Most AI video ads lose the viewer in the first second. Not because the model was bad, but because someone opened a generator before deciding what the ad was supposed to do.
We produce AI commercials, product films and Reels for brands every week. The pattern never changes: the ads that look expensive were planned like a real shoot, and the ads that look cheap were simply “generated”. This is the checklist we actually work from, plus what recent research says about where audiences still push back on AI, so you can decide where AI should carry your ad and where it shouldn’t.
Key points
- In a May 2026 YouGov survey of Australians, 45% said they would trust a brand less if its ads were mainly made with AI.
- Synthetic people were the most disliked AI element in Cint’s study (46%). Realistic landscapes were the most liked.
- Let AI lead on product macros, locations and impossible camera moves. Keep real people in testimonials and emotional categories.
- Storyboard on stills, give each shot one camera move and one action, and treat sound and a final grade as half the production.
Start with where audiences push back
People are not neutral about AI ads, and pretending otherwise is how brands end up in a comment-section storm.
- In a YouGov survey of Australians run in May 2026, 45% said they would trust a brand less if they found out its ads were mainly made with AI.
- Cint’s Authenticity Audit found synthetic people were the most disliked AI element in ads (46% disliked them), while realistic landscapes were the most liked (40%).
- Research from the Nuremberg Institute for Market Decisions, covered in The Drum, found AI-made ads are accepted more easily for high-tech products and resisted in traditional, emotional categories.
Put those together and you get a simple rule we use on every brief: let AI do what it is visibly great at, and keep it away from the moments where viewers want to feel a real person.
Let AI lead on product macros, textures, liquids and pours; on landscapes, locations and impossible camera moves; and for tech, gaming, software and future-facing brands.
Use AI carefully for lifestyle scenes with people in the background, stylized or animated characters, and food and hospitality hero shots.
Keep real people in front for testimonials and founder stories, for emotional categories such as family, health and weddings, and for anything a viewer could mistake for a real customer.
The 10-point checklist we run before anything is generated
1. Write the ad, not the prompt
One sentence: who is this for, and what should they do after watching? If you can’t write that sentence, no model will save the ad. Prompts come last, not first.
2. Lock the first two seconds
Short-form ads are judged before the sound comes on. Decide the opening image and the on-screen line before anything else. A strong product shot with a clear promise beats a slow cinematic reveal almost every time.
3. Storyboard on stills, then animate
We iterate framing, lighting and composition on still images, because stills are cheaper and faster to change. Only locked frames get turned into video. This single habit decides most of the “expensive” look.
4. Build one visual world
Write a short style sheet: palette, lens feel, lighting, era, texture. Reuse the same references for every shot. When each clip is prompted on its own, the edit looks like five different ads stitched together.
5. Protect the product and the characters
Use reference images for the hero product and any recurring character, then check every frame for warped labels, extra fingers and drifting faces. Packaging is where AI lies the most, so for pack shots we often composite the real product.
6. One camera move, one action per shot
Ask a shot to pan, zoom and show a hand pouring sauce at the same time, and it will start to melt. Give each shot one move and one action, then cut.
7. Treat sound as half the production
Most cheap-looking AI ads are silent clips with a stock track on top. Real sound design changes everything: foley, room tone, small hits on the cuts, and music edited to the picture rather than laid underneath it.
8. Do an “un-AI” pass
Grade the whole edit together, add a touch of grain, keep motion blur consistent, soften over-sharpened skin, and trim clips before they drift. In our experience generated shots tend to degrade after a few seconds, so short cuts help.
9. Keep the claims real
AI will happily invent features, ingredients and results. Every claim in the ad has to be true and provable, and the product on screen has to match the product in the box.
10. Label it where it should be labeled
Platforms such as TikTok and Meta ask advertisers to disclose realistic AI-generated people, voices or scenes; color grading or background cleanup is a different matter. Rules change, so check each platform’s current ad policy before launch. Audiences expect it too: in the same Cint study, 74% said brands have a duty to disclose AI use in advertising.
Six tells that make an AI ad look cheap, and the fix
| The tell | Why it happens | The fix |
|---|---|---|
| Waxy, plastic skin | Default model aesthetics | Add texture and grain, stylize, or keep faces small in frame |
| Morphing hands and objects | Too much action in one shot | One action per shot, cut earlier |
| Warped logos and text | Models can’t spell reliably | Add text and logos in the edit, composite the real pack |
| The same floaty push-in on every shot | Default camera motion | Mix locked-off shots, real moves and hard cuts |
| Light that changes between shots | Each clip prompted in isolation | Style sheet plus one grade across the whole edit |
| Silence with a stock track | No sound design | Foley, room tone and music cut to picture |
What this looks like in a real project
For a vertical restaurant promo, we kept AI on the things it does beautifully: slow sauce pours, smoked cocktails, macro textures and the venue’s bold interior art. The edit opens on a single appetizing hero shot and runs 39 seconds for Reels and TikTok. For The Island Sports Club, the brand identity itself, the logo wall and the venue, anchors a short landscape film, so it feels like the club and not like a generic sports ad.
Notice what’s not in either piece: AI actors pretending to be customers.
Where the time really goes
People assume generation is the slow part. It isn’t. Most of the hours go into the brief, the storyboard, selecting the right takes and the edit. If an agency quotes you a fixed price for an AI ad without asking about your audience, your offer and where the ad will run, treat that as a warning sign.
Quick answers
Can AI video ads perform as well as filmed ads?
In categories where audiences accept them, yes, and they make testing cheap: you can try five hooks instead of one. In emotional categories, use AI for the world around your people, not for the people.
Do I have to say an ad was made with AI?
When it shows realistic AI-generated people, voices or scenes, most major ad platforms expect a label, and many viewers expect honesty anyway. Check the current policy of each platform you run on.
What is the most common mistake?
Generating before briefing. Every other problem on this list grows out of that one.
If you want a second pair of eyes on an AI ad before it goes live, or want us to produce one with you, get in touch with the Nexysm team. You can also see how we approach AI video production and AI advertising.
How this article was made: written by Farzad Torabi with AI assistance, based on our production work at Nexysm, and fact-checked against the sources linked above.




Comment(01)