What lands and what annoys
Lesson 20 of 22 in the course “AI Video, No Flops”. Free, with animations and a quiz.
Two videos, two reactions
The last level is about posting a video without a wave of hate, a takedown or a ban. Let’s start with how people react. In November 2024 Coca-Cola released an AI remake of its legendary Christmas trucks ad, which had been running since 1995. It was made with generative models and posed as a normal shoot. On social media it was quickly called soulless and creepy. Meanwhile the Kalshi ad from the previous level, also fully AI, was discussed as a breakthrough. Both are AI, yet the reactions were opposite. Let’s see why.
The uncanny valley
Back in 1970 Japanese roboticist Masahiro Mori described the “uncanny valley”. The more a robot looks like a human, the more we like it, but only up to a point. When it’s almost human but something is off, liking drops sharply and turns into unease. A cartoon character seems cute, a real person seems normal, but an almost-human with frozen eyes and odd fingers is unsettling. AI video often lands right in that pit, especially when it tries to look like real footage. That’s where the Coca-Cola ad fell: it wanted to look real, and the eye caught small glitches.
What irritates
Stories like these add up to a list of things viewers won’t forgive in AI ads. First: AI pretending to be a real shoot. When the trick is exposed, people get angry not at the technology but at being fooled. Second: faked emotions about a topic people hold dear, like a holiday or family. Third: made-up “customer reviews”. That’s not just cringe: in the US, an FTC rule in force since October 2024 explicitly bans fake reviews, including AI-generated ones. Fourth: faces and hands in the uncanny valley. You already know how to check all of this, and it directly affects a brand’s reputation.
What lands
Now the flip side. Kalshi worked not despite AI but with it: the ad was deliberately absurd, and the model’s weirdness became part of the joke. So rule one: humour and absurdity, where AI weirdness works for you. Rule two: impossible shots. A sneaker running up a skyscraper wall can’t be filmed, viewers know it’s AI and enjoy it. Rule three: show the process. People love seeing how something that surprised them was made. Rule four: an honest “made with AI” label. It answers “are we being fooled?” before anyone asks.
The process is content too
Every one of Mira’s projects now has two videos. The first is the ad itself, 15 seconds of result. The second is a breakdown: drafts, prompts, failed takes with extra fingers and how she fixed them. The material is already in the /drafts and /prompts folders from the previous level, so the second video is almost free. The creator of the Kalshi ad did the same: he posted a breakdown of his process, and that’s exactly where we know about the 300–400 generations and the Gemini prompts from. A breakdown builds trust: viewers see a person and their work behind the video, not a single button.
Comments are a free focus test
Nobody hits the audience perfectly the first time. So after posting, Mira runs a simple loop. In the first hours she reads the comments, especially the ones calling the video cringe or “obvious AI”. She notes what exactly bothered people: eyes, hands, voice, skin that’s too smooth, or just the idea. And she fixes it in the next video instead of arguing in the comments. Over five or six videos this loop does more than any course, this one included. Just remember: comments show the loudest people’s reaction, so read them together with the watch-through numbers.
Mira’s brand rules
From this whole lesson Mira put together four brand rules and wrote them into the agent’s brief so they apply to every video. Style: bold AI style and a bit of absurdity, with no passing pictures off as a shoot. Faces: only Mira and characters invented from scratch. Emotions: no made-up reviews or “real customer stories”. Label: always “made with AI”. The last rule isn’t only about reputation. In Europe and on major platforms, AI labelling is now an obligation, and a video can be removed without it. That’s the next lesson.
Check yourself
- A brand wants an AI remake of its most beloved old ad, “so it looks like a real shoot”. What’s the main risk?
Correct answer: The uncanny valley and a sense of substitution on something people love. An almost-real picture on a topic people cherish triggers irritation, as with Coca-Cola in 2024. - Mira is offered AI “happy customer reviews” for the ad. What should she say?
Correct answer: No: it deceives viewers, and in the US such reviews are explicitly banned by an FTC rule. A made-up review is a lie about the product. It hurts both trust and legality. - Which video idea makes best use of AI’s strengths?
Correct answer: A sneaker runs up a skyscraper wall while the city folds below. An impossible shot can’t be filmed, so viewers see value in it, not a fake. - Lots of comments under the video say “doll eyes”. What should she do?
Correct answer: Note it and fix the gaze in the next video: real blinks and micro-movements. Comments are a free focus test. A repeated complaint shows exactly what to fix.