Sound is 40% of the ad and you are treating it as 5%

01Why does sound matter so much in a short ad?

Because it is the only channel that can make somebody physically want the video to stop. Picture can bore you. Sound can hurt. We shipped a film with a bed that had been tuned to pass a midrange check, and the client's note arrived before any note about the picture: a high pitched sound that bothers me in my ears. Other people had told him the same. It is logged as the worst item on that film's defect list.

It is the worst item because we did it to ourselves. A lint had failed at 4% midrange, so the drone went up an octave and partials went on until the number read 29%. The number moved. The film got worse.

A lint is a floor, not a target. When a check fails, fix the thing, not the number.

The rule that came out of the worst soundtrack we have shipped

02Audio rejects before it attracts

The picture does the stopping. The sound decides whether the stop holds. That ordering is not a preference, it follows from the fact that a large share of feed impressions begin muted, so the audio has not been heard yet at the moment the thumb decides. A great bed will not rescue a dead opening frame. A bad bed will lose a viewer the picture already caught.

Which reframes the budget question. A soundtrack buys permission to keep the attention the picture already won. That is worth less than people who love music think, and far more than people who ship stock beds behave as though it is.

03The first sound is an event, not a texture

Something has to start at frame zero. Not exist at frame zero, start. We measured 48 short ads: the median time to first sound is 0.018 seconds, forty of them are under 0.12, and not one of them fades in. A bed that swells is a bed that arrives after the decision has been made.

The same distinction runs through the whole mix, and treble is where it shows. Four films full of pouring water put no high-frequency content on any of the pours. One dragged a metal blade across a frozen tube for four and a half seconds with nothing in the mix acknowledging the tool. The two films that got it right did it with one sound each.

One sound on the one sharp thing beats a treble-rich bed.

Our own audio doctrine, after counting the pours nobody scored

04Where should a drop actually land?

Inside the beat it belongs to, and never on that beat's first frame. A drop on the first frame of a cut reads as the cut, so the ear files it as a transition noise and the moment is spent on nothing. A drop a few frames later reads as something that happened because of what you are now looking at, which is the whole point of putting it there.

05Is silence a legitimate choice?

Yes, and it is a register rather than an absence. Five films in our corpus put their biggest transitions inside measured near-silent gaps rather than on a hit. One engineers a five-frame void and starts the picture moving four frames before the sound comes back. The breath before the sentence does the work that a crash cymbal is usually asked to do, and it costs nothing.

In a talking film the gaps arrive for free, which is why sixty-plus of sixty-four creator files we studied carry no music bed at all. In a wordless film you have no speech gaps, so you have to build them. That is the whole instruction: if there is no voice, the silences are a thing you author.

06Can one music bed make a campaign feel consistent?

It makes a campaign feel like one asset. We put a single bed under three good ads and the set came back reading as one thing shown three times, which is the same disease that killed eight of our finished films in a single afternoon for sharing an argument, a subject, a world, an energy, a point of view and a borrowed genre.

Tempo is the same trap one level down. Twenty ads on a single 90 BPM grid were audible as sameness regardless of instrument. Across our reference sets tempo runs 65 to 144 BPM in one study and 51 to 208 in another. Per film, measured off the rendered audio, never asserted in a brief.

07Commit the sound before the first cut

This is the ordering rule and it is the one that actually changes output. The bed is chosen and committed as a file at direction lock, before a single frame is cut. Not a genre label, not a reference track, an artifact. Then the cut is derived from it and it from the cut table, so changing a shot length cannot drift the sound off picture.

Two craft details make that order cheap to hold. Put the cuts on a whole-frame grid and the sound lands for free: 15 frames at 24 frames a second is exactly 96 beats per minute, the cleanest integer grid available. And measure your measurements. Our own tempo detector once read a 90 BPM build as 165 because it was computing onsets across the full spectrum, where the bed's own low end drowns every transient.

The honest version of the 40% claim is that we cannot prove the number. Nobody has run the experiment and our films carry no outcome data, so the split stays a working number we argue for rather than a finding. What we can prove is the shape of the mistake: every film we have killed for its sound was killed on the sound alone, while the picture stayed exactly as good as it was. That is what multiplying looks like from the inside. The same file, halved, by the one element nobody put in the budget. The rest of it is in how a short ad is built.