The short answer
Muse writes in one polite corporate voice because that is its default, and it keeps writing that way until you show it your voice. The fix is a style profile built from your own sent messages, one per audience, plus hard rules about length and commitments, plus a short correction loop every time it drifts. In a discussion on October 14, 2026, a user who had spent two weeks letting Muse draft their message replies named the problem exactly: the drafts did not sound like them. This guide turns the fixes traded in that discussion into a method, with the exact prompts to copy.
Why does Muse sound like a company?
The October 14 report listed four pains, and anyone who has let an agent draft messages will recognise them. First, one voice for everyone: the same warm, polished tone for a manager, a client, and a best friend. Second, length: drafts that run long and explain themselves. Third, over commitment: drafts that agree to times, prices, and promises you never approved. Fourth, friction: approvals happen one message at a time, and the draft lives inside the Muse window, so sending still means copying it into the app where the conversation actually is.
None of this is a talent problem. Muse has simply never seen how you write. It fills the gap with the safest average voice it knows, which is the voice of business email everywhere. The moment you hand it real samples of your writing, and rules drawn from your habits, the average disappears and your patterns take over. That is the whole method: samples, rules, corrections.
There is a second reason the corporate voice is sticky. Politeness is safe. A draft that is too warm rarely offends anyone, so the default survives every casual test and only fails the real one, which is a friend reading a message that sounds like it came from a customer care department. Your standard is not whether a draft is pleasant. It is whether the person receiving it would ever guess you did not write it.
Step one: build a style profile for each audience
Gather 10 to 20 messages you have actually sent, for one audience at a time. Your messages to your manager are one profile. Your messages to close friends are another. Family group chats are a third, if you use them. Pull the samples from your sent folder, and take out anything sensitive before pasting; you are teaching tone, not sharing secrets. Users in the October discussion worked with 10 to 20 sent messages per context and kept the profiles separate, which is the detail that makes this work: one blended profile recreates the original problem in a new costume.
Then hand the samples over with this prompt.
Here are messages I have actually sent to [my manager / my clients / my close friends]. Read them and write a style profile: how long my messages usually run, how I open and close, words and phrases I use a lot, words I never use, how direct I am, how I say no, and how much warmth I show. Save it as my voice for this audience. From now on, draft messages for this audience in that voice, and show me the profile so I can correct it.
Read the profile it writes back. Correct it in plain words where it misreads you. The profile is a living document, and the corrections you make now save you a hundred corrections later. Repeat the process for each audience you write to often, and give each profile a clear name you can call in a draft request: your work voice, your friends voice, your family voice. Naming matters, because a named profile is one you can summon in a single line: draft this in my work voice.
Choose the samples with a little care. Recent messages beat old ones, because your voice this year is not your voice from your last job. Include the unglamorous kinds: the short reply under time pressure, the message where you said no, the one where you fixed a mistake. A profile trained only on your most polished emails learns the polish and misses the person. Ten to twenty messages that sound like your ordinary Tuesday will teach Muse more than fifty of your greatest hits.
Step two: set the hard rules
A style profile captures how you sound. Rules capture what you never do, and they matter more, because the costliest draft errors are not tonal. They are commitments: a time agreed, a price implied, a promise made in your name while you were making tea. Set the rules once, in the same conversation as the profile.
Rules for every draft, whatever the audience. Keep it under 40 words unless I ask for longer. Never agree to a time, a date, a price, or a promise without asking me first. Never add a greeting or a sign off that I do not use. Match the voice profile for the audience. If you are unsure what I would say, give me two short options instead of guessing.
Adjust the numbers to your life. If your messages run to three lines, say three lines. If you would never ask a client a question in a first reply, write that down. Rules are cheap to state and expensive to skip: every rule you set is a category of correction you stop making.
Other rules users reach for, once the first drafts land: never use exclamation marks; never open with a greeting when writing to friends; if a reply needs more than the word cap, draft the short version and flag that a longer one exists; never soften a no into a maybe. Your list will differ. The pattern is what matters: each rule names a thing you caught yourself deleting, and deletes it forever.
Step three: run the draft, correct, teach loop
Users in the discussion described their working rhythm with a phrase worth stealing: draft, correct, send, teach. Muse drafts. You correct, in plain words, the way you would brief a person. You send the corrected version yourself. And you tell Muse what changed, so the next draft starts closer to you than the last one did.
That draft is too formal and too long. Here is how I would actually say it: [your version]. Remember what changed and why: shorter, plainer, and no promise at the end. Use that in the next draft for this audience.
Two habits make the loop compound. Correct the pattern, not just the message: if the draft added a pleasantry you never use, say so once, as a rule, instead of deleting it thirty times. And keep the approval step manual. The users who do this well are explicit that nothing sends itself. A draft is a suggestion wearing your voice, and the final tap stays yours, which is what makes the whole arrangement safe to use with real relationships.
Expect the loop to feel slow for the first few days. It is not. Every correction you make once is a correction removed from every future draft, and the arithmetic compounds quietly. The users who abandon the method usually abandon it here, one message before it would have started saving them time.
Step four: tune it for the people you write to most
Audiences are categories. Your actual inbox is people, and the people you write to every week deserve their own notes inside the profile. A line per person is enough.
For messages to [name]: keep it warm and short. We joke around, and a one line reply is normal between us. Never sound like an announcement.
This is where the drafts start feeling uncanny in the good way. Muse learns that your manager gets complete sentences and your oldest friend gets fragments, because that is how you already write to each of them. Keep the list short: the five or six people who fill your week. A long list dilutes the profiles; a short one sharpens them.
How do you use the profiles day to day?
The routine that emerges from the reports is short. Paste or forward the message you need to answer. Name the profile in the same breath: draft a reply in my work voice. Read the draft against two checks only: would I send this, and does it commit me to anything. If it passes, send it. If it fails, correct the pattern out loud, not just the sentence, and let the correction join the profile.
Two checks sound too simple for a method, but they map exactly onto the two failure classes users reported: drafts that sound wrong, and drafts that promise wrong. Tone failures get corrected into the profile. Commitment failures get corrected into the rules. After a few dozen cycles, most drafts pass both checks on the first read, and the ones that fail, fail in new and more interesting ways, which is what progress looks like with a writing tool.
What about humanizer skills?
The October discussion also traded a shortcut: humanizer style skill packs, pasted into Muse as skills, which push drafts away from generic machine polish and toward natural writing. One pack named in the thread is the blader/humanizer skills collection on GitHub. Treat these the way you would treat a borrowed style guide. A pack can sand down the corporate edges, but it cannot know your voice, because your voice lives in your sent folder, not in a repository. If you try one, run it alongside your own profile rather than instead of it. One caution belongs with any pasted skill: a skill is instructions Muse will follow, so read one before you install it, the same way you would read a contract written in your name. Our guide to Muse skills explains how skills work and how to add one, and the setup guide covers the memory and connector picture your drafts draw on.
The honest limits
- Approval is one message at a time. Users in the October discussion asked for an approve all queue and do not have one. Budget the taps into your routine.
- Drafts live in the Muse window. For apps Muse cannot post into directly, sending still means a copy and a paste. The friction shrinks; it does not vanish.
- Profiles drift. Samples age, roles change, and the corporate voice creeps back if you stop correcting. The loop is the maintenance, not a phase.
- This is about text. Matching your spoken voice is a different feature entirely, covered in our voice chat guide.
Set expectations by the calendar, not by the demo. The user whose report anchors this guide was two weeks into using Muse for replies and still frustrated, and two weeks is roughly when the profile method starts paying back its setup cost. The first days feel like teaching. Somewhere in the second week of corrections, the drafts cross a line: you stop rewriting them and start merely approving them, with the occasional pattern correction as maintenance.
Set against those limits, the gain is real. A user who fixes the voice problem stops editing every draft from scratch and starts approving most of them, which is the difference between an agent that drafts for you and an agent that drafts at you.
Get started
Want drafts that sound like you wrote them? The profiles and prompts in this guide run inside Muse, so the first step is a Muse account. If you are in the United States or Canada, get a code from the pool and start with your sent folder. Anywhere else, the waitlist holds your place in line for launch day.
Get a code Join the waitlistFrequently asked questions
- Why does Muse sound like a corporate email writer?
- Because the corporate voice is its default when it knows nothing about you. It has never seen your messages, so it writes in the safest average register it knows. Feeding it your own sent messages is what replaces the average with your patterns.
- How many of my messages does Muse need to learn my style?
- Users who have done this work with 10 to 20 sent messages for each audience. Fewer can work for a narrow context, like short replies to one team. Keep the sets separate per audience, or the voices blend back into one.
- Can Muse keep different voices for work and friends?
- Yes. Build a separate style profile for each audience, give each one a clear name, and ask for drafts by that name. The separation is the entire trick: one blended profile is what produced the corporate voice in the first place.
- Can Muse send the replies itself once it sounds like me?
- Keep the approval step anyway. The users who do this well run draft, correct, send, teach with no automatic sending, because a draft that sounds like you and commits you to something is worse than a stiff one. Approvals run one message at a time today.
- What are humanizer skills for Muse?
- Skill packs that users paste into Muse to push drafts toward natural, personal sounding writing. One named in user discussions is the blader/humanizer collection on GitHub. They can soften the corporate polish, but your own sent messages teach more than any pack can.
- My drafts are drifting back to corporate. What now?
- Refresh the profile. Add recent sent messages, restate the hard rules, and correct the next few drafts in plain words while telling Muse to remember the changes. Drift is normal over a month or two; the correction loop is the fix.
Built with Muse is an independent community project. Not affiliated with or endorsed by Meta.