Sound off captions: three to five words is your whole sentence

Every one of the six packs captions the ad as part of the batch. There is no separate step and no separate charge.

What differs between packs is the shape: one soft line in clean-ugc, stacked capitals in bold-impact, a sweeping highlight in karaoke-flow.

The word ceiling differs too, from three in the fast packs to five in editorial.

That ceiling is the practical size of a sentence for a muted viewer, and it is the number to write against.

You can switch the caption style, colour, weight, size and position afterwards without re-cutting the ad.

So the decision is not whether to caption. It is what the words say when they are the only channel working.

A muted viewer decides in the first second. In that second they get one caption group and one frame.

So the opening group has to carry a claim rather than a preamble. As a busy parent I found that is four words of nothing.

Put the specific first. A number, a price, a name, a result. Abstract openings do not survive being read at a glance.

Write the opening headline yourself rather than accepting the drafted one, because it is the highest yielding text in the ad.

Then emphasise one word so it pops in the captions, which works in every pack and does the job a stressed syllable would have done in speech.

That emphasis is the only tone of voice available with the sound off, so spend it deliberately.

Speech carries information that captions drop: emphasis, pace, hesitation, warmth.

The compensation is visual. Highlight a phrase and a clip lands over exactly those words, so the picture carries what the tone would have.

At normal pace up to 45 percent of the runtime can be covered, and at fast pace 52 percent, which is a lot of room to show rather than tell.

So a muted ad should have more footage than a heard one, and picking a faster pack raises that ceiling.

Numbers on screen work harder than numbers spoken, because reading a figure is more reliable than hearing one.

Every caption group is a chance to change what is behind it, and most people use a fraction of them.

Go through the transcript once and mark every phrase that names something a viewer could see.

Each of those marks is a place where the picture can carry meaning the muted audience would otherwise miss entirely.

Highlight the phrase and a clip lands over exactly those words, so the pass takes minutes rather than an afternoon on a timeline.

Most people mark four or five phrases and stop. Going through the whole transcript once usually finds twice that many.

The ones people miss are the small concrete nouns rather than the obvious product mentions, and those are the marks that make a muted ad legible.

Play the finished export on a phone, muted, held at arm's length, once.

If you cannot say what the product does after four seconds, a viewer who did not make it certainly cannot.

Watch where your eye goes. If it goes to the captions and stays there, the footage is doing nothing.

Then check readability. Small type on a bright frame is the most common failure. Caption colour, weight and size are all adjustable.

None of those changes re-cut the ad, so testing costs a minute rather than another batch.

This test catches more problems than any amount of reviewing on a monitor with sound.

A script written to be heard. Long sentences broken into four word groups become a slideshow of fragments. Rewrite it for the eye and the captions do the rest.

A testimonial whose power was in the delivery. Read as text the same words are ordinary, so the picture has to carry the warmth instead.

An ad where the product makes a sound worth hearing. That evidence is gone with the sound off, and b-roll on the right phrase is what replaces it.

All three fixes happen inside the same batch you already paid for. None of them is another shoot.

Three plain facts to plan around. Nothing of yours is stamped on the export, so no logo and no closing frame carries the brand at the end.

The output is 9:16 only, and there is no timeline for frame-level typographic work.