What an ad format actually is

When people hear “ad format” they think of placements: a Reel, a Story, a 9:16 video. That is not what we mean.

For us a format is the recipe behind an ad that keeps working. Street interview. Car selfie. Clone comparison. Claymation. A crochet character explaining a symptom. Each one is a set of choices that stays the same no matter who is selling what.

What goes into a format

• The hook shape. What happens in the first two seconds, and who says it.
• The camera. Front-facing phone, a mic in someone’s face on the street, a slow push on a clay set.
• The pacing. How long each beat runs and where the cuts land.
• How the product shows up. Held to camera, compared side by side, or never shown at all.
• The slots. The person, the product and the place. These are the only things that change.

Everything except the slots is fixed. That is the whole trick. When we build a template from an ad, the prompt is written with the person, product and place as blanks, and the rest is locked in.

Why build around formats

Because a format comes with evidence. In our ad library we track how long ads in each format keep running, since nobody keeps paying for an ad that loses money. The medians are not close: UGC to camera runs about 24 days, street interviews about 41, podcast clips about 46.

So the question changes. You stop asking “what should my ad be?” and start asking “which proven format fits my product?” That is a much easier question, and the answer is already sitting in other brands’ ad accounts.

What this means when you generate

A model given a blank prompt has to make a hundred decisions, and it will make some of them differently every time. A format makes most of those decisions for it. That is why one take is usually enough: you are not asking the model to be creative, you are asking it to fill in three blanks.

Copy the structure, never the ad. The structure is what sells.