Short answer: Built-in Mac and iPhone dictation has no dedicated spelling mode. You can say letters one at a time, slowly and evenly, and they usually appear, but single letters sound like common words, so results are unreliable. For recurring names and terms, a personal dictionary fixes spelling permanently.
Every dictation user eventually hits the word that will not come out right. A surname with an unusual spelling, a company name that is almost a real word, a medication, a street in another language. The instinct is universal: you slow down and start spelling it out loud, letter by letter, the way you would to a person on the phone.
And then dictation types "are ee why ay in" instead of R-E-Y-A-N, and you learn the hard way that spelling out loud is one of the things speech recognition is structurally worst at. This guide explains why, shows the technique that works about as well as anything can, lists the workarounds that do not work at all, and finishes with the approach that removes the need to spell in the first place.
Why spelling out loud breaks dictation
Dictation engines transcribe by weighing what they heard against what people usually say. That context-weighting is why full sentences transcribe so much better than isolated words: every word constrains its neighbors, and the engine uses those constraints to resolve ambiguous sounds.
Spoken letters destroy that mechanism twice over.
First, most letter names are homophones of common words. "B" is "be" and "bee." "R" is "are." "U" is "you." "Y" is "why." "I" is both the letter and the most common pronoun in the language. "C" is "see" and "sea." When you say a letter in isolation, the engine has to choose between a one-character token that almost nobody dictates and an everyday word that everybody dictates. It picks the word, because on the numbers it is usually right.
Second, a spelled sequence has no context to lean on. The letters of a surname are, from the engine's point of view, random. Nothing about "R-E-Y" predicts that "A" comes next. So the one tool the engine relies on to rescue ambiguous audio is exactly the tool a spelled word takes away. This is the same reason serial numbers and license plates dictate so badly, a problem we cover from the numeric side in how to dictate numbers and dates.
Understanding this mechanism matters because it tells you what has a chance of working (anything that makes letters less ambiguous) and what is hopeless (saying them faster, louder, or more crisply, which changes nothing about the ambiguity).
The letter-by-letter technique that half works
Sometimes you genuinely need to spell something once, into a form, with no time to set anything up. Here is how to give letter-by-letter dictation its best shot:
- Dictate the sentence first, then fix the word. Do not try to spell mid-sentence. Speak the full sentence with your best pronunciation of the difficult word, let it land wrong, and then repair just that word. Mid-sentence mode-switching between prose and spelling confuses the transcription and you.
- Select the wrong word before re-entering it. On iPhone, tap the misspelled word (or tap-and-hold to select it); on Mac, double-click it. With the word selected, whatever you produce next replaces it cleanly.
- Say the letters in a slow, even rhythm. About one letter per second, identical spacing, no trailing off. An even cadence is the strongest signal you can send that these are letters, not words.
- Say "capital" before a letter that needs it. "Capital R" at the start of a name works reasonably well on both platforms. The full set of capitalization commands lives in our guide to dictating capital letters.
- Expect casualties and fix them by keyboard. Even done perfectly, expect the odd "are" or "you" in the output. Delete those two characters and type the letter. For a six-letter name, fixing two letters by hand still beats retrying the whole word four times.
Honest assessment: for a one-off word, this gets you there in about the same time as just typing the word. Which leads to the actual rule.
The real rule: speak the sentences, type the strings
A word you have to spell is not really a word. It is a string: an exact character sequence that carries information your voice cannot transmit. We keep arriving at this same boundary across this series, whether the string is an email address or URL, a promo code, or a part number, and the conclusion is always the same. Dictation is the fastest way ever invented to produce sentences, and a clumsy way to produce arbitrary strings.
So split the work. Speak the prose around the difficult word at full conversational speed, 130 to 150 words per minute, and when you reach the string, use the keyboard for eight keystrokes. The hybrid is faster than either pure mode, and it eliminates the most dangerous property of spelled-word errors: they are silent. A misrecognized sentence usually reads as obvious nonsense. A misspelled name reads as fine, and ships.
What does not work (so you can stop trying it)
- The phonetic alphabet. Saying "Romeo Echo Yankee" gets you the words "Romeo Echo Yankee." Built-in dictation has no mode that interprets NATO-style call signs as letters. It works on humans, not transcription engines.
- "As in" constructions. "R as in Robert" types exactly that, seven words of it, into your text.
- Saying the word more slowly and loudly. Over-articulating a word the engine does not know cannot teach it the word. It usually makes recognition worse, because stretched, stressed speech matches the engine's expectations less than your normal voice does. If you find yourself doing this repeatedly, that strain is a real cost; we wrote about it in our piece on voice strain from dictation.
- Hoping it learns from your corrections. Built-in dictation does not durably learn your corrections. Fix "Reyan" today and you will fix it again tomorrow, and every day after. This is the design ceiling of stock dictation, not a settings problem.
Fix the spelling at the source instead
The pattern behind almost every spelling emergency is that it is not a one-off. It is your client's name, your product's name, your street, your industry's jargon. You will need that exact string hundreds of times. Every at-the-source fix below converts a recurring spelling problem into a one-time setup step.
1. Add people to Contacts
On both iPhone and Mac, dictation reads your Contacts and strongly prefers those spellings when it hears a matching name. Adding a contact card for anyone whose name you dictate regularly, with the exact spelling in the name fields, is the highest-leverage free fix for people's names. It does nothing for non-name words, though, and it will not help with the surname of someone you are writing about but would never store as a contact. We cover the full name-handling toolkit in how to stop Mac dictation autocorrecting names.
2. Text Replacement for recurring strings
Settings → General → Keyboard → Text Replacement (iPhone) or System Settings → Keyboard (Mac) lets you map a short trigger to an exact string. Dictate the trigger word, get the string. It works, with known sharp edges: replacements fire on the whole trigger only, some browser and Electron text fields ignore them, and stale entries become saboteurs that rewrite text you did not want rewritten.
3. A personal dictionary in your dictation tool
The strongest fix is a dictation tool whose vocabulary you can extend directly. Voice Keyboard Pro's Smart Vocabulary is a personal dictionary with replacement rules: add "Reyan," "Xylotech," or "hydroxychloroquine" once, and the transcription engine produces the exact spelling every time you say the word, in any app, with no trigger phrases and no spelling out loud. Replacement rules go further for strings that are not pronounceable at all: define a spoken phrase like "my case number" and the rule inserts the exact character sequence you stored.
This is the difference in kind, not degree. Contacts and Text Replacement patch specific holes; a personal dictionary changes what the engine believes a word is, which is where spelling errors actually come from. On Mac, hold the hotkey, speak naturally, release, and the corrected spelling is simply in the text that appears at your cursor.
4. On iPhone, fix by voice instead of by cursor
The worst part of a misspelled word on iPhone is not the respeaking, it is the cursor: fingertip-hunting for a position between two letters on glass. Voice Keyboard Pro's keyboard includes Voice Edit, which lets you speak the change instead: say what should change and to what, and the text updates without touching the cursor at all. For the broader repair toolkit on both platforms, see our guide to correcting dictation mistakes.
A workflow for spelling-heavy work
If your writing is full of proper nouns (case files, patient notes, sales pipelines, academic citations), do not fix words as they break. Batch the problem:
- Harvest for one week. Every time you correct a spelled word, add it to a running note. Do not fix anything permanently yet, just collect.
- Load the dictionary once. At the end of the week, sit down for ten minutes and add the whole list: names to Contacts, terms and codes to Smart Vocabulary or Text Replacement.
- Watch the corrections drop. Most people find a few dozen entries cover the overwhelming majority of their recurring misspellings, because your vocabulary of difficult words is much smaller than it feels while you are fighting them one at a time.
After that, spelling out loud becomes what it should be: a rare fallback for genuinely novel strings, not a daily tax.
What never to spell into a microphone
A boundary that matters more for spelled strings than for anything else you dictate: do not spell credentials out loud. Passwords, two-factor codes, card numbers, account numbers, API keys. Partly because voice is the most overhearable input method ever invented, and partly because a silent one-character error in a credential fails in ways you will not see until it matters. Type them, always, and preferably from a password manager rather than your memory.
For everything else you dictate, it is fair to ask where the audio goes. Voice Keyboard Pro's server stores only operational pings: no audio and no transcript content. What you speak lands in your document and stays yours.
FAQ
Is there a spelling mode in iPhone or Mac dictation?
No. Neither platform's built-in keyboard dictation has a dedicated mode that interprets speech as spelling. Letter-by-letter entry rides on the ordinary transcription path, which is why it inherits all the homophone problems described above. (macOS Voice Control, the separate accessibility feature, does have richer letter-entry commands, but it is a different tool with its own trade-offs and conflicts with normal dictation.)
Why does dictation get the word wrong even after I fixed it yesterday?
Because built-in dictation does not durably learn from your corrections. Its vocabulary is shared and fixed; your corrections edit the document, not the engine. Only vocabulary-level fixes (Contacts, replacement rules, a personal dictionary) change tomorrow's output.
How do I dictate a name with punctuation in it, like O'Brien or Anne-Marie?
Common punctuated names are usually in the engine's vocabulary and come out correctly when spoken naturally. Rare ones misfire, and spelling will not save you because the apostrophe and hyphen are not letters you can say mid-sequence reliably. These are precisely the names to put in Contacts or your personal dictionary once.
Does spelling work better in a quiet room?
Somewhat. Single letters are the shortest possible utterances, and short utterances are the first casualty of background noise, especially other people's speech. Quiet helps the technique; it does not repair the underlying ambiguity.
The bottom line
Spelling out loud is the one phone-call habit that does not survive contact with speech recognition, because single letters are acoustically identical to the most common words in the language. Use the slow-even-rhythm technique when you are cornered, type the string when it is quicker, and never spell credentials at all. But mostly, stop treating recurring words as spelling emergencies. Put them in a personal dictionary once and let the engine spell them for you forever. Voice Keyboard Pro's free tier is enough to load your ten worst words and feel the difference today; Pro is $4.99 a month or $34.99 a year if voice becomes how you write.