The Custom Dictionary: Teach Dictation Your Words

Every dictation tool eventually trips over the same words: your colleague's name, your product, the acronym your team lives by. A custom dictionary fixes that once and for all. Here is what it is, why it works, and how to build a short list that makes Mac voice to text spell your world correctly.

What a custom dictionary actually does

Speech recognition models like Whisper and Parakeet are trained on huge amounts of general language. That makes them excellent at everyday words and surprisingly weak at the words that are unique to you. Ask them to transcribe "Siobhan reviewed the Kubernetes rollout for Vonsel" and a general model may produce "Shavon reviewed the coober netties rollout for vonsell." The audio was clear. The model simply had no reason to prefer your spellings.

A custom dictionary closes that gap. You give the app a short list of the exact spellings you use, and it biases the transcription and cleanup toward them. Instead of guessing phonetically, the app recognizes that the sound it just heard maps to a word you have told it about. It is the same idea as adding a contact to your phone so autocorrect stops fighting you. If you want the background on how the underlying open model handles vocabulary, OpenAI documents its Whisper speech model on GitHub.

What belongs in your dictionary

The words worth adding share one trait: they are common in your life but rare in the training data. Good candidates fall into a few buckets:

• People and place names. Colleagues, clients, cities and streets that are spelled in unexpected ways.
• Brand and product names. Your company, your tools, competitors you mention constantly.
• Acronyms and initialisms. Team shorthand like SOC 2, RAG, KPI or a project codename.
• Industry jargon. Medical, legal, engineering or scientific terms that sound nothing like they are spelled.

Resist the urge to dump an entire glossary in on day one. A short, high-signal list keeps the app fast and accurate. Add a word the first time you catch it wrong, and your dictionary grows naturally around your real vocabulary. This same discipline shows up in our look at how accurate voice to text is in 2026: the biggest accuracy gains come from tuning for your words, not chasing a bigger model.

Dictionary or AI prompt: which fixes what

People often confuse two features that solve different problems. A custom dictionary controls spelling of specific terms. A custom AI prompt controls style: tone, formatting, how much filler gets removed, whether output is a bulleted list or a tidy paragraph. You want both, working together.

If you want to go deep on the second column, our power guide to custom AI prompts for dictation covers how to write instructions the on-device model actually follows. Pair a tight dictionary with a good prompt and the raw transcript arrives almost ready to send.

The trade-offs, honestly

A custom dictionary is not magic, and it helps to know where it shines and where it does not.

Used with intent, the upside far outweighs the limits. And because dictation lets you work at speaking speed, and most people speak around three to four times faster than they type, every correction you no longer make by hand compounds across your day.

Common myths, cleared up

Building yours in five minutes

You do not need a plan or a spreadsheet. Open your dictation history from the last week and scan for words that came out wrong. Add the ones you will say again, skip the one-offs. Because BlaBlaType maps a single shortcut to your dictation, you can test a new word the moment you add it. For more on that workflow, see how we think about one shortcut and three modes for voice input, and if you are weighing built-in options, our Apple Dictation comparison shows why a tunable dictionary matters. Within a couple of sessions your most-used terms just work, and you stop noticing the feature at all, which is exactly the point.