People who sent work emails in Justin Kruger’s 2005 experiments predicted their readers would pick up the intended tone 78 percent of the time. The readers managed 56 percent. Roughly a coin flip.
Now put an objection inside that email. Every guide on this topic hands you a phrase swap: “I disagree” becomes “I see it differently,” “that’s wrong” becomes “help me understand.” Softer words, same message, same result. None of them account for the fact that a disagreement typed into an email behaves nothing like the same disagreement typed into a Slack DM, a channel, a doc comment or a pull request review. Only one of those lets the other person answer you halfway through.
Softening the words is not the fix
Phrase-swap advice comes from spoken disagreement, where softening genuinely helps, because the other person is watching your face and ready to cut in. Move it into writing and it changes the wrapper while the mechanics stay exactly where they were.
The variable that actually matters is the channel. Each one decides two things before you type a word: how much of your case lands in front of the reader at once, and who else is standing there when it does.
This isn’t about winning your manager over (there’s a separate guide to pushing back on your manager), or about writing a clearer ask in general, or about claiming credit in a thread when somebody else has taken it, which has a script of its own. That script covers a claim on the record: one move, one channel, and the argument is over who did the work. This post is about objecting to a decision, run across five channels.
The numbers behind a claim every guide repeats
Every page on this topic asserts that tone doesn’t survive text. Almost none of them says where that comes from. It comes from a set of experiments by Justin Kruger, Nicholas Epley, Jason Parker and Zhi-Wen Ng, published in 2005.
They had people write short statements, some sincere and some sarcastic, then predict how accurately a reader would classify each one. In the email study, senders predicted 78 percent accuracy and readers hit 56 percent. A later study found the gap higher up the scale: senders predicted 88 percent, readers managed 62.8 percent, and face to face landed at 73.9 percent. Spoken aloud over the phone in the earlier comparison, the same statements landed at 73.1 percent. Our async guide covers why a friendly message can read cold on a screen.
The second finding is about who does the misreading. A study of 276 working adults across Australia and New Zealand, reported by SmartCompany, collected real workplace emails that had caused bad feeling and had both the recipients and uninvolved observers rate them. Recipients read the ambiguous ones far more negatively than the observers did, and the effect got stronger when the sender held more power. An absence of overtly negative features turned out to be a poor predictor of whether the reader would take it badly.
The same objection, rewritten for five channels
Five people, five channels, one objection each, all illustrative composites from coaching conversations rather than real individuals or accounts. Each “before” is what phrase-swap advice produces. Each “after” is shaped by what that channel does to your case on the way in. Assess your written communication first if you want to know which of the five you already get wrong.
Email: one reader, no interruptions, a crowded inbox
Miriam doesn’t think the onboarding survey should ship with the paywall screen inside it. The team will measure the redesign against completion rate, and the paywall drags that number down for reasons unrelated to the redesign. Colin owns the survey.
Before:
Hi Colin, hope the week’s going well. I’ve been thinking about the survey and wanted to share a few thoughts. I know a lot of work has gone into this and I might be missing context. I’m slightly concerned the paywall placement could affect our numbers, and I wondered about the question order too. No rush at all.
After:
Subject: Paywall inside the onboarding survey will move completion rate
Colin, one thing before Thursday. If the paywall screen stays inside the survey, the completion rate we measure the redesign against will include everyone who dropped at the paywall. That’s the baseline we agreed to judge the redesign on. Moving the paywall after the last question fixes it, if that’s cheap. If it isn’t, say so and I’ll work out what else we can use as a baseline.
One objection, one sentence on what it costs, and an exit. Miriam has three concerns about this survey. Two of them stayed in her drafts folder.
A Slack DM: send it in pieces and let them answer
Curtis wants to merge the two support queues into one rotation. Alicia thinks the SLA breaches visible in the smaller queue will disappear into the combined average. In a DM she doesn’t have to make that case in one go.
Alicia: before you write up the merge, can I check one thing about the small queue?
Curtis: sure, what’s up
Alicia: four of the five SLA breaches last month came from it. merged, that’s 5 out of about 400, so they stop showing in the weekly. is there another way we’d see them?
Curtis: hm. not really, the dashboard’s queue-level.
The first message is fifteen words and contains no objection at all. It buys Alicia the thing email can’t give her: Curtis answers before he’s heard the whole thing, so by the time the real point arrives he’s already in the conversation rather than receiving a verdict on his plan. Breaking it into pieces is the entire move.
A Slack channel: you’re asking six people to pick a side
“I typed the whole thing, read it back, and deleted it, because I’d written a message that made my own HR director defend herself in front of her team.”
Meredith’s pay review letters had been moved to the Thursday before a public holiday, and nobody was covering the People team inbox on the Friday. The channel had agreed to the date the day before.
Before:
I really don’t think sending the letters Thursday is a good idea given the holiday. Has anyone thought about what happens if people have questions?
After:
On Thursday’s letters: who’s on for Friday? If it’s nobody, anyone who opens theirs and wants a conversation waits until Monday morning. I’m fine with the date if someone’s covering. If nobody is, I’d rather we went the Tuesday after.
Nothing in the second version is softer. What’s gone is the demand that somebody defend the choice. The first version puts a question to the room that has one honest answer, and whoever picked Thursday has to give it in front of everyone who agreed with them yesterday. The second asks for a name on a rota, which anyone can supply without conceding anything.
A doc comment: what you highlight is what you’re objecting to
Our async guide has a template for feedback on somebody’s draft, and that’s a different job. Tessa’s Q4 plan doesn’t need improving. It records decisions, and one of them is wrong.
The line in the doc:
Q4 people ops: retire the tuition benefit (Nov), move everyone to the learning stipend.
Gordon knows the November date lands halfway through an academic year. He’s also the reason the people still on the old benefit were never moved across when the stipend launched, which he leaves out of the comment and Tessa knows anyway.
Where he puts the comment decides what he’s arguing with. Highlight the whole “Q4 people ops” block and he has objected to the quarter, and Tessa has to defend the quarter. Highlight the five words “retire the tuition benefit (Nov)” and the disagreement is five words wide, visible to everyone, and impossible to inflate:
Eleven people across three teams are mid-degree on the old benefit and the stipend starts them from scratch. Can Nov move to Feb, or can we carry the eleven to the end of their course?
No other channel does that for you. An email arrives whole, and the reader works out for themselves how much of the plan you’re unhappy with. A highlight settles it before they’ve read a word.
A PR comment: the only one of the five that can stop the work
Bennett’s pull request wraps a payment call in a retry loop with no idempotency key, so a timeout that actually succeeded gets retried and the customer is charged twice. Jonah has two instruments available and they aren’t interchangeable.
A comment joins the queue with everything else, and Bennett decides what to do with it:
This retry has no idempotency key on it, so a request that already succeeded server-side before the timeout gets charged a second time. Worth threading the request id through?
Requesting changes stops the merge until Jonah comes back and clears it. His disagreement now has a lock attached, and Bennett can’t route around it without escalating:
Requesting changes on the retry block: without an idempotency key this can double-charge on a timeout. Happy to pair on it this afternoon if that’s faster than a round trip.
Nothing else on this list hands you a lock. You can write the best email of your career and the thing still ships on Thursday. Use the lock where being wrong costs a customer money or loses data. Everything else, including the things you’re fairly confident about, is a comment.
The two signs the rewriting has stopped helping
You’ve started quoting their own words back at them.
Watch for the moment you paste a line from their earlier message above your reply. It feels like precision. It’s the one escalation move that exists only in writing, because you can’t quote somebody verbatim out loud, and it changes what you’re doing. You’ve stopped answering a person and started assembling a case against a transcript.
The second sign is quieter. Somebody who hasn’t written in the thread gets added to it. From then on you’re both writing for the person watching, and neither of you can give ground without doing it in front of them.
The standard advice is to move to a call, and our async guide sets out which conversations belong in writing and which don’t. Holloway’s practitioner guide for remote teams lands in the same place, then adds an option we’ve seen work less often than it should: longer writing rather than less of it, with both sides setting the whole thing out properly.
We don’t have a clean answer for what to do when the call isn’t available, which on a team spread across time zones is often. Gemma’s thread ran 31 comments over four days. They found an hour in week two. It helped with the tone. The decision didn’t move, and neither did what either of them privately thought of the other.
Why your objection arrives heavier than you sent it
Here’s the part that explains all five rewrites, and it has nothing to do with tone.
In a spoken disagreement the other person interrupts you. Almost always, and usually before you’ve finished. That interruption was doing something for you that you never noticed: it stopped you at one objection. You made the first point, they came back at it, and the second and third points you had lined up either turned out to be unnecessary or arrived as a reply instead of an assault.
Nobody can interrupt a message. The reader receives your complete case in one piece, which is why the polite version still lands hard. They’re reacting to the fact that you clearly finished the whole thought before they got a word in, and a complete case reads as a decision you’d already made about them. Your wording had very little to do with it.
So send less than you have. One objection, not your strongest three. Stop somewhere the other person can still change the outcome.
Read the five rewrites again with that in mind. Miriam sent one of her three concerns. Alicia sent fifteen words and waited. Meredith asked for a name on a rota. Gordon highlighted five words instead of a section. Jonah picked the instrument that gave Bennett an afternoon to fix it himself. None of them found a better phrase. Every one of them sent a smaller piece of the argument than they were holding.
There’s a test for this, and it costs nothing:
In coaching we ask people to read the message they’re worried about out loud, and it always sounds fine, because they read it in their own voice with their own pauses. Then we ask them to read it as the person who opened it at seven in the evening after a bad day. Most people stop somewhere in the second sentence.
Where they stop is the sentence to cut.
Run that on your own draft tonight. If you’d rather hear it from the other side, Merlin lives in Slack and Microsoft Teams and will read your message back as the person receiving it, bad evening included, then work the draft with you until there’s room in it for them to answer. Or start with the assertive communication assessment and bring the result to the message you’ve been putting off.