How to dictate in Word and Google Docs in any language

Picture an old government office with a row of service counters, one for each language. There is a counter for Spanish, a counter for English, a counter for French. You can only stand at one counter at a time. If your business is entirely in Spanish, fine, you never leave that line. But the moment your paperwork has one English word on it, you have to gather your things, walk over to the English counter, wait, then walk back. That walk, over and over, is what native dictation feels like the instant you write in more than one language.

And almost everyone writes in more than one language. Think about the last message you typed. A Spanish email that mentions Next.js and ChatGPT. A note with a French client's name in it. A sentence you deliberately wanted in English, dropped into an otherwise Spanish paragraph. That is normal writing, not an edge case. This guide explains why Word and Google Docs make you pick one language at a time, what that costs you, and the honest fix if your writing is genuinely multilingual.

Real writing is not monolingual

Here is the thing nobody designs dictation around: the mixed-language sentence everyone actually writes. 'Te mando el link del repo por Slack, el deploy ya está en Vercel.' Four of those words are English brand and tech terms living inside a Spanish sentence. You did not code-switch to show off. That is just how a bilingual professional talks about work.

A monolingual dictation tool has to guess. It was told 'this session is Spanish', so when you say 'Slack' or 'Vercel' or a French name, it runs those sounds through a Spanish model and writes down the closest Spanish-sounding thing it can find. The result is a sentence that is mostly right and embarrassingly wrong in exactly the words that mattered, the product name, the client, the term. And you only notice after you have moved on.

Why native dictation makes you pick ONE language

Let us be fair and precise, because this is often described wrong. Word and Google Docs are not missing languages. They both support a lot of them. The limitation is that you select one at a time from a menu, and there is no automatic detection of a second language mid-sentence.

In Microsoft Word's Dictate feature, you set one Spoken Language at a time from the dictation settings (the gear icon). It then listens in that single language. Word covers roughly 20+ fully supported languages plus more in preview, which is generous, but you are always at exactly one counter. There is no code-switching and no auto-detection. See Microsoft's Dictate support page.

Google Docs Voice typing supports over 100 languages and accents, which is even broader. But you still pick one language per session from the dropdown next to the microphone, and there is no auto-detection, so switching languages mid-writing means pausing and changing the menu. See Google's Voice typing support page. One more catch worth knowing: Google Docs states its voice editing and formatting commands (select, bold, delete, move around) are available only in English, and both your account language and document language must be English for those commands to work at all.

Windows has its own system-wide dictation too, launched with the Windows key + H shortcut. Its input language follows your active keyboard or input language, again one at a time. See Microsoft's Windows voice typing page. Same pattern everywhere: many languages available, one active at once, no mixing.

The language-switch tax

Because there is no auto-detection, a mixed-language sentence leaves you two bad options, and both cost you.

1. 1

   Option A: stop and switch the setting

   You feel an English term coming, so you pause your thought, open the language menu, change the counter, say the word, then change it back. Do that three times in a paragraph and dictation is slower than typing. The whole point was to not stop.

2. 2

   Option B: let it guess and fix it after

   You just keep talking and let the Spanish model take a swing at 'Next.js' and your French client's name. Then you go back and repair every foreign word by hand. The editing pass eats the time you saved.

That is the language-switch tax. It is not one big failure, it is a hundred small ones, a mangled brand here, a wrong name there, a stop-and-menu dance in the middle of a sentence. It is exactly the friction that makes multilingual people quietly give up on dictation and go back to the keyboard.

What you actually need

If your writing really is multilingual, the counter model is the wrong model. What you want instead is a single clerk who simply speaks every language, so you never move between counters at all. In practical terms, that means four things:

• An engine trained on many languages at once, so it is not locked to a single one you had to declare up front.
• Graceful handling of a mixed-language sentence, so an English brand name inside a Spanish sentence lands right instead of being forced through the wrong model.
• Dictate in one language, output in another, for the times you think in Spanish but need the text in English (native dictation does not do this, it only transcribes the language you selected).
• Teach it names and brands, so 'Next.js', a client's surname, or a product name spells correctly no matter which language surrounds it.

Where Golem fits (and honest alternatives)

Golem is one tool built for exactly the multilingual case. It runs OpenAI's Whisper, which was trained on and can transcribe about 99 languages (with the honest caveat that accuracy varies a lot by language), and then adds an AI cleanup pass on top. Because it is one many-language engine rather than one counter you pre-select, it handles a mixed-language sentence more gracefully, it can dictate in one language and output in another, it has custom vocabulary so you can teach it names and brands, and it drops the finished text into any app, Word, Google Docs, email, chat, a terminal, not just one program. It needs a login and is free to start with no card.

Golem is not the only option, and it would be dishonest to pretend otherwise. Wispr Flow supports 100+ languages and offers code-switching with automatic language detection too, though its own docs note code-switching works best with the occasional foreign word rather than full sentence-by-sentence alternation, and that Chinese-English is the hardest case. Superwhisper is a strong Mac-first option with on-device modes. And the most honest point of all: if you truly only ever write in a single language, the native tools in Word and Google Docs are genuinely fine, and free with software you already have. The multilingual engine only earns its keep when you actually cross languages.

So which suits you?

• Write in exactly one language, always? The native tools in Word and Google Docs are enough. Pick your language in the menu and go.
• Mix a second language, brands, or foreign names into your writing? You want a many-language engine with code-switching so you stop paying the switch tax.
• Need to think in one language and write in another? Native dictation cannot do this. Look at a tool with dictate-in-one, output-in-another, like Golem.
• Care about privacy or working offline? Consider an on-device option such as Superwhisper.
• Want single-language, step-by-step help first? Read our companion guide, How to dictate by voice in Word and Google Docs.