You sound sharp on the call. Then you rewrite the Slack message four times.
This isn't a language problem. This is a medium problem.
Maria is a senior account executive at a European B2B SaaS company. She speaks English fluently. Her last two performance reviews praised her "excellent exec presence on calls." They also flagged "could be more concise in writing." She knows the written feedback isn't a real gap. It's the medium.
On a call, Maria is present. She laughs mid-sentence, corrects herself, lets thoughts unfold. People feel like they're talking to someone real. In Slack, something shifts. The sentences become more formal. The directness flattens. The voice that survives her accent, the confidence, the humor, the ability to think out loud, disappears into something safer but smaller.
This is the tax that async writing collects from non-native speakers. You're already code-switching all day. Async writing demands one more version of yourself.
The Problem: Async Writing Triggers a Different Kind of Self-Editing
Grammarly, ChatGPT, Hemingway, and other writing tools assume the issue is grammar or word choice. They're wrong.
Maria's English isn't the problem. Her grammar is correct. Her vocabulary is native-like. The problem is psychological, not linguistic.
Native speakers write in a register slightly different from how they speak, but the gap's small enough they don't feel it. They reread once, hit send, and move on. The voice stays theirs.
Non-native speakers hit a different wall. The moment you shift from synchronous (a call) to asynchronous (Slack), you become aware of yourself writing. You notice the gap between how you sounded on the call and how you read on the screen. You rewrite to close the gap. As you rewrite, the voice gets quieter.
The rewrite loop doesn't stop after one pass. It spirals. Three edits, four edits, five. Each pass tries to fix the discomfort of reading yourself as a non-native speaker in a permanent, written form. Each pass makes the writing safer. Safe writing feels smaller. So you rewrite again.
This is why Grammarly and ChatGPT make the problem worse. Grammarly offers grammar fixes that sound more "correct" but less like you. ChatGPT rewrites entire sentences, erasing your voice and replacing it with generic corporate English. Translation tools capture the literal meaning while stripping out the directness and rhythm that actually makes you heard.
Maria tried all three. She stopped using them. "They take my voice and give me back someone I don't recognize," she said.
Why Voice Tools Built for Transcription Don't Help
You might think: why not just use voice-to-text? Something like Otter.ai?
Otter.ai is excellent if you're recording a meeting or call and need a transcript. It captures what was said. But transcription tools assume the voice is the finished product. They're built to preserve the spoken word, not to help you compose async writing that needs to land differently than speech.
There's a category difference. Otter.ai hears what you said and writes it down. A voice composition tool hears what you're thinking and helps you shape it into async-native form: shorter sentences, stripped filler words, structured for scan-reading in Slack, still recognizably yours.
Otter.ai also lives in the cloud. It uploads audio, processes it remotely, and returns text. That latency, that journey to the server and back, breaks the momentum of composition. You feel the delay. The moment stretches. That stretched moment is where self-doubt creeps in.
The tools that exist for async composition on Windows, like Wispr Flow, are Mac-first, if they exist at all. So you end up with a choice: use a transcription tool that misunderstands the task, or go back to typing and hoping the self-edit spiral gets tired.
What Actually Happens: The Gap Is Latency, Not Language
The insight comes from watching how the best non-native speakers actually work.
On a call, Maria speaks once. The moment passes. She can't rewind and rewrite. So her voice comes out at its actual fluency level.
On Slack, she's got infinite time before hitting send. That infinite time is a trap. It invites doubt. It gives the careful version of yourself room to live.
If she drafts her thought by speaking first, the way she does on calls, the voice comes through immediately. The draft has filler words, maybe a false start. It's rough. But it's hers. Then she does a single, light editing pass to remove the obvious bits and clean up the rhythm. The message takes eight minutes instead of thirty.
One concrete example. Kristian's a senior product manager from Germany, 12 years in SaaS. He drafted a Slack message about a product roadmap decision. Speaking it aloud took 45 seconds. The rough draft had three run-on sentences and two "actually"s he didn't need. He removed those in two minutes. Total time: seven minutes. Without voice, Kristian said he'd have spent 25 minutes rewriting that message, cycling through three different framings before landing on something that felt safe to send.
The difference isn't the tool type. The difference is latency and context. Voice drafting that stays local, that doesn't bounce to a server, that knows it's composing async text not transcribing a call, collapses the time between thinking and writing. When the time's short, the self-editing loop can't start. When the loop can't start, the voice doesn't disappear.
What Changes When You Draft Async by Voice
Three shifts happen once Maria started voice-drafting her Slack messages.
First: speed. She went from 30 minutes on a single message to eight minutes. The self-edit spiral doesn't have room to exist.
Second: her team's perception. They described her Slack presence as "more like how she actually is on calls." This isn't because the writing improved grammatically. It's because the voice didn't flatten. Her directness, her humor, her ability to catch herself mid-thought and course-correct, the wit that survives her accent, all of it made it through to the async channel.
Third: her performance reviews. The "could be more concise in writing" comment didn't appear in her last review. Not because she became more concise. But because conciseness that emerges from a self-edit spiral reads as evasive or guarded. Conciseness that comes from actual clarity, speak once and light edit, reads as confident and direct.
Recitey works for this because it runs locally on your Windows device, so there's no upload delay or cloud processing to interrupt the flow. You speak, it drafts in Slack directly, you edit, you send. One voice. No careful version in between.
The Trade-Off: Speed and Voice Against Polish
Voice drafting's faster than typing, but the first draft's rougher. If you need formally reviewed prose, contract language, or pixel-perfect copy, voice might not be your first tool. The trade-off is acceptable if you're already deep in rewrites.
But if you're Maria, if you're spending 30 minutes in Slack because the self-edit loop won't release you, or if translation tools have stolen your voice, or if you know the gap between how you sound on calls and how you write isn't a language gap but a medium gap, then voice drafting isn't about grammar lessons or vocabulary expansion. It's about recovering your actual voice in async channels.
The tool's the vehicle. The real shift is this: the careful version of yourself isn't a feature. It's a tax, and you don't have to pay it.