Numbers in French audio: should you type digits or words?

When you write an article that will be read aloud, one small decision recurs at every figure without your noticing: do you type « 20 % » or « vingt pour cent », « 1er » or « premier », « 2 000 » or « deux mille »? On the page the two forms are interchangeable. To the ear, sometimes they are too, and sometimes not. Here is how to decide without turning every number into a dilemma.

The default rule: write for the eye

The most common instinct is the wrong one: spelling everything out in words "to help the machine". It is pointless and counterproductive. A competent reading engine knows that « 20 % » is said "vingt pour cent", that « 1 200 € » is "mille deux cents euros", and that « 15 h 30 » is "quinze heures trente". It does this text-to-words step before speaking, and it does it well on standard French forms. Spelling « vingt pour cent » out in the body of an article changes nothing in the audio and weighs the reading down on screen, where most of your visitors will land anyway.

So the rule fits in one sentence: write the number in the form that reads best to a human eye, and let the engine say it. An amount is written in figures in a news article, a year is written in figures, and so is a percentage. You would not write "the revenue rose by twenty-three point four percent" in published copy, so do not do it for the audio either. This is exactly the spirit of our stance on reading French: preparing the text is the tool's job, not the writer's.

The three cases where you do have to step in

The "write for the eye" rule has three exceptions, and they are the only ones. All three come from the same fact: certain shorthand spellings are ambiguous even to an attentive human reader, and ambiguity on the page becomes an error out loud.

First, the English decimal point. If you type « 3.5 » with a dot instead of « 3,5 » with a comma, you have written an English-style number inside French text: the engine may read it "trois virgule cinq", but it may also read "trois point cinq", depending on the surrounding context. The French comma removes the doubt at once, and it is the correct spelling in any case.

Second, glued magnitude abbreviations: « 150K », « 2M », « 50k abonnés ». They are clear in a dashboard, far less so aloud, where « 150K » risks coming out "cent cinquante K". In text meant to be heard, write « 150 000 » or « deux millions »: it also reads more plainly for a hurried eye.

Third, ranges joined by a dash: « 10-12 », « pages 30-45 ». A dash between two numbers has no obvious reading, and it can surface as "dix tiret douze". Write « de 10 à 12 », « pages 30 à 45 ». You help the ear and lose nothing on the page.

The everyday hesitations, and what they produce

Here are the choices writers face most often, and the form to prefer. In every one of these, both spellings are spoken correctly; the right-hand column simply flags the one that costs nothing on screen and leaves no ambiguity.

| You hesitate between | What the audio does | Prefer |
| --- | --- | --- |
| 20% / 20 % / vingt pour cent | "vingt pour cent" in all three | 20 % (clean on screen, said right) |
| 3.5 / 3,5 | dot: uncertain reading | 3,5 (French comma) |
| 2000 / 2 000 / deux mille | "deux mille" | 2 000, or « en 2000 » for a year |
| 1er / premier | "premier" | 1er (expected written form) |
| 10-12 / de 10 à 12 | dash: "dix tiret douze" possible | de 10 à 12 |
| 150K / 150 000 | "K": uncertain reading | 150 000 |
| N°3 / n° 3 | "numéro trois" | n° 3 (spaced, safer) |

This table is the writing-side mirror of the one we give on the reading side in reading numbers aloud: one says what the reader does with a written form, the other says what you are better off writing.

When a form still resists

Some cases are specific to your field: a regulatory reference, a stock code, a rare unit, a numeric acronym peculiar to your sector. For those, there is no need to contort your published text: the pronunciation lexicon lets you fix, once and for all, how a term is said, without the displayed article carrying any trace of it. You keep clean spelling on screen and a correct reading in the ear, which is exactly the goal.

The overall line stays the same, and it matches writing for the ear: you do not write a different text for audio, you write a good text, legible on screen, and you only hand-correct the three forms that genuinely mislead the ear. Everything else the engine handles, and that is its job.

Read next

• Adding audio to your site without a developer: what is actually no-code
• How long does an article's audio actually last?
• Changing the voice of an already-generated audio: what happens, and what it costs