Clean up Wispr Flow dictation in one click
Wispr Flow transcribes but does not tighten. One hotkey sends the text to a model with your style guide, shows the result, and pastes it back on a second key.
Put the cleanup behind one key. You dictate with Wispr Flow as normal, then press a hotkey. The text goes to a language model with a short style guide built from your own past messages. A tighter version comes back in a small review window. You read it, press one more key, and it lands where your cursor was. I'm building exactly this for a team lead at Chimpy now, September 2026.
Four steps, twenty times a day
The team lead at Chimpy talks faster than they type, so they dictate. Chimpy is Europe's largest powerbank rental service, running since 2013, and the person I work with there runs a team on Google Workspace, Slack and Wispr Flow. Most of their day is messages. Slack to the team, email to partners, a note to a supplier.
Here's the problem in their own words: "I use Wispr Flow to dictate messages, but it only transcribes. It doesn't tighten up the language." So they paste the transcript into a chat window, ask for a rewrite, and copy it back out.
In full: dictate into the Slack box. Select all, copy. Switch to a browser tab, paste, type "tighten this up", wait. Copy the result. Switch back, paste, send. Four steps, and the switch to the browser is the expensive one, because Slack is still open behind it with three other threads moving. Gloria Mark at UC Irvine puts average attention on a screen at about 47 seconds before a switch, and finds that people interrupt themselves as often as anything else interrupts them. Each paste into a chat window is a self-interruption you chose.
Say you do it twenty times a day. That's twenty trips out of the app where the work is, and twenty trips back. The words come out fine. The trip is the waste.
What the one-key version does
The build is small on purpose. Here's the whole thing.
You dictate as before. Wispr Flow puts the transcript in the text field. You press one hotkey. The text in that field, or whatever you have selected, goes to a language model along with a short style guide. I use Claude for the rewriting step; the model matters less than the style guide does, and I'll come to that. The tighter version comes back in a small floating window next to your cursor. You read it. Press one more key and it replaces the original. Press Escape and nothing changes.
That's the rule I've given the build: one key in, one key out. No prompt to type. No tab to find. No clipboard to babysit. If it ever needs a third key, I've built it wrong.
Where the hotkey lives depends on where you write. Two honest options:
- A global shortcut in the operating system. It works in any text field on the machine: Slack, Gmail, a Google Doc, a CRM note. This is what the Chimpy build uses, because the messages go everywhere.
- A shortcut inside the app. Slack, for example, lets an app add a global shortcut to the message composer. It's cleaner if all your dictation happens in one place, and it does nothing outside that app.
Either way, the model never sends anything. It hands you text. You send it. That's the same shape as the meeting recap to Slack build I'm doing for the same client: the machine prepares, the person posts.
Why the review window stays
I could have made it paste straight back with no window. It would look slicker in a demo. I didn't, and I'd argue you shouldn't either.
A transcript can be wrong. Dictation software hears "Ola" as "Oslo", "fourteen" as "forty", a product name as three words that mean nothing. The model then tightens a sentence that was already wrong, and does it confidently, because that's what models do. Nothing in the chain knows what you meant. You do. Only you.
So the review window is not a safety feature bolted on at the end. It's the step where the only person who knows the intent checks the output. It costs you two seconds of reading per message. A wrong number sent to a supplier costs a good deal more.
The window is small on purpose, so the review feels like glancing, not editing. Most of the time you'll press the key without changing anything. Sometimes you'll fix a name. Now and then you'll press Escape and send what you dictated, because the model made it stiff. All three are the system working.
The style guide comes from your messages, not from the model
This is the part that decides whether the result sounds like you or like a chatbot.
Every model has a house voice. Ask it to "tighten this up" with no other instruction and you'll get the same polite, slightly formal, slightly padded paragraph everyone else gets. That's fine for a first draft. It's wrong for a Slack message to someone who has worked with you for three years and knows you don't write like that.
So before the build goes live, the client sends me twenty or thirty messages they've already written and were happy with. Slack, email, mixed. I read them and write a style guide on one page. How long the sentences run. Whether they open with the ask or with the context. Which words they never use. Whether they sign off, and how. Whether "thanks" goes at the top or the bottom. A short list of names and product terms so the model stops mangling them.
That guide goes with every request, invisibly. You never type it and you never see it. The model gets the transcript plus the guide, and its job is narrow: shorter, clearer, same voice, same meaning. Not more polite. Not more complete. It is not allowed to add a point you didn't make.
The guide is a text file. When your voice shifts, or you want a stricter version for email and a looser one for Slack, it's a five-minute edit, not a rebuild.
What Wispr Flow already does, and what this adds
Be honest about the overlap before you pay for anything.
Wispr Flow is not a dumb transcriber. As you dictate, it drops filler words and adds punctuation, so the raw transcript is already cleaner than what a phone keyboard produces. It also has a Command Mode: select text, speak an instruction like "make this shorter", and it rewrites in place. For plenty of people that's enough, and if it's enough for you, stop reading and keep your money.
Pricing, so you can weigh it: Wispr Flow's pricing page lists a Free plan with 2,000 words a week on desktop, and Pro at $15 a user a month, or $12 a month on an annual plan, with unlimited dictation.
What the custom step adds is three things, not ten:
- A fixed house style in one press. Command Mode does what you tell it that time. The hotkey applies the same one-page guide every time without you saying anything, so message forty sounds like message one.
- A review window before anything changes. Command Mode rewrites in place. The hotkey shows you the result first and waits.
- No prompt. Not typed, not spoken. One key.
If those three are worth a build to you, it's because you send a lot of messages and care that they all sound like the same person. If you dictate five things a day, they aren't, and Command Mode is the right answer.
What it's worth, estimated
I'll show the sum so you can swap in your own numbers. Nothing here is measured yet. The build is being built now, September 2026, and I'll replace these with real timings after thirty days in use.
Say you dictate twenty messages a day. Say the old loop of paste, wait, copy, paste ran 45 seconds each, which is about right when the browser tab is already open and you don't get distracted on the way. Twenty times 45 seconds is 15 minutes a day. Five days is 75 minutes, call it an hour and a quarter a week.
That's the arithmetic. It's modest and I'd rather say so than dress it up. The part I can't put a number on is the twenty context switches that no longer happen, and if you've read the Gloria Mark figure above you know that's probably worth more than the minutes.
Across the four builds for this client, my estimate is roughly five to six hours a week for one person, all four workflows added up, and all four sums are shown in the Chimpy case study. The dictation piece is the smallest of them. It's also the one they'll feel most, because it runs twenty times a day.
Do this this week
Three things, cheapest first.
Try Command Mode properly for two days. Select a dictated message, say "make this shorter and keep my tone". If the results are good enough, you're done, and it cost you nothing beyond the plan you already have.
Write your own style guide anyway. Pull ten messages you were happy with. Write down, in plain sentences, what they have in common. Keep it under a page. Even if you never automate anything, you'll paste it into whatever model you use and get better rewrites straight away.
Count the trips. For one day, put a tally mark every time you leave the app you're writing in to fix a message somewhere else. If the count is under five, you don't need a build. If it's twenty, the tally sheet has done my sales pitch for me.
If you're at twenty and want it built, the shape of the job is on the implementation page: a fixed price for the build, then a monthly retainer to keep it running. If you'd rather start with the whole picture of what a small team should automate first, that's here. Before I was a builder I was a plumber, and the habit that stuck is the same: diagnose before you buy the part.
Questions people ask
Does Wispr Flow clean up dictation on its own?
Partly. As you dictate, Flow drops filler words and adds punctuation, so the transcript is already tidier than raw speech. It also has a Command Mode: select text, speak an instruction, and it rewrites in place. What it does not do is apply one fixed house style every time without being asked, or show you the result before it replaces your text. That gap is what the one-key build fills.
How much does Wispr Flow cost?
According to Wispr Flow's pricing page (source: wisprflow.ai/pricing, read 5 September 2026), the Free plan is $0 with a limit of 2,000 words a week on desktop and 1,000 a week on iPhone. Pro is $15 a user a month, or $12 a month if you pay annually, with unlimited dictation. Growth and Enterprise plans add SSO and audit logs and start higher.
Why keep a review step if the point is one click?
Because a transcript can be wrong, and the model will tighten a wrong sentence just as confidently as a right one. It hears a name as a city, or fourteen as forty. Nothing in the chain knows what you meant. You do. Two seconds of reading in a small window is cheaper than a wrong figure sent to a supplier. Most of the time you press the key without changing anything.
Where does the cleanup hotkey run?
Two options. A global shortcut in the operating system works in any text field on the machine, so Slack, Gmail, a Google Doc and a CRM note all get the same treatment. Or a shortcut inside one app, for example a Slack shortcut in the message composer, which is cleaner if all your dictation happens there and does nothing anywhere else. The Chimpy build uses the global one.
How much time does the one-key cleanup save?
An estimate, not a measurement. Say you dictate 20 messages a day and the old paste, wait, copy, paste loop took 45 seconds each. That is 15 minutes a day, about an hour and a quarter a week for one person. The bigger cost is the 20 switches out of the app you were working in, which no timer captures. I replace the estimate with a measured figure after 30 days in use.
Sources
- 1.Wispr Flow: Pricing · Free plan 2,000 words a week on desktop; Pro $15 a user a month monthly or $12 annual, unlimited dictation
- 2.Chimpy: About us · Europe's largest powerbank rental service, since 2013
- 3.University of California: Can't pay attention? You're not alone (Gloria Mark) · Average attention on a screen about 47 seconds; people interrupt themselves as often as they are interrupted
- 4.Slack developer docs: Implementing shortcuts · Global shortcuts run from the shortcuts button in the message composer; up to 5 global and 5 message shortcuts per app
Written by Ivar André Knutsen
I build and run AI systems, internal tools and workflow automation. You work directly with me from the first conversation through implementation and support. About Ivar
Want this looked at in your business?
We look at where you want the business to go, what is slowing you down and where AI could make a useful difference. You get a clear recommendation: a tool to try, a focused automation, a broader system or a closer look at the process. Any build is scoped and quoted before work starts. Free, no obligation.
Single automations are quoted on the call: a setup fee plus a monthly retainer to run them. Full systems start at $4,500, fixed scope, fixed price.
Book a free 30-minute call