Switching Dictation Apps on a Mac: The Setup Checklist

You have a dictation app on your Mac and you are about to replace it. Maybe the price went up, maybe you want audio to stay on the machine, maybe it falls apart when you mix Chinese and English. The move itself is the same set of steps whether you are switching from Wispr Flow, superwhisper or Typeless. This is the list, in the order that avoids redoing work. Feature-by-feature differences are on the compare page; this post is only about the mechanics.

Before you uninstall: save the word list and note the key

Two things in the old app are worth keeping, and both are easier to get while it is still installed.

The vocabulary. If your old app has a custom words or replacements list, get it out as plain text now: copy it from its settings window, or use whatever export it offers. Any readable list will do — you will retype these one at a time, so the format only has to be legible to you. ListenIn’s Import… button reads only ListenIn’s own backup file, not other apps’ files.

The trigger. Write down which key you press today and whether you tap it or hold it. Muscle memory is the part of a switch nobody plans for; you will either reproduce the old key or change it on purpose.

Then quit the old app fully, not just close its window. Two apps watching the same key both fire, and you cannot tell which one produced the text.

Pick a trigger key that isn’t already taken

ListenIn’s default trigger is fn: tap to start, tap again to finish, or hold and release. macOS may already have a job for that key. System Settings → Keyboard has a menu, “Press fn key to” (or “Press 🌐 key to” on some Macs), whose single-tap options include the emoji picker, input-source switching and Apple Dictation. If it is set to anything other than Do Nothing, every tap goes to macOS first.

Since v0.3.4 a fresh install checks that setting. If fn is bound to a system action, ListenIn starts on Right ⌘ instead and says so in onboarding: fn is taken by macOS — ListenIn uses Right ⌘ Command instead. You can leave it there. Right ⌘ is the one modifier macOS ships with nothing bound to it, which is why it is the fallback.

If you want fn, set the macOS menu to Do Nothing first; onboarding has a step for this, ③ Set the fn key to “Do Nothing”. To change the key later, open Settings › Dictation and use Trigger key. The list has fn and the four right-hand modifiers (Right ⌘ Command, Right ⌥ Option, Right ⌃ Control, Right ⇧ Shift), plus a Custom shortcut you record, such as ⌥Space or F13. ListenIn refuses a shortcut that overlaps one of its other keys and tells you so. The full map of who claims fn, including WeChat IME, is in who owns the fn key.

Grant two permissions

Onboarding asks for them in order:

① Microphone + Speech Recognition. Recognition runs through Apple’s speech engine with on-device processing forced. Audio is processed on the Mac and never reaches ListenIn’s cleanup service.
② Accessibility. This is what lets a background app see your trigger key and paste at your cursor. Without it, the key never reaches ListenIn, and any text you do produce arrives as a copy card instead of a paste.

If you skipped the second one, the menu bar shows ⚠️ Grant Accessibility access…, and the path is System Settings → Privacy & Security → Accessibility.

macOS 14–15: download the dictation language first

This catches people coming from apps with their own recognition model. ListenIn uses Apple’s engine and forces the on-device path on both of Apple’s engines: the newer one on macOS 26 and the classic one on macOS 14 and 15. On the classic engine the trade-off is blunt. If the on-device model for your language has not been downloaded, recognition fails outright. It does not quietly send your audio to Apple’s servers instead.

So before your first dictation on Sonoma or Sequoia, open System Settings › Keyboard › Dictation Languages and make sure the language you speak is downloaded. If you skip this, the symptom is an orb that ripples while you talk and no text at the end, and after repeated failures a HUD message that points at the same setting. Version-by-version steps are in Mac dictation shows nothing? Check the language download.

Rebuild your vocabulary: paste a list, then let corrections fill the rest

Open the dictionary: choose Dictionary from the menu bar, or Settings → Cleanup → Manage Personal Dictionary…. Both show the same list. Add the terms one at a time: type a word into New word and press Return. The field keeps focus, so a short list goes quickly. There is no bulk paste — Import… only reads a ListenIn backup. Keep it to names, products and terms you actually say; a short list works better than a long one, because these are repair candidates, not a spellchecker.

The list feeds the text cleanup pass that runs after recognition (and recognition biasing on the classic engine). When the recognizer writes “lama” and Ollama is in your dictionary, cleanup corrects it, because the sounds match and the context fits. It does not retrain Apple’s recognizer, and it will not invent a word the recognizer dropped.

The rest of your vocabulary rebuilds itself. After a paste, ListenIn watches the field for a short while. Fix a misrecognized word in place and, if the two sides sound alike, the corrected spelling goes into the dictionary with an on-screen Added to dictionary, and the pair is stored locally as a correction. If the same wrong word is still elsewhere in that text, a Replace All offer finishes the job. Fix a word once, and it sticks. The details, including what does and does not count as a correction, are in teaching Mac dictation new words.

When the list is in shape, back it up. In the same view, under Backup, Export… writes your words and corrections to one file and Import… reads it back. Import merges and never erases what is already there, which makes it the way to carry the dictionary to a second Mac.

Day-one tests

Run these before you trust the app with real work. Each one exercises a failure path, and you want to see it fail while nothing is at stake.

Test	What to do	What should happen
Password field	Click into a login password box, dictate	No paste. A card: Can’t paste into a password field · click to copy
Terminal	Dictate in iTerm2 or Terminal with Secure Keyboard Entry on	Either a normal paste, or Another app has secure input on · click to copy
Clipboard	Copy something, dictate, then press ⌘V	Your original clipboard content, not the dictation
Mixed 中英	Say 「我用 Claude Code 写了个 component」	The same sentence, both languages, no mode switch
Offline cleanup	Turn Wi-Fi off, dictate	The raw transcript still lands, labeled as raw

The clipboard test is the one most people skip. ListenIn snapshots every pasteboard type before it pastes and restores the snapshot afterwards, unless you copied something new in the meantime. The terminal case has its own post, Secure Keyboard Entry and dictation, because the fix is usually another app holding secure input, not the terminal itself.

Two more cards you may meet in the first week: You switched apps — click to copy if you change windows mid-dictation, and Couldn’t auto-insert · Click to copy when a paste cannot be verified. The rule behind all of them is the same. When insertion is not safe, you get the text on a card instead of a paste into the wrong place.

Nothing to migrate: there is no account

If your old app had a login, you may be looking for the equivalent here. There is none. The free plan is 8,000 cleaned words per week with no account and no card; the first cloud cleanup starts a 14-day full trial. Pro is a license key: paste it into License Key (Pro) and press Activate Pro. Current prices are on the pricing section.

That also means nothing syncs on its own. Everything ListenIn knows about you is local to the Mac: the dictionary, the corrections, Dictation History, your settings. On a second machine, run the same checklist and bring the dictionary over with Export and Import. What leaves the Mac in each mode, and what never does, is on the security page.

If you are weighing the move rather than making it, the FAQ covers the usual questions, system-wide dictation describes the daily loop, and the Typeless page is for people coming from there. If you are mid-switch, the order above is the whole job: save the list, pick the key, grant two permissions, check the language, add the words, run five tests.