Every caption tool will happily colour, scale and animate as many words as you point at. That’s the trap. Emphasis works by contrast, which means it’s a fixed budget you spend rather than a decoration you apply — and almost every clip that feels noisy rather than punchy has overspent it.
The budget
A workable rule: about one strong emphasis per 25 words of speech. On a typical 45-second talking-head clip that’s three to six words in the entire video — far fewer than the one-per-line most templates default to.
The mechanism is habituation. A visual system exposed to repeated identical salience signals stops routing attention to them, and it does so fast — within a few seconds. Emphasis on every line doesn’t produce a video where every line is important; it produces a video where nothing is, plus extra noise the viewer now has to filter.
Which words earn it
Ranked by how reliably they pay off:
- Numbers and quantities. “Three settings”, “40%”, “twice”. Specific, countable, and the eye finds digits before it finds words.
- The surprising noun. The thing the sentence is actually about, especially when it contradicts expectation.
- The word you stressed out loud. Speech has prosody; matching the visual emphasis to the vocal one makes the two channels reinforce rather than compete.
- The turn. “But”, “instead”, “until” — the hinge where the argument changes direction.
- Names and proper nouns, when the name is the point.
Which words don’t
- Generic intensifiers. “Literally”, “insane”, “crazy”, “absolutely”. They feel emphatic and carry no information, which is exactly the wrong trade.
- Function words. Emphasising “the” or “is” reads as an automated system that scored the wrong token.
- Two words in a row. Adjacent emphasis is one long emphasis with a gap in it. Let a few plain words sit between picks so each one has something to contrast with.
- Every instance of your keyword. Say it five times, emphasise it once.
Pick one device, not four
Emphasis can be colour, size, a filled box behind the word, or motion. Each is legible on its own. Stacked, they don’t multiply — they cancel, because the viewer can no longer tell what the signal is.
| Device | Reads as | Careful with |
|---|---|---|
| Colour | Category — “this word is a different kind of thing” | Needs enough luminance contrast to survive busy footage |
| Size | Volume — the visual equivalent of saying it louder | A word that grows can shove its neighbours around mid-line |
| Filled box | Object — the word becomes a thing rather than marks on video | The bluntest instrument; overuse reads as a template |
| Motion | Onset — reflexive attention before conscious processing | Costs the most attention; spend it least often |
The most expensive emphasis of all
Pulling a word behind the speaker is the strongest device available — it changes the word’s apparent position in space, not just its appearance. Which means it should be the rarest: one hero word per clip, on the line that would be the headline if the clip were an article. Two in a row reads as an effect rather than a point.
How Moonshot allocates it
Emphasis is chosen against a budget rather than per line: a language model reads the transcript and proposes the payoff words, and a global cap of roughly one per 25 words keeps only the best-evidenced picks. Emojis run on a separate, tighter budget. Behind-the-subject hero words are rarer still, and never land on consecutive clips.
You can always override it — mark a word as a hero and it stays one. The default just errs toward scarcity, because that’s the side of the trade nobody chooses by hand.
Quick answers
How many words should I emphasise in a caption?
Think of it as a budget rather than a per-line rule: roughly one strong emphasis per 25 words of speech, so a 45-second clip gets a handful, not one per line. The second you exceed the budget, emphasis stops meaning anything because there's nothing left to contrast against.
Which words should be emphasised?
The ones carrying the payoff: numbers, names, the surprising noun, the word that would be bolded if you wrote the sentence down, and the one you actually stressed with your voice. Not generic intensifiers — 'literally', 'insane', 'crazy' feel emphatic and carry no information.
Should emojis be added to captions?
Very sparingly. An emoji every few words reads as automated, because it is. Treat them like exclamation marks: rare enough that one lands. Roughly one per thirty words is a defensible ceiling, and zero is a perfectly good answer for serious content.
Why do my captions feel noisy even though each word looks fine?
Because emphasis is a contrast effect and you've spent it everywhere. Every device — colour, size, a box, motion — competes with every other. Habituation sets in within a few seconds and the viewer is back to reading flat text with extra visual noise on top.
Emphasis chosen for you, on a budget
Moonshot reads the transcript, picks the payoff words, and keeps them scarce enough to still mean something — with a one-tap override when you disagree.
Open Moonshot free