← Back to Blog

Short answer: A handful of math symbols have spoken names you can dictate directly, including "percent sign," "plus sign," "minus sign," "equals sign," "greater than" and "less than." Most others, and all layout like fractions and exponents, cannot be dictated. Say the math in words, then format it.

Dictation is very good at sentences and very bad at notation. That is not a bug you can configure away, and understanding why saves you an hour of trying spoken commands that were never going to work.

A transcription engine turns sound into a stream of characters. Math is not a stream of characters. It is a two-dimensional layout, where meaning lives in position: what sits above the fraction bar, what is raised into an exponent, what is tucked under a radical, what the limits of a sum are. There is no sound for "the following expression is one level up and half size." So the honest split is this. Symbols that are just characters in a line are dictatable. Structure is not.

This guide covers what actually works, in order: the symbols that have reliable spoken names, the ones that do not, the say-it-in-words workflow that most people should use for most documents, how to handle real notation in Word, Pages, Google Docs and LaTeX, and how to make the symbols with no spoken command reachable anyway using replacement rules.

The symbols you can dictate

These have spoken names in Apple's dictation on both Mac and iPhone, and they work in ordinary text fields. Say the name of the symbol as its own phrase, with a small beat before and after, the same way you would say "comma" or "period."

That list is enough for a surprising amount of everyday quantitative writing. "Revenue grew twelve percent sign" is silly, but "revenue grew twelve percent" produces "12 percent" or "12%" depending on context, and either is fine in an email. For inequalities, "x greater than sign five" gives you x > 5. For a simple formula in a comment, "open parenthesis a plus sign b close parenthesis forward slash two" gives you (a + b) / 2, which is perfectly readable code-style math.

One thing to watch: spoken symbols usually arrive with spaces around them. You will often get x > 5 when you wanted x>5, or ( a + b ) with padding inside the parentheses. That is the same behaviour that makes email addresses and URLs painful to speak, and we covered the mechanism in detail in how to dictate email addresses and URLs. For prose math it does not matter. For anything a machine will parse, it matters a lot, and you should type it.

The symbols you cannot dictate

There is no spoken command for most of the notation that makes math look like math. On stock Mac and iPhone dictation, do not expect any of these to appear when you say their names:

Say "pi" and you get the word "pi," or possibly "pie." Say "square root of x" and you get the words "square root of x." Say "x squared" and you get "x squared," not x². This is worth internalising early, because the failure mode is quiet: you get plausible-looking text instead of an error, and you only notice when you reread.

The workflow that actually works: say it in words, format it after

Here is the reframe that makes dictated math practical. Stop trying to speak notation. Speak the meaning, at full speed, and treat symbol formatting as a separate, later pass done with your hands.

Dictate this:

The variance scales with the square root of n, so quadrupling the sample size only halves the standard error. If theta is the mixing parameter, the estimator is unbiased whenever theta is between zero and one half.

That paragraph transcribes cleanly at speaking speed, because every token in it is an ordinary English word. Then, if the document needs real notation, you go back and convert the four or five places that need it. In most documents the answer is that you do not convert anything, because words were the right choice all along.

This is the single highest-leverage habit in this guide, and it has a second benefit. Written math in prose is frequently over-notated. Reviewers, colleagues and students often understand "the square root of n" faster than they parse √n dropped mid-sentence. Dictation nudges you toward the readable version by default.

The three modes, and when each is right

  1. Prose math. Words only, no symbols. Correct for emails, Slack messages, ticket comments, meeting notes, most documentation, and the explanatory paragraphs of a paper. Fully dictatable at speaking speed.
  2. Inline symbol math. Plain-text symbols in a line: (a + b) / 2, x > 5, n^2. Correct for code comments, spreadsheets, quick notes and chat where formatting is not available. Mostly dictatable, with hands for the spacing cleanup.
  3. Real notation. An equation editor or LaTeX. Correct for papers, problem sets, textbooks, slides with displayed equations. Not dictatable. Type it.

Almost every frustrating dictation session with math is someone trying to do mode 3 by voice. Decide the mode before you start talking and the friction disappears.

Numbers, which are their own problem

Getting 1,250,000 instead of "one million two hundred fifty thousand" depends on how you say it, not on any setting. The short version: read large numbers the way you would read them off a page, say "point" and not "dot" for decimals, and give units their full name so the engine knows which formatting to apply. The full treatment, including dates, currency, and why sequences of digits are structurally the hardest case, is in how to dictate numbers and dates correctly. If you write anything quantitative, read that one next; it pairs directly with this guide.

The one habit worth repeating here: number errors are silent. A misheard word looks wrong and you catch it. A misheard digit looks exactly like a number. Always reread dictated figures before you send, and never dictate a figure you have not checked against the source.

Variables and single letters

Single letters are the worst case for any transcription engine, because letters are homophones of extremely common words. Dictate "let a equal b" and you may well get "let a equal be." Dictate "solve for x" and you usually get it, because "x" has no common homophone; dictate "solve for e" and you can get "solve for eat," "solve for E," or "solve for he."

Three things help:

Real notation, app by app

Microsoft Word and Apple Pages

Both have equation editors that accept a linear typed syntax and render it into layout. You cannot drive those editors by voice: the equation field expects keystrokes that build structure, and dictated characters arrive as plain text. The workable pattern is to dictate the surrounding paragraph normally, leave a placeholder such as "EQ ONE" where the equation goes, and do a single typing pass at the end to build every equation in one sitting. Batching by input type is far faster than switching between talking and typing every two sentences.

Google Docs

Same story. The equation toolbar is keyboard-driven. Dictate the prose, mark the gaps, fill them by hand.

LaTeX and Overleaf

LaTeX is interesting because it is the one environment where math is genuinely a stream of characters. In principle you could speak it. In practice you should not: backslashes, braces, carets and underscores in dense sequence are exactly the string-like content that dictation handles worst, and one wrong brace breaks the compile. The productive split is to dictate the prose between the equations and type the math environments. We laid out the full workflow, including which LaTeX special characters to watch for, in how to dictate in Overleaf.

Notion, Obsidian and Markdown notes

These accept inline LaTeX between dollar signs and Markdown superscript conventions. Dictate the note, type the math blocks. The same placeholder trick applies.

Spreadsheets

Never dictate formulas. A spreadsheet formula is parsed strictly, so a stray space or a smart quote breaks it outright, and the failure is loud but time-wasting. Dictate the cell notes, the column headers, the explanatory text and the commentary. Type the formulas.

Anki and study notes

Flashcard fronts and backs are mostly prose with occasional notation. Dictate the prose side at speed, type the notation side. Card creation is the real grind in spaced repetition, and voice removes most of it even if the math itself stays typed.

The fix for symbols with no spoken command

Everything above is about working within the limits of stock dictation. There is a way around a large part of it.

Voice Keyboard Pro includes Smart Vocabulary, a personal dictionary with replacement rules. A rule maps something you say to exactly the text you want inserted. That means you can invent spoken names for the symbols the operating system never gave you one for:

Build the ten or fifteen you actually use in your field and a large slice of your notation becomes speakable. The same mechanism handles recurring strings that are tedious to build symbol by symbol: a rule that expands "standard error formula" into the exact expression you type forty times a semester, or one that expands a field-specific abbreviation into its full form.

Smart Vocabulary also solves the other half of the technical-writing problem, which is not symbols at all. It is proper nouns. Field jargon, author surnames in citations, dataset names, model names, chemical and drug names, internal project code names: these are the errors that come back every single time, because a general transcription engine has no reason to know them and stock dictation never learns from your corrections. Teach them once and they stop being errors.

On Mac: dictating into any of these apps

Voice Keyboard Pro lives in the menu bar. You hold a hotkey, speak, release, and the text appears at your cursor in whatever app is focused. There is no per-app plugin, no browser extension, and nothing to install into Word, Overleaf or your LMS. That matters for math writing specifically, because technical work is scattered across a text editor, a browser tab, a PDF reader and a notes app, and a tool that only works in one of them is not much use.

Two setup notes. First, pick a hotkey that does not collide with a shortcut in your editor; technical apps tend to claim a lot of the keyboard. Second, because the model is hold-to-talk, you decide where each utterance ends, which pairs well with the talk-then-format workflow: one held press per paragraph, release, then type the equation that follows.

On iPhone: capture now, format later

Phone math is where the say-it-in-words rule is least negotiable. There is no equation editor worth using on a phone screen, and hunting for a symbol in the keyboard's third layer costs more time than the idea was worth.

Use the phone for capture. The Voice Keyboard Pro keyboard has a mic button built in, so it works inside any iOS app: dictate the observation, the derivation sketch in words, the note to self about which term you got wrong. Format it on a real screen later.

Two features earn their keep here. Voice Edit lets you speak a change instead of placing a cursor by hand, which is the difference between fixing "the second term should be negative" in three seconds and giving up. And the two-way translation while dictating, across 24 languages, is genuinely useful for technical collaborators who read a field's notation fluently but write English less comfortably.

Troubleshooting

It types the words "plus sign" instead of the symbol

Almost always a pause problem. Symbol commands need to be recognised as commands, not content, and that recognition depends on the phrase sitting on its own. Leave a short beat before and after. If it still types the literal words, check your dictation language variant: symbol command sets differ between variants, and a mismatch between the variant you selected and how you actually speak degrades command recognition along with everything else.

The spacing around symbols is wrong

Working as designed. Spoken symbols are inserted as separate tokens with word spacing. If the output has to be exact, either fix it in a cleanup pass or use a replacement rule that inserts the whole expression with the spacing you want.

Everything degrades when I dictate technical content

Technical vocabulary sits far from everyday speech, so the engine has less statistical support for each word, and people unconsciously slow down and over-articulate when dictating unfamiliar terms, which makes it worse. Speak technical sentences at normal conversational pace and in full phrases rather than word by word. Then fix the recurring proper nouns at the source with vocabulary rules rather than by correcting them again each time.

A symbol appears, then changes after I stop

That is an autocorrect or text-substitution layer acting on the inserted text, not the transcription. Check your text replacement list for an entry that matches, and check whether the app itself is doing smart-substitution on quotes and dashes.

A two-minute drill

Open a plain text document and dictate this, one phrase at a time, pausing around each symbol:

open parenthesis a plus sign b close parenthesis forward slash two equals sign the mean period new line x greater than sign five and y less than sign ten period new line the sample grew twelve percent this quarter period

You should get three usable lines. Now try saying "x squared plus y squared equals z squared" and watch it come out as words. That contrast, in your own hands, is the whole lesson: characters yes, structure no.

Frequently asked questions

Can I dictate an entire equation into a document?

Not as rendered notation. You can dictate it as a linear expression using the symbols that have spoken names, and you can dictate it in words. Anything with a fraction bar, exponent, subscript or radical has to be built with a keyboard.

Is there a math mode for dictation?

No. Apple's dictation has no notation mode on Mac or iPhone, and no toggle exposes one. Specialised math input tools exist, but they are separate applications rather than a setting inside dictation.

Does saying "superscript two" work?

No. Superscript is a formatting state, not a character, and dictation inserts characters. Use the caret convention (x^2) if plain text is acceptable, or apply superscript formatting by hand.

What about Greek letters?

They have no spoken commands. Replacement rules are the practical answer: define "theta symbol," "lambda symbol" and the handful of others your field uses, and they become one-phrase insertions.

Should I dictate at all if my writing is heavily mathematical?

Yes, for the prose, which is usually most of the word count even in a technical document. Speech runs at roughly 130 to 150 words per minute against about 40 for average typing, so the explanatory paragraphs, the abstract, the discussion, the comments and the emails all get much faster. The equations were always going to be typed. Separating the two is what makes the speed real.

Where does my audio go?

Voice Keyboard Pro's servers store only operational pings. No audio and no transcript content is retained. As with any dictation, do not speak credentials, keys or account numbers into a text field regardless of the tool.

The short version

Dictation writes sentences, not layouts. Roughly fifteen symbols have spoken names and work fine inline. Everything with a shape to it, including fractions, exponents, radicals, sums and every Greek letter, does not. Say the math in words, dictate the prose around it at full speed, batch the notation into one typing pass, and use replacement rules to reclaim the symbols the system never gave you a way to speak. Voice Keyboard Pro has a free tier if you want to try the workflow on your next set of notes.

Keep reading