Custom vocabulary for dictation: teach it your words

Every dictation app has a short list of words it always gets wrong for you: a colleague's name, your product, the acronym you use twenty times a day. The fix is a custom vocabulary. Add each word once, spelled exactly how you want it, and the app stops guessing. For phrases you want swapped for something else entirely, use replacements.

Here's how both work, what the built-in Mac options can and can't do, and how to set up a list that actually saves you time.

#The quick version

1. Open your dictation app's vocabulary or dictionary list. In Sayburst it's Settings → Dictionary.
2. Add the names, products and terms you correct most often, spelled and capitalized the way you want them.
3. Add replacements for phrases you want expanded, like "my email" → your address.
4. Dictate a sentence that uses them and check the result.

Start with the ten words you fixed most last week, not a 500-word glossary. A short list you trust beats a long one you never check.

#Vocabulary vs replacements

They solve different problems, and mixing them up is the most common setup mistake.

#What a vocabulary list fixes

A custom vocabulary handles three kinds of mistakes.

• Spelling. Names like "Siobhan" or "Nguyen" that a model has no way to guess from sound alone.
• Capitals. "github" becomes "GitHub", "iphone" becomes "iPhone", "gad" becomes "GAD".
• Near-misses. The model hears something close but wrong. Parakeet once heard our own name as "Saberst". With "Sayburst" in the vocabulary, it comes out right.

Acronyms, and why "GAD" needs help

Acronyms are where dictation trips most, because the same three letters sound like an ordinary word. In a therapist's notes, GAD means generalized anxiety disorder; said aloud, it sounds exactly like the word "gad". A dictation model has no way to know which one you meant.

Two ways to handle it:

• Keep the acronym. Add "GAD" in capitals to your vocabulary. Sayburst fixes the capitals whenever it hears it.
• Expand it. Add a replacement with a trigger phrase you'd never say otherwise, like "gad full" → "generalized anxiety disorder (GAD)". Don't make plain "gad" the trigger if the word could ever mean something else in your writing.

The same goes for any acronym with a second life as a word: "SAP", "CRUD", "PEP", "RICE".

#Custom vocabulary on a Mac: what's built in

Apple Dictation doesn't have a custom word list in Apple's guide. What it does give you: words it wasn't sure about are underlined in blue, and you can click one to pick an alternative. That fixes the moment, not the next time.

Voice Control has a real vocabulary list. Open System Settings → Accessibility → Voice Control → Vocabulary and add up to 1,000 terms (Apple's Voice Control vocabulary guide). If the option is dimmed, your language doesn't support it. The catch: while Voice Control is on, standard Dictation isn't available, and you dictate through Voice Control instead, and it listens for commands as well as text. That's a bigger change than most people want just to fix a few names. Our Voice Control vs Dictation guide explains the difference.

Third-party apps generally have one: Wispr Flow calls it the Dictionary (60 characters per entry on desktop, per Wispr's help center), and Superwhisper calls it Vocabulary. Both checked October 31, 2026.

#How vocabulary works in Sayburst

Add words in Settings → Dictionary → Vocabulary. What happens next depends on the speech model you use:

• Whisper models are nudged toward your words while they transcribe.
• Parakeet models (the default) transcribe first, then Sayburst corrects near-misses to your words afterwards. It never changes a real word into one of your terms, so adding "Dana" won't turn every "data" into "Dana".
• Capitals for your words are fixed either way.
• Writing providers, including the optional on-device Sayburst Writer, are told to spell your words exactly as listed.

Here's a sprint update with three words the model hasn't met:

"um so Siobhan merged the, the Kubernetes fix, and the Postgres migration goes out Thursday after Nguyen signs off"

Names like Siobhan and Nguyen are exactly what a model can't spell from sound alone. With all four in the vocabulary, what's typed is:

So Siobhan merged the Kubernetes fix, and the Postgres migration goes out Thursday after Nguyen signs off.

It learns from your corrections

Turn on Suggest words when I correct a dictation. After you dictate, Sayburst looks back at that text field a few times over the next minute, on your Mac. If you fixed a name by hand, it asks whether to add it. It never reads password fields, and it only keeps the words you accept (plus the ones you declined, so it doesn't ask twice).

#Replacements in Sayburst

A replacement turns an exact phrase into whatever you want, matched as whole words and applied after cleanup.

So "send the invoice to my email and link sayburst dot com in the footer" becomes:

Send the invoice to alex@example.com and link sayburst.com in the footer.

Because matching is whole-word, a rule for "eta" won't touch "beta". Each rule has a case-sensitive toggle for the times capitals matter. The Try it box at the bottom of the pane lets you type a sample sentence and see the result before you dictate anything.

#Import, export and moving Macs

Already keep a glossary in a spreadsheet? Import it as a two-column CSV: find, replace. Rows with an empty replace column become vocabulary words. JSON export keeps everything, which is the easy way to move your list to another Mac or share it with a teammate.

#Tips for a list that pays off

• Add words when they go wrong, not in advance. The learning prompt does most of this for you.
• Use the exact spelling and capitals you want. "macOS", not "MacOS".
• Keep replacement triggers unusual. "my email" is safe; "email" alone would fire constantly.
• Re-test after switching models. Parakeet and Whisper make different mistakes, so a word you never needed on one may need adding on the other. Our Parakeet vs Whisper benchmark shows how they compare.
• Check punctuation separately. Vocabulary fixes words, not sentence structure. For that, see Mac dictation punctuation commands.

#Limits

A vocabulary list can't make a model hear a word you mumble or say over loud background noise; it helps the model pick the right spelling of what it did hear. Near-miss correction works best when your word sounds reasonably like what the model guessed. And replacements are literal: they do exactly what you told them, every time, so review them when your writing changes.

Questions people ask

What is custom vocabulary in dictation?

It's a list of words you add so the dictation app spells them your way: colleagues' names, product names, acronyms and jargon. Depending on the app, the list steers recognition, fixes near-misses afterwards, or both.

Can I add custom words to Apple Dictation on Mac?

Apple's Dictation guide doesn't describe a custom word list. Voice Control has a Vocabulary list of up to 1,000 terms, but it only applies while you dictate through Voice Control, which replaces standard Dictation while it's on.

How do I get dictation to type acronyms like GAD in capitals?

Add the acronym in capitals to your app's vocabulary. In Sayburst, add GAD in Settings, Dictionary, and it fixes the capitals. If you want the expanded form instead, add a replacement with a trigger phrase only you would say.

What's the difference between vocabulary and replacements?

Vocabulary teaches the app how a word is spelled so it recognizes it. A replacement swaps one exact phrase for another after transcription, like turning my email into your address.

Does custom vocabulary work offline?

In Sayburst, yes. The vocabulary, replacements and speech recognition all run on your Mac, so they keep working with no connection.