email can be reread ten times, each time with a slightly darker inflection. Anything you write about someone's work will be examined more closely than anything you say, which is one more reason the words should point at the work and not at the person.
The shift: describe the gap, not the person
Helpful written feedback does one thing. It makes the distance between what was delivered and what was needed visible, specifically enough that the person can close it without guessing what you think of them.
That usually takes three moves, and none of them involves an adjective about the human being.
First, name what you saw in the work. Not a summary, not a feeling about it, the actual observable thing. "The intro runs eleven paragraphs before the reader learns what the product does." "The deck uses three different fonts." "The client email went out without the pricing attachment."
Second, state the standard it was measured against, so the gap is between the work and a bar rather than between the person and your approval. "The brief asked for the value proposition in the first screen." "We're aiming for one typeface so it reads as one voice." Standards can be questioned, negotiated, and met. Approval cannot.
Third, say what happens next, as a specific ask with a time attached, and leave room for them to answer. "Could you cut the intro to two paragraphs and send it back by Wednesday? If something in the brief made that hard, tell me and we'll adjust the brief."
Notice what is absent. There is no "careless," no "you always," no "I expected better." There is also no praise sandwich. Kluger and DeNisi found that praise is itself self-level attention, and wrapping a correction in it does not neutralize the correction. It just makes the praise untrustworthy forever after. If you genuinely admire something, say so in a separate sentence that could stand on its own, and let it be true.
One more test, and it is the fastest one available. Before you type anything, say the feedback out loud as if the person were in the room. Your mouth will not produce four paragraphs of hedging, because talking to a face does not permit it. It will produce something like "the numbers in column C don't match the invoice." That sentence is the one to write down. Everything you were going to add to it was for you, not for them.
Your next moves
- Pick the one message you have been avoiding. Before you type, say the feedback aloud once, as if across a table, and write down the exact sentence that came out. That is your first draft, not your fourth.
- Run the self-versus-task check. Read your draft and cross out every word that describes the person rather than the work: careless, sloppy, brilliant, thorough, lazy, great. Replace each with a noun about the thing itself.
- Put the standard in the message. Swap "this isn't quite there" for "the brief asked for X; this has Y." If you cannot name the standard, that is a sign the feedback is really about your taste, and you should say so honestly instead.
- Delete the cushions and see what is left. Remove "just," "a bit," "maybe," "I hate to," and any opening praise you added after the fact. If the result is clear and specific, send that. If it reads as cold, add one sentence of real context, not padding.
- End with a specific ask, a time, and a door. "Can you revise the intro by Wednesday? If anything about this doesn't make sense, call me." The question keeps it a conversation instead of a ruling.
Say it first, then send it
The reason feedback comes out cleaner when spoken is that speech is anchored to a listener and writing is anchored to your anxiety. Quill is built for exactly that gap. It is on-device, private dictation that works in any app, so you can speak the feedback the way you would say it across the table and watch it land in the email or comment box as clean text, before the hedging instinct gets a chance to pile on. If the raw version sounds a shade too blunt in print, one tap rewrites it warmer or more formal without reintroducing the fog. The next time a draft is on its fourth rewrite, try saying it instead at quill.lumenlabs.works.