A video ad for a physical product has one job in its opening seconds: show a stranger the single thing they need to understand before they would buy. For a stain remover, that is the stain lifting. For a knitted throw, it is the texture up close. For a supplement, it is often the person who makes it, because the result takes weeks and no camera can capture it in fifteen seconds.
That choice decides the format. It also decides how much of the ad AI can generate, and where a claim rule or a disclosure rule starts to apply.
Start from what a first-time viewer must understand
A good product page answers eight questions: how it works, how it feels, what problem it fixes, who else uses it, who makes it, why buy now, why switch, and what arrives in the box. A 15-second ad has room for one. Pick the question your best customers asked before their first order, and the format follows from it.
A worked example runs through the rest of this piece. Say you sell a EUR 45 insulated steel bottle, and the product page promises drinks stay cold for 24 hours. That promise is a claim, and it takes a full day to prove, so no 15-second demo can show it honestly. The weight in the hand, the click of the lid and the condensation on a cold morning can each be shown in a single shot.
Guessing costs a whole batch.
Which video ad formats suit physical products?
Nine formats cover the eight questions, with the "how it works" question split in two: results you can film in one shot, and results that take days. The table maps each format to its question and to the frame it opens on, because the opening frame is where a viewer decides whether to keep watching.
| Format | Question it answers | What frame one shows | What must be true |
|---|---|---|---|
| Product demo | How does it work? | The product mid action, already in use | The result on screen is the real result |
| First-use routine | What is it like to use? | A hand reaching for it in a real setting | No result claim when results take days |
| Sensory close-up | How does it look and feel? | Texture, drape, pour or finish filling the screen | Colour and texture match the real product |
| Problem and solution | Does it fix my problem? | The problem on screen before anyone names it | The fix works as shown |
| Comparison with the old way | Why switch? | The old item beside the new product | A competitor is named only with evidence |
| Unboxing | What arrives? | The closed box, a hand on the lid | The contents match what ships |
| Founder story | Who makes it, and why? | The founder with the product | A real person telling a true story |
| Customer or creator review | Do people like me use it? | A real person holding the product | They used it; any connection is disclosed |
| Offer spotlight | Why buy now? | The product with the offer on screen | The offer is live and the end date real |
No column in that table ranks formats by results. No platform publishes conversion rates by creative format for physical products, so treat any ranking you are shown as a claim that needs its data.
Which formats can AI make well, and which need a camera or a person?
AI video copies how a product looks far better than how it behaves. That splits the nine formats into groups by what production they need.
- Generate from references. Sensory close-ups, unboxings, first-use routines and demos where nothing is being proven. Supply front, back and side photos plus the flat label artwork, then check every frame against them.
- Generate the scene, film the proof. Demos and problem-solution ads that show a result. The setting, the problem and the pack shot can be generated; the moment the stain lifts or the lid seals has to be real footage whenever the ad presents it as proof.
- Cast a real person. Founder stories and customer or creator reviews. The person, their words and their experience have to be real.
Offer spotlights and comparisons belong to the first group for production and to the second for claims. A generated steel bottle beside a generated plastic one is an illustration. A generated shot of ice still floating after a day is a claim demonstration.
Material matters as much as format. Clear glass, liquids and fine repeating patterns are the hardest things for current models to keep accurate, and a hand gripping the product fails more often than the product standing alone. The guide to products that are hard for AI video rates materials and scores individual shots.
Which formats carry claim or disclosure risk?
Reviews, demonstrations and problem-solution ads in health categories carry legal weight on top of creative risk. The rules below bind advertisers, which puts them in a stronger evidence grade than any platform tip.
Reviews and testimonials. The US Federal Trade Commission's final rule on fake reviews and testimonials, announced on 14 August 2024, bans reviews that misrepresent that they come from someone who does not exist, and the announcement names AI-generated fake reviews. The FTC's Endorsement Guides add two limits: an endorser cannot talk about experience with a product they have not tried, and a relative, employee or paid endorser needs that connection disclosed clearly. An AI presenter styled as a happy customer fails both.
AI-generated ads in a UGC style stay usable when they are labelled as generated and never pose as a customer. The explainer on UGC ads covers where that line sits.
Demonstrations. In FTC v. Colgate-Palmolive (1965), the US Supreme Court held that it is deceptive to give viewers the false impression they are watching an actual test that proves a product claim when an undisclosed mock-up is used. The shaving cream ad in that case shaved plexiglass covered in sand. A generated clip of your stain lifting is a mock-up by that definition if the ad offers it as proof.
Health, weight and skin. Meta's health and wellness ad standards require ads for dietary and weight loss products to target people aged 18 or older, and they forbid statements that attack a person's appearance, body parts or hygiene. A problem-solution ad for an acne patch has to show the situation without telling the viewer something is wrong with their face.
Check each market you sell in. These are the US and Meta baselines.
