ElevenLabs Emphasize Word: 4 Methods That Work (2026)

To emphasize a word in ElevenLabs, capitalize it — the docs say capitalization increases emphasis. Pair caps with an audio tag like [shouting], and set the word apart with an ellipsis or dash. SSML emphasis tags are not supported on v3 or v4. Each emphasized character still bills 1 credit; a 10-character word costs 10 credits. Checked 4 Nov 2026.

How does word emphasis work in ElevenLabs?

Emphasis in ElevenLabs is interpretive, not parametric. There is no emphasis slider and no loudness field — instead, the model reads typographic signals the way a human actor reads a marked-up script. ALL CAPS reads as stress (the docs say capitalization increases emphasis); an ellipsis before a word reads as a held breath; an audio tag like [shouting] reads as a delivery instruction for everything after it. The signal is the text itself, which is why every emphasis method on this page costs exactly the characters it adds — nothing more.

That interpretive nature is both the advantage and the limit. It makes emphasis available on every plan, in every language the model supports, with no settings to learn. It also means no method guarantees byte-identical stress across regenerations: emphasis is a nudge to the performance, not a value written into the file. If you need laboratory-grade repeatability, read the “what is not published” section below before committing a budget to it.

Method 1: Capitalize the word

This is the primary documented mechanism. The official prompting guide says capitalization increases emphasis — “this is HUGE news” lands harder than “this is huge news.” Caps work on v3 and v4 alike, in the web app and through the API, and they cost nothing extra: the characters were already in the script.

Two rules keep caps effective. First, one stressed word per sentence — a sentence with three capped words emphasizes none of them, because the model spreads the stress. Second, never capitalize a whole paragraph; sustained caps read as shouting, and a paragraph of shouting is just noise.

Method 2: Add an ellipsis or dash before it

The docs say ellipses add pauses and weight to the words around them. A pause before a word is a spotlight: “The winner is… MARIA.” The ellipsis costs three characters (three credits).

Dashes do the sharper version of the same job — “The winner is — MARIA” reads as an interruption rather than a held breath. Colons work too (“The winner is: MARIA”) for a formal, announced feel. Combine a pause mark with caps and you get the two most reliable emphasis signals stacked on one word — still without a single audio tag.

Method 3: Put an audio tag in front of it

When caps and pauses are not enough, audio tags change the delivery itself. [shouting] before a word makes it loud; [whispering] makes it intimate; [sarcastic] twists its tone. Tags are inline — “[shouting] Run!” — and each tag bills its characters ([shouting] is 10 characters, about 10 credits).

Tags are stronger than caps but less predictable, because they steer the whole line that follows, not just the word. Scope them tightly: put the tag immediately before the word and close the thought right after. And remember the docs’ caution — tags follow the voice. A [shouting] tag on a soft-spoken narrator gives you a soft-spoken person trying to shout, which may be exactly the effect you want or exactly the wrong one.

Method 4: Rebuild the sentence around it

Sometimes the cleanest emphasis is structural. Short sentences hit harder than long ones: “Maria won. Nobody else came close.” beats “Maria won and nobody else came close.” An exclamation mark adds lift (“Maria won!”), and front-loading the word puts it where the ear expects the point: “MARIA won the race” vs “The race was won by Maria.”

This method costs nothing — same words, new order — and it survives every model change, because sentence structure is not a model-specific trick. If caps and tags ever stop behaving the way you expect, a rebuilt sentence is the fallback that always works.

Emphasize a word in 6 steps

Six steps, each with its credit cost.

1. Step 1 — Write the sentence plain, then read it aloud. Mark the one word your voice naturally stresses. Cost: $0 — this step is thinking, not generating.
2. Step 2 — Capitalize that word. One word only. Cost: $0 extra — the characters were already in the script.
3. Step 3 — Add a pause mark before it. An ellipsis (…) or em dash (—) gives the word a runway. Cost: 1–3 credits for the mark.
4. Step 4 — If it still needs more, add an audio tag immediately before the word. [shouting], [whispering], or [excited] are the common choices. Cost: about 10–12 credits for the tag.
5. Step 5 — Generate a short test. The sentence plus one neighbor, about 150 characters. Cost: ~150 credits. Listen: does the stress land on the right word?
6. Step 6 — Only then generate the full passage. Cost: 1 credit per character of the full script. Never test emphasis inside a 5,000-character generation — every tweak re-bills the whole thing.

What emphasis costs in credits

Emphasis is the cheapest effect in ElevenLabs, because the signals are characters you mostly already had:

A fully dressed emphasized word — caps + ellipsis + tag — adds about 13 characters, roughly 13 credits. The expensive mistake is testing emphasis inside full-length generations: a 5,000-character script re-billed three times is 15,000 credits, while the same three tests at 150 characters cost 450.

What is not published about emphasis

Honesty matters here, because emphasis is where marketing claims usually outrun documentation. ElevenLabs does not publish a loudness value for capped words (no dB figure), does not publish an emphasis tag or level for v3 or v4, and does not guarantee identical stress across regenerations of the same script. The docs describe the mechanisms — caps increase emphasis, ellipses add pauses and weight — but the exact internal weighting is not public. Treat every emphasis method as a direction to the performer (“stress this word”), not a parameter (“+6 dB on this word”), and your expectations will match the product.

Common problems and fixes

The wrong word gets stressed

Another signal in the sentence is louder than yours — a second capped word, an exclamation mark, or a mis-scoped audio tag. Remove every other emphasis signal in the sentence, keep one, and regenerate a short test. One stressed idea per sentence fixes most cases.

Caps do nothing audible

The sentence may be too long or too flat for the stress to register, or the voice’s natural cadence may swallow it. Shorten the sentence, add an ellipsis before the word, and if that fails, add an audio tag. Our pause timing guide covers the punctuation side in more depth.

The tag changes the whole paragraph

Audio tags steer everything after them, not just the next word. Scope the effect: tag, word, then a period or line break to close the delivery before the next thought. Tight scoping is the difference between an emphasized word and an emphasized monologue.

Emphasis differs between regenerations

That is expected — emphasis is interpretive, and the docs publish no guarantee of identical stress across runs. If a take lands perfectly, download it right away so you keep the exact performance.

Who should NOT chase emphasis tricks

If your script is technical narration — documentation, legal text, course material — skip emphasis tricks entirely and let clean sentence structure carry the meaning; marked-up stress in a manual sounds like a sales pitch. And if your pipeline requires SSML emphasis levels or audited, repeatable loudness values, ElevenLabs’ interpretive emphasis is the wrong tool — those numbers are not published and cannot be audited. For general prompt craft, the prompt formatting guide is the better starting point.

Frequently asked questions

How do you emphasize a word in ElevenLabs?

Capitalize the word — the official docs say capitalization increases emphasis. For more force, add an ellipsis or dash before it (the docs say ellipses add pauses and weight), or place an audio tag like [shouting] immediately before it. Each signal bills only its characters at 1 credit each; caps add no extra cost at all.

Can you bold or italicize a word in ElevenLabs?

No — ElevenLabs reads plain text, so bold, italics, and other rich-text formatting have no effect on delivery. Use capitalization for stress, punctuation for pacing, and audio tags for delivery changes. If you paste formatted text from a document, the formatting is silently ignored, so convert it to caps and punctuation first.

Does ElevenLabs support SSML emphasis?

No. The docs state that Eleven v4 and Eleven v3 do not support SSML break tags, SSML support is experimental overall, and phoneme tags only work with eleven_flash_v2. There is no <emphasis> tag on v3 or v4 — caps, ellipses, and audio tags are the documented emphasis mechanisms.

How many credits does emphasizing a word cost?

Emphasis costs only the characters it adds at 1 credit per character: capitalizing costs 0 extra, an ellipsis adds ~3 credits, and an audio tag like [shouting] adds about 10. A fully dressed word (caps + ellipsis + tag) adds roughly 13 credits. Always test on a ~150-character sentence first — retesting inside full scripts is the expensive error.

Why is the wrong word emphasized?

A competing signal is louder — another capped word, an exclamation mark, or a mis-scoped audio tag. Strip the sentence to a single emphasis cue and rebuild; one stressed idea per sentence fixes most cases. If caps alone will not register, shorten the sentence and add a pause mark before the word.

Does emphasis work in other languages?

The mechanisms are typographic, not language-specific — caps, punctuation, and audio tags work in any language the model supports (70+ on v3). One caution: all-caps conventions differ by language, and some scripts have no case distinction at all, so in those languages lean on punctuation and audio tags instead of caps.

Sources checked

Emphasis and pause rules (capitalization, ellipses), audio tags, and SSML limits: official ElevenLabs prompting best-practices guide (elevenlabs.io/docs/overview/capabilities/text-to-speech/best-practices), checked 4 Nov 2026. Model facts and 70+ languages: official ElevenLabs models documentation (elevenlabs.io/docs/overview/models), checked 4 Nov 2026. Plan credit quotas: official ElevenLabs pricing (elevenlabs.io/pricing), checked 4 Nov 2026.