AI creative production at scale — and the disclosure rules

0:00 AI creative at scale

You can now produce in an afternoon what used to take a studio a week. That's the exciting part. The dangerous part is that you can also produce a misleading ad, a fake testimonial, or an undisclosed deepfake in the same afternoon. In this lecture you'll learn a production workflow that scales AI creative safely, what the major platforms and laws require you to disclose as of September twenty twenty-six, and a simple creative log that protects you when someone asks how an ad was made.

0:37 Why it matters

Why does this matter? Because AI has changed the economics of creative. Producing fifty variations used to be a budget decision. Now it's an afternoon. That sounds like pure upside, but it's like giving everyone in the office a printing press. Suddenly the question isn't whether you can print, it's whether what you print is true, legal and on-brand. The teams that win with AI creative aren't the ones generating the most. They're the ones with a workflow that turns fast generation into trustworthy ads, every single time.

1:15 Where AI helps

Let's map where AI helps. Research: summarizing reviews and clustering objections. Scripting: drafting hooks and localized versions. Visuals: generating backgrounds, expanding images, placing products in scenes, even generating video. Voice: synthetic voiceover and dubbing. Editing: resizing to vertical and square, captions and cut-downs. And variation: headline and description options. At every step, a human still owns the truth, the brand fit, the rights, and compliance.

1:43 Choosing tools

You'll find these tools inside the platforms and outside them. Google Ads can generate assets. Meta's Advantage plus creative can generate backgrounds, expand images and vary text. TikTok has Symphony. Then there's a whole world of external image, video and voice tools. When you choose, look at output quality, commercial-use terms, brand controls, and whether the tool embeds provenance metadata, like C2PA content credentials, that tells platforms how the content was made.

2:14 Six-step workflow

Here's a workflow that scales. One, brief each concept from your angle matrix, including which claims are allowed. Two, generate three to five drafts using templates and a brand kit. Three, human edit: check the product looks exactly as it really does, fix odd hands and garbled text, and check cultural fit. Four, compliance: claims, disclosures and rights for music, likeness and trademarks. Five, label and log. Six, launch into a structured test.

2:46 Simple example: Dubai perfume (illustrative)

Here's a simple worked example of the workflow. A Dubai perfume brand wants a lifestyle image for a new fragrance. They generate a desert sunset background with an image model, then composite their real bottle photo into it without changing the bottle. A designer fixes the lighting and checks that the label text is exactly right. The compliance checklist asks: are we claiming anything? No, just a mood. Is there a realistic person? No. Is metadata preserved? Yes. On Meta, the platform may apply an AI info label, which is fine. Every step goes into the creative log in two minutes.

3:30 Platform disclosure

Now disclosure, and remember these rules change, so always check current policies. Meta automatically adds an AI info label to ads made or significantly edited with some of its own generative tools, and can detect third-party AI content using signals like C2PA metadata. Political and social issue ads need explicit disclosure of realistic altered content. TikTok requires labels on realistic AI-generated content. Google requires disclosure of synthetic content in election ads.

4:01 The law

And then the law. In the European Union, Article fifty of the AI Act applies from the second of August twenty twenty-six. If you publish a deepfake, meaning realistic AI-generated or manipulated images, audio or video of people, places or events, you must disclose it. Consumer protection law matters just as much. In the US, the FTC, and in the UK, the ASA, judge the overall impression. If an AI render makes your product look better than it is, a label won't save you.

4:38 Safe default policy

There's also a hard line on testimonials. In the US, the FTC's rule on fake reviews and testimonials prohibits presenting fabricated testimonials, and an AI-generated person presented as a real customer falls squarely into that. So here's a safe default policy for any market. Label realistic synthetic people, voices and scenes. Never depict the product doing what it can't do. Never present synthetic people as real customers. Keep provenance metadata intact. And get written consent for any cloned voice or likeness.

5:13 Example: Manchester skincare (illustrative)

Here's how a Manchester skincare brand does it, illustratively. Real customers film testimonials with signed consent, and AI only handles captions and resizing. Product-in-scene images are generated around the real packshot, never altering the product. Before and after images are never generated. A licensed synthetic voice reads explainer ads, labeled where required. And every asset gets a row in a creative log listing tools, synthetic elements, claims checked, disclosure applied and the human reviewer. When a platform or regulator asks, they can answer in minutes.

5:50 Mistakes + try this now

Let's go through the mistakes. Generating a synthetic person and presenting them as a real customer. Letting AI change the product's appearance or performance. Using a cloned voice without written consent. Stripping provenance metadata when exporting. And letting AI translate regulated claims without native legal review. Try this now: pick your last five ads and fill in a creative log row for each: which tools were used, what's synthetic, what claims are made, and whether a disclosure applies. If you can't answer for any ad, that's where your process needs work.

6:30 Watch me do it: one ad, end to end (illustrative)

Watch me do it. Let's produce one AI-assisted ad for an illustrative Lahore furniture brand, end to end. The brief: concept, work from home without the back pain, persona remote workers, proof, ergonomic chair specs. Step one, generate: three background scenes of a bright home office using an image model, with our brand palette in the prompt. Step two, composite: our real chair photo goes into the best scene. I check the chair's shape, color and armrests exactly match the product. Step three, copy: an LLM drafts five headlines; I keep two and delete one that says eliminates back pain, because that's a health claim we can't substantiate. Step four, voice: for the video version I use a licensed stock synthetic voice, not a cloned one. Step five, compliance: no synthetic person, no altered product, claims checked, music licensed. Step six, label and log: the creative log row lists the image model, the voice tool, synthetic background and voiceover, claims reviewed, and the reviewer's name. Total time, under an hour. And if a platform or regulator asks how it was made, the answer takes one minute.

7:51 Recap and next step

Recap. AI makes production fast, but people own truth, rights and compliance. Use the six-step workflow. Follow platform labels and the EU AI Act, and remember that misleading is misleading, label or not. Your next step: write a one-page AI creative policy for your team, covering approved tools, what must be labeled, what can never be generated, and the fields in your creative log.