STAYSPICY · JOURNAL 17 Two halves
Craft

Where AI helps in a 150-video month, and where it ruins it.

Every studio now says it uses AI, so the answer means nothing. The useful question is where. A 150-video month has two halves: the hours the viewer never sees and the seconds they do. We put AI in the first, keep it out of the second, and the feed now labels anyone who does not.

The month has two halves.

The month in numbers: 150 or more videos, 15 to 40 creators booked inside a week, four languages produced natively in Polish, English, German and Arabic, and 5 to 10 working days from approved brief to first batch. Most of those hours are not creative. They are logistics, transcripts, first-pass translations, selection and admin. That is where the machine earns its keep: nothing there reaches the viewer without a person in between.

A model drafts every transcript and a person checks every caption before it is burned in, because the Careless Whisper study (Cornell, 2024) found roughly one transcript in a hundred contained phrases nobody said, often seeded by a pause. Translation works the same way: a machine draft, then a native writer who rebuilds the line rather than corrects it, which is why one system runs every market. Draft by machine, decision by human.

Why the viewer-facing half is different.

Because the feed labels it. TikTok has read Content Credentials since 2024, had labelled 1.3 billion videos by November 2025, and is giving viewers a dial to see less of it. Meta has carried an AI info label since 2024 and from July 2026 applies it to ads automatically. YouTube has required disclosure of realistic synthetic content since 2024. From 2 August 2026, Article 50 of the EU AI Act requires a synthetic face or voice to be disclosed to the viewer, visibly.

Kantar's Media Reactions 2025 study found over half of consumers uneasy about AI-generated ads and over 60 percent worried it will produce misleading ones. The Nuremberg Institute for Market Decisions (2025) showed people the same ad under two labels, and the AI-generated version was liked less. Shi and Jiang (SAGE Open, 2026) found the label adds novelty but subtracts authenticity, and the loss bites hardest on hedonic products. Premium is hedonic by definition.

The feed pays for a human doing something specific, and the label now tells it when nobody did.

Where it ruins the month.

The damage is never in the admin, always in the seconds. A Gulf audience hearing a dubbed voice in the wrong dialect from a face they trust. Coca-Cola's 2024 holiday film was mocked for flickering wheels and uncanny people, McDonald's Netherlands pulled its AI Christmas ad within days in November 2025, and Guess's AI model in Vogue's August 2025 issue drew fury from readers and models. None of them lacked budget, only a person in the frame.

Then the law. Tennessee's ELVIS Act (2024) added a person's voice to the rights the state protects, and Denmark drafted a right to one's own face and voice in 2025. Getty Images v Stability AI (High Court, 2025) showed how little copyright could do about training data, which leaves the contract. So we do not clone a creator's face or voice in any language, which is the deal behind 15 to 40 creators bookable inside a week.

Where the argument stops.

The other column. AI-generated creative genuinely wins in three places. Performance variants at volume, when paid needs forty packshot versions by Thursday. Product-only shots, where nothing can be uncanny. And markets where the label does not depress performance: Shi and Jiang found utilitarian products gain from novelty, and the IAB with Sonata Insights (2026) found 73 percent of Gen Z and millennials said disclosure would raise or not dent their intent to buy.

And the line will move. System1 testing reported in November 2025 put 18 AI-assisted ads at an average of 3.4 stars against a database average of 2.3: a small, flattering sample, but a result. Tools improve yearly, labels fade into menus, generations stop caring. So we test the line rather than defend it, in the only honest creative test there is, and the day a labelled AI video beats a creator in our numbers, we rewrite this article.

What this means for your brief.

Ask any studio where AI sits in its month. The right answer names the invisible half: transcripts, translation drafts, cutdown selection, scheduling, reporting. It names a person for the hook, the face, the voice and the joke. And it settles the label before the shoot, not after the platform does. Our organic record stands above 100 million views for one brand. The machine did much of it. The viewer never met it.

Where does AI sit in your next 150 videos?

Tell us about your brand and your goal. We will come back with a plan and the first step.

Book a call
Keep reading