Updated September 8, 2026 · By Sumbat.T
Dictation That Spells Your Jargon Right: 8 Tools Tested
The short answer
Wispr Flow, Superwhisper, Dragon Professional and Otter all let you add custom words, and each caps it differently: 60 characters an entry on Wispr Flow, 1,000 terms on Apple Voice Control, 100 names plus 100 terms on an Otter Pro seat. Two of the names people expect are not on the list. Apple Dictation has no custom vocabulary at all, because that feature belongs to Voice Control, and Windows Voice Typing on Win+H has none either, though Windows Voice Access does. BlabbyAI approaches it from the other end: custom spellings per language plus custom modes, which are free-form AI instructions that can enforce a spelling as part of whatever else the mode is doing, at $8.49 a month with a free tier of 60 credits weekly and no card. All prices and limits below were read off each vendor's own pages on 8 September 2026.
Every dictation tool on the market transcribes ordinary English well enough that the difference between them barely shows. Then you say the name of a colleague, a drug, an internal system or a client, and the gap opens instantly. This is where accuracy is actually won or lost, and it is the part of the category that published accuracy percentages tell you nothing about.
The confusing part is that vendors use one word, dictionary, for two mechanisms that behave very differently. Getting that distinction right matters more than picking the tool with the biggest number next to it, so this piece starts there, then goes through what each tool actually gives you and where the menu is.
A dictionary and a replacement rule are not the same thing
A recognition hint is handed to the speech model as part of transcription. It shifts the probabilities the model assigns to similar sounding candidates, so it can catch forms you never wrote down, including ones you would not have predicted.
A replacement rule runs afterwards. The audio is already text, and the rule swaps one string for another. It is exact, it is predictable, and it only ever fires on the specific wrong output you thought to list in advance.
Superwhisper is the clearest illustration because it ships both, named separately: Vocabulary Words are hints, Replacements are post-processing. Voice In sits entirely on the replacement side. This matters in practice because a name you say fifty times a week can come out wrong in four or five different ways, and a replacement rule catches only the ones already on your list.
The test that tells them apart
Dictate the troublesome term ten times in normal sentences, then look at the failures. If the tool produces several different wrong spellings, you need recognition hints and a replacement list will always be one step behind. If it fails the same way every time, a replacement rule will fix it completely.
The eight tools, compared
Every row was checked against the vendor's own documentation or pricing page on 8 September 2026, not against other comparison articles. Where a vendor publishes no figure, the table says so rather than guessing.
| Tool | What it is called | Where it lives | Limit | Price |
|---|---|---|---|---|
 BlabbyAI | Custom spelling, plus custom modes | Settings › Languages › pick a language › Add Custom Spelling | No documented per-entry cap | $8.49/mo unlimited, free tier 60 credits weekly |
 Wispr Flow | Dictionary | App › Dictionary › Add new | 60 characters per entry; bulk import 1,000 items / 3 MB | Pro $15/mo, $12/mo annual |
 Superwhisper | Vocabulary Words and Replacements | Settings › Vocabulary | None published; docs advise a short list | Pro $8.49/mo, $84.99/yr, $249.99 lifetime |
 Dragon Professional | Vocabulary Editor | DragonBar › Vocabulary › Open Vocabulary Editor | None published | Not listed on the vendor site |
 Otter.ai | Custom Vocabulary | Account settings › General › Manage Vocabulary | Pro 100 names + 100 terms; Business 800 + 800 team | Pro $16.99/mo, $8.33/mo annual |
Voice In | Custom Voice Commands (replacement only) | Options › Custom Commands › Add Command | None published | Plus $9.99/mo, $59.99/yr |
 Apple Voice Control | Custom vocabulary (not in Dictation) | Accessibility › Voice Control › Vocabulary › Add | 1,000 terms | Free with macOS |
 Windows Voice Access | Add to vocabulary | Voice access settings › Add to vocabulary | None published | Free, Windows 11 22H2+ |
Sources: each vendor's own documentation and pricing pages, checked 8 September 2026. Full links at the end of this article.
Spelling that follows the language, not a word cap
Teach it once, in the language you said it in
BlabbyAI custom spellings attach to a language, so the same term can behave differently across 90+ languages with auto-detect. Custom modes can enforce spellings too, as part of the grammar or formatting pass they were already doing. $8.49/mo unlimited, 60 credits a week free, no card.
BlabbyAI: custom spellings, plus an instruction layer that can enforce them
BlabbyAI handles the problem in two places, which is unusual. The direct route is custom spelling: Settings, then Languages, then pick a language, then Add Custom Spelling. You type the term exactly as you want it written and it is matched when you transcribe in that language. Because spellings hang off a language rather than a single global list, a term that should look one way in English and another in Spanish is handled without collision.
The second route is custom modes, and it is the reason the ceiling here is higher than a word list. A mode is a free-form instruction applied to your speech after transcription, using a model you pick. Because the instruction is just text, spelling rules sit inside it alongside everything else the mode does, so one mode can clean up grammar, format the result as an email and enforce that a product name is always written a particular way, in a single pass on one shortcut.
Source: BlabbyAI desktop app. The AI Instructions field is free-form, so spelling rules go in alongside whatever else the mode does.
Best for
Anyone who dictates in more than one language, or who wants spelling rules to run alongside grammar cleanup and formatting in the same pass.
Watch out for
Modes process one recording at a time after you stop speaking, so this is not live word-by-word correction as you talk.
Price
$8.49/mo unlimited. Free tier of 60 credits weekly, around 2,000 words, no card required.
Wispr Flow: a Dictionary with a 60 character ceiling
Wispr Flow calls it the Dictionary and puts it in the left sidebar of the desktop app, where Add new takes a word and saves it. It also supports misspelling correction rules, so it covers both mechanisms. Each entry is capped at 60 characters, duplicates are rejected, and bulk import accepts up to 1,000 items or 3 MB per import.
Wispr Flow’s own two pages disagree
The pricing page presents adding words to the dictionary as a Pro differentiator, with the free tier described only as learning your names and jargon as you go. The dictionary help article says the personal dictionary is available on all tiers and that only team sharing needs a Team or Business plan. Separately, the bulk import documentation requires an active paid or trial subscription. If the dictionary is the reason you are choosing this tool, confirm it on the tier you intend to buy.
Best for
People who already have a written list of terms and want to load it in one go, since bulk import is the most generous here.
Watch out for
Bulk import is Mac and Windows only and needs an active subscription. Android drops starring, sorting and bulk import, and its replacement rules are read-only.
Price
Free tier available. Pro $15/mo, or $12/mo billed annually.
Superwhisper: the only one that names both mechanisms separately
Settings, then the Vocabulary tab, and you get Vocabulary Words and Replacements as distinct features. That separation is the most honest treatment of this problem in the category, because it makes the user choose the right mechanism rather than hiding both behind one word.
It is also the only vendor that publishes the downside. Its documentation warns that adding too many words can confuse the transcription model and advises keeping the list short, which contradicts the instinct to paste in an entire glossary. The documentation goes further on what a bloated list costs you: as well as recognition, it can degrade punctuation, language detection and formatting. That is a wider blast radius than most people expect from a word list.
The commercial detail to note is that custom vocabulary sits behind Pro. Superwhisper's own feature comparison marks it with a dash in the free column, alongside local voice models, custom modes and speaker separation.
Best for
Anyone who wants to understand and control which mechanism is fixing their terms, rather than trusting one combined dictionary.
Watch out for
Custom vocabulary is Pro-only, and the vendor warns that long lists can degrade punctuation, language detection and formatting as well as recognition.
Price
Free tier available. Pro $8.49/mo, $84.99/yr, or $249.99 lifetime.
Dragon Professional: the deepest vocabulary tooling, and no published price
Dragon has been doing this longest and it shows in the design. DragonBar, then Vocabulary, then Open Vocabulary Editor gives you entries with a written form and a separate spoken form, which is the correct model for anything whose pronunciation and spelling diverge: an acronym you say as letters, a name that is not spelled the way it sounds. Custom entries are flagged in the list, and Dragon can also learn from documents you point it at.
Dragon is also blunt about the consequence of not doing this work: its documentation states that it cannot transcribe words that are not in its vocabulary. The catch for anyone shopping today is commercial rather than technical. As of 8 September 2026 neither the product page nor the store URL lists a price, so any figure you see quoted in a roundup is not coming from the vendor.
Best for
Specialist vocabularies where spoken and written forms genuinely differ, and where the vocabulary work is worth doing thoroughly once.
Watch out for
No price is published on the vendor site as of 8 September 2026, and it is Windows only.
Price
Not listed on the vendor site. Treat quoted figures elsewhere as unverified.
Otter.ai: vocabulary priced by the entry
Otter splits Custom Vocabulary into names of people and other terms, reached through your account settings, then General, then Manage Vocabulary. It is the only tool here that turns the size of your vocabulary into a pricing tier: a Pro seat carries 100 names plus 100 other terms, while Business raises that to 800 plus 800 at team level with 200 plus 200 per user.
Two caveats worth knowing. Otter is built around meetings and recordings rather than typing into whatever field your cursor is in, so it is solving a different problem from most of this list. And its own sources disagree on the free tier, with the pricing page listing no custom vocabulary for Basic while the help centre has described a small allowance.
Best for
Teams transcribing meetings who want one shared vocabulary of client and staff names across everybody’s recordings.
Watch out for
Caps are strict and tier-linked, and this is meeting transcription rather than dictation into your own text fields.
Price
Basic free. Pro $16.99/mo or $8.33/mo annual. Business $30/mo or $19.99/mo annual.
The built-in tools: two widespread myths
Both operating systems are routinely described as having features they do not have, because in each case the vendor ships two voice products and only one of them holds the dictionary.
Apple Dictation does not have custom vocabulary
The feature lives in Voice Control, an Accessibility feature, not in Dictation. The path is System Settings, Accessibility, Voice Control, Vocabulary, Add, and the cap is 1,000 terms. Dictation is involved only because it must be enabled to record pronunciations, and the Vocabulary option is dimmed in languages that do not support it. An article telling you Apple Dictation learns your custom words is describing a different product surface than the one you dictate with.
Windows Voice Typing does not have a dictionary either
Voice Access, on Windows 11 version 22H2 and later, has Add to vocabulary in its settings and supports spoken commands including "add to vocabulary" and "spell that". Voice Typing, the Win+H tool most people mean by Windows dictation, has no custom vocabulary feature at all: Microsoft documents its settings, its languages and its voice commands with no mention of adding words.
Best for
Anyone who wants a zero-cost option and is willing to use the specific product that actually holds the vocabulary feature.
Watch out for
Both are easy to get wrong: the feature is in Voice Control on macOS and Voice Access on Windows, not in the dictation tool sitting next to it.
Price
Free, built into macOS and Windows 11 22H2 and later.
Voice In: replacement rules wearing a dictionary’s coat
Voice In offers Custom Voice Commands, at Options, then Custom Commands, then Add Command, and they map a spoken phrase to output text. Useful, but it is post-transcription substitution rather than anything that changes what the model hears, so it belongs on the replacement side of the distinction at the top of this article. It is a Chrome extension only, and the feature requires the paid Voice In Plus tier.
Best for
Browser-only workflows where the same handful of phrases need expanding or correcting in a predictable way.
Watch out for
It cannot improve recognition, only rewrite output, so unpredictable misspellings of a name will slip through.
Price
Voice In Plus $9.99/mo or $59.99/yr.
How to choose, in four questions
Do your terms fail the same way every time, or differently each time? Consistent failures are fixed completely by replacement rules. Inconsistent ones need recognition hints, which rules out the replacement-only tools.
How many terms do you actually need? Be honest, and count the ones you say weekly rather than everything in your field. Twenty to thirty terms fits inside every cap on this page, which makes the published limits mostly irrelevant and the other three questions decisive.
Do you dictate in more than one language? Check whether the tool keeps one shared list or one per language. Apple maintains separate lists per language and hides the feature entirely in unsupported ones; BlabbyAI attaches spellings to a language by design.
Where does the text need to land? A tool that types into any focused field covers your EHR, your CRM, Slack, your IDE and your email with the same vocabulary. A meeting transcription product solves a different problem, however good its dictionary is.
What a word list cannot do
A vocabulary entry fixes one term. It does not fix the surrounding text, and the corrections most people make after dictating are not spellings at all: they are removing the false starts, breaking one long spoken run-on into paragraphs, and turning speech into the register the message actually needs.
This is where an instruction layer does work a dictionary cannot, because the instruction is ordinary language rather than a row in a table. A BlabbyAI mode can carry the spelling rules for your product names, strip filler words, and format the result as a support reply, all in the same pass. It can also handle the case a word list has no answer for, where a term should be written one way in a clinical note and another way in a message to a patient, because that is a sentence you can simply write into two different modes.
A limit worth knowing before you switch
Modes process one recording at a time, after you stop speaking, and they have no memory between recordings. This is not live correction as you talk. In exchange, what the instruction can ask for is open-ended rather than a fixed feature list.
Frequently asked questions
Which dictation software lets you add custom words?
Most of the serious ones, but they do not all mean the same thing by it. Wispr Flow has a Dictionary with a 60 character limit per entry. Superwhisper has a Vocabulary tab that is Pro-only and splits into recognition hints and post-transcription replacements. Dragon Professional has a Vocabulary Editor with separate written and spoken forms. Otter has Custom Vocabulary capped by plan, 100 names and 100 other terms on Pro. Apple offers custom vocabulary through Voice Control rather than Dictation, capped at 1,000 terms. Windows Voice Access has Add to vocabulary, while Windows Voice Typing on Win+H has no dictionary at all. BlabbyAI takes a different route: custom spellings per language plus custom modes, which are free-form AI instructions that can enforce spelling as part of a wider cleanup, at $8.49 a month with no per-entry cap.
What is the difference between a custom dictionary and a replacement rule?
This is the distinction that decides whether a feature will actually work for you, and almost no roundup makes it. A recognition hint is fed to the speech model before or during transcription, so it biases what the model hears and can catch variants it has never seen written down. A replacement rule runs after transcription and swaps one string for another, so it only fires when the exact wrong text appears. Superwhisper is the clearest illustration because it ships both separately: Vocabulary Words are hints, Replacements are post-processing. Voice In sits entirely on the replacement side, since its Custom Voice Commands map a phrase to another phrase after the fact. If your problem is that a drug name comes out as three different wrong spellings, replacement rules will only catch the ones you predicted.
Does Apple Dictation support custom vocabulary?
No, and this is the single most common error in articles on this topic. The custom vocabulary feature on macOS belongs to Voice Control, which is an Accessibility feature, not to Dictation. Apple’s own documentation puts it under System Settings, Accessibility, Voice Control, Vocabulary, and the cap is 1,000 terms. Dictation only enters the picture because it has to be enabled in order to record pronunciations. So a roundup that tells you Apple Dictation lets you add custom words is describing a different product surface than the one you would actually be dictating with.
Can Windows voice typing learn new words?
Not the one most people use. Windows has two separate voice products and only one has a dictionary. Voice Access, on Windows 11 version 22H2 and later, has an Add to vocabulary setting and voice commands including "add to vocabulary" and "spell that". Voice Typing, the Win+H tool that most people mean when they say Windows dictation, has no custom vocabulary feature at all: Microsoft’s own voice typing page documents settings, languages and voice commands with no mention of adding words. These two get conflated in almost every comparison article.
Does adding more custom words make dictation more accurate?
Not automatically, and one vendor is refreshingly honest about it. Superwhisper’s own documentation warns that "adding too many words can confuse the AI transcription model" and advises keeping the list short. That is worth taking seriously, because the intuition most people have is the opposite: that a longer list is a better list. A vocabulary hint works by shifting the probabilities the model assigns to similar sounding options, so a list stuffed with rare terms can pull ordinary words toward strange ones. The practical approach is a short list of the terms you actually say often and that actually come out wrong, rather than a pre-emptive dump of your entire field’s glossary.
How do I get a dictation app to spell someone’s name correctly?
Three routes, in rough order of reliability. First, add it to whatever custom vocabulary or spelling list the tool provides, which teaches the transcription layer the exact form you want. In BlabbyAI that is Settings, Languages, pick a language, Add Custom Spelling, and you type the spelling exactly as you want it written. Second, if your tool has an AI instruction layer, write the rule there: a BlabbyAI custom mode can carry an instruction such as always spelling a colleague’s name a particular way, alongside whatever else that mode does. Third, a plain replacement rule as a fallback, though it only catches the specific misspellings you thought to list.
Is custom vocabulary a paid feature?
Usually yes, and the gate sits in different places. Superwhisper documents custom vocabulary as Pro-only, at $8.49 a month or $84.99 a year. Voice In requires Voice In Plus at $9.99 a month or $59.99 a year for Custom Voice Commands. Otter scales the caps by plan, with no custom vocabulary listed for Basic on its pricing page. Wispr Flow is the awkward one: its pricing page presents adding words to the dictionary as a Pro differentiator while its help documentation describes the personal dictionary as available on all tiers, so the two vendor sources disagree. Apple Voice Control and Windows Voice Access are both free, being built into their operating systems. BlabbyAI includes custom spellings and custom modes on the $8.49 a month unlimited plan and gives 60 credits a week free without a card.
How many custom words can I add?
The published limits vary enormously and several tools publish none at all. Apple Voice Control caps at 1,000 terms. Otter caps by plan, 100 names plus 100 other terms for a Pro user, 800 plus 800 at team level on Business. Wispr Flow limits each entry to 60 characters and bulk imports to 1,000 items or 3 MB per import. Dragon, Superwhisper and Windows Voice Access publish no numeric limit, though Superwhisper explicitly advises against long lists. BlabbyAI does not impose a documented per-entry cap on custom spellings. For most people the limits are academic, because the list that actually helps is the twenty or thirty terms you say every day.
What if my dictation tool has no custom vocabulary feature?
You have two workable options and one that is not worth the effort. The good options are to switch to a tool that has one, or to use a tool with a free-form AI instruction layer that can enforce spellings as part of its normal processing. BlabbyAI custom modes work this second way: because a mode is an ordinary instruction to an AI model, an instruction such as always writing a particular product name in a particular form applies every time that mode runs, and it can sit alongside the grammar cleanup or formatting the mode was already doing. The option not worth the effort is retyping the same corrections by hand, which is the cost you are already paying and the reason you are reading this.
Does custom vocabulary work in other languages?
Sometimes, and this is a real limitation to check before you commit. Apple Voice Control only offers the Vocabulary option in languages that support the feature, and the menu item is dimmed otherwise, with separate lists maintained per language. Dragon ties some of its learning features to the profile language. BlabbyAI attaches custom spellings to a specific language, which is by design rather than a restriction: you pick a language then add the spelling, so the same term can be handled differently across the 90 plus languages it supports with auto-detect. If you dictate in more than one language, check whether your tool keeps one shared list or one per language, because the answer changes how much setup work you are signing up for.
Stop retyping the same six words
Custom spellings per language, plus modes that enforce them while they clean up grammar and shape the result. Works in any text field on Windows or in Chrome. $8.49/mo unlimited, 60 credits a week free, no card.
Sources
- Wispr Flow, Teach Flow your words with the Dictionary and pricing, checked 8 September 2026.
- Superwhisper, Vocabulary documentation, checked 8 September 2026.
- Nuance, Adding words to your vocabulary, checked 8 September 2026.
- Otter.ai, pricing, checked 8 September 2026.
- Dictanote, Custom voice commands, checked 8 September 2026.
- Apple, Create a custom vocabulary for Voice Control, checked 8 September 2026.
- Microsoft, Dictate text with voice access and Use voice typing, checked 8 September 2026.
Related reading: the nine AI dictation apps compared and priced, dictation that formats as you speak, and what dictation software does.