You have to approve an AI-made ad and you have never made one. Look at five things, in this order: the pack, the people, the light across the cuts, the first three seconds, and whether the idea survives without the technique. The first three decide whether it embarrasses the brand. The last two decide whether it sells.
I review AI work from both sides of the table — as the director who made it and as the person inside an FMCG group who has to put it in front of a brand team. The questions below are the ones that actually move a film between versions. None of them need you to know a single tool.
What should I check first in an AI ad?
The pack. Pause on every frame where the product appears and read it like a shopper would: logo, colours, the name, the flavour line. It is the one thing a viewer has to recognise on the shelf, and it is the thing a generated frame is most likely to get subtly wrong — a letter swapped, a colour shifted, a shape softened.
On Tiny — One Bite, Another World the film crosses from a real kitchen into an animated chocolate world. Everything around the pack was allowed to become fantasy. The pack itself was held photoreal in every frame, on purpose. If you remember one rule from this page, make it that one: the product is not a creative variable.
How do I spot a character that keeps changing?
Watch the film once with the sound off and look only at faces and mascots. Does the child in shot two have the same hair, the same age, the same clothes as in shot six? Does the mascot keep its proportions when it turns? In my productions, continuity takes more effort than any other part of the film, and it is the note most reviewers only notice on the fourth viewing — after it has already been approved.
A mascot is a promise in a costume. On Mooo Greek Yogurt the pilot-cow has to be the same cow in every frame, or the brand is introducing a stranger. Ask to see all the character shots side by side, as stills. Differences that hide in motion are obvious in a grid.
Why does an AI ad sometimes feel “off” when I can’t say why?
Usually it is the light. Each shot can look beautiful on its own and still cut badly, because the light comes from a different side or the colour temperature jumps between shots. The viewer reads it as “cheap” without knowing why.
The hardest part of the Tiny film was exactly this: the live-action side and the animated side had to share lighting direction and colour temperature, or the cut reads as two different films instead of one journey. When a film feels off, ask one question: “Is the light coming from the same place in every shot?” It is a precise note, and a director can act on it.
What matters most for social: the first three seconds?
Yes. A small product in a vertical feed has about three seconds to earn a scroll-stop. Review the film at phone size, in the feed format it will actually run in, not on a meeting-room screen. Then stop it at second three and ask: would I stop here? If the answer depends on something that happens at second eight, the film is in the wrong order.
Also check the safe area. A film built for 9:16 from the first frame keeps everything important away from the edges where the platform puts its buttons. A film cut down from a horizontal edit usually does not.
Does the idea still work without the AI?
Describe the film to a colleague in one sentence, without mentioning how it was made. If the sentence is “it’s a cool AI video of a cake”, there is no idea yet — only a technique. If the sentence is “two kids sneak a wafer, dad catches them, and then he joins the secret”, that is an idea, and it would work as a live-action film too.
That second sentence is Hit Wafers: the whole film hangs on one sign, a finger on the lips, that changes owner halfway through. The technique makes it affordable. The sign makes it work.
Which notes are cheap to fix, and which mean starting again?
This is the part nobody tells brand teams, and it decides whether your second round takes a day or a week.
- Cheap: pacing and cut order, music, supers and end-card copy, the grade, one shot that needs regenerating.
- Expensive: changing what the main character looks like after the film is built, changing the setting, changing the idea. Each of these touches every shot that has already been approved.
- Not negotiable, whatever it costs: anything wrong on the pack.
The expensive notes are cheap if you give them at the right moment. Approve the character and the key frames as stills before any motion is made. A face change at the storyboard stage costs an afternoon. The same note after the edit costs the film.
What to do on your next review
Ask for two things before the first frame moves: a page of non-negotiables — pack, character, line — and a grid of key frames to approve as stills. Then review the film in order: pack, people, light, the first three seconds, the idea. If the pack is wrong, stop there; nothing else matters until it is fixed. If you want this done with someone who has shipped these films on real deadlines, this is how I scope it.
Questions brand teams ask
Do I need to understand AI tools to review an AI ad?
No. Review it the way you review any film: the product, the people, the light, the opening, the idea. The difference with AI is timing — approve characters and key frames as stills early, because changing them later touches every shot.
What is the most common mistake in AI-made ads?
Treating the pack as part of the creative. The product must be exactly right in every frame; everything around it can be imagined. Check logo, colours, name and flavour line frame by frame before anything else.
How many review rounds should an AI ad take?
It depends less on the tool than on when you give the expensive notes. Approve the idea, the characters and the key frames as stills first; after that, most notes are cheap ones — pacing, music, copy, the grade.