AI creative that doesn’t look like AI

The first wave of AI-generated ads was easy to spot. Glossy faces that didn’t quite blink right, hands with the wrong number of fingers, product shots that looked like the product but not exactly. Audiences learned quickly, and “this looks AI” became a reason to keep scrolling.

The tools have improved a lot since then. But the lesson still holds: the goal isn’t to make AI ads. It’s to make good ads faster, with AI doing the parts it’s genuinely good at. Here are the principles we work by.

Start from something real

Our strongest AI-assisted ads almost always begin with real material. That might be a creator filming with the product, real product photography, actual customer reviews or footage from the founder. AI then extends that material: new hooks, alternative edits, different aspect ratios, translated voiceovers, background changes, fresh openings for a proven ad.

Starting from scratch with fully generated scenes is sometimes useful for concept exploration or simple product visuals. As the core of a performance ad, it rarely holds up.

Use AI for variation, not fabrication

There’s a clear line we don’t cross:

• We do use AI to vary, adapt, resize, translate and speed up production of real creative.
• We don’t use AI to invent customers, testimonials, before-and-after results or claims.

Fabricated proof is a bad idea on every level. It misleads people. It’s increasingly restricted by platform policies and advertising rules. And when audiences notice, they lose trust in the brand. A single real review read by a real person is worth more than a polished fake.

Keep the product accurate

AI image and video tools can subtly change a product’s shape, colour, texture or labelling. Small differences matter when someone receives the real thing. We check every frame featuring the product against the actual item, and we prefer workflows where the real product image or footage is kept intact while the surroundings change.

Watch for the tells

Before anything goes live, someone reviews it specifically for signs of generation:

• Faces and hands, especially in motion.
• Text in the scene, which models often garble.
• Lighting and shadows that don’t match between product and background.
• Voices that sound slightly too smooth or oddly paced.
• A general “stock” feel: too perfect, too clean, too generic.

When something feels off, it usually is. The fix is often to use less generation, not more.

Match the platform

Native-looking content outperforms polished content on most social platforms. AI can make ads look too finished. We often push in the opposite direction: shorter cuts, simpler text overlays, real-sounding audio, phone-camera framing. The aim is to look like something a person would post, because that’s what people stop for.

Label honestly where it matters

Platforms increasingly label or require disclosure of AI-generated or significantly altered content, and rules vary by market. We follow platform requirements and err on the side of transparency, particularly for anything realistic involving people.

Where AI shines

Used well, AI dramatically expands what a small team can do:

• Dozens of hook variations on a proven ad in an afternoon.
• Voiceovers and subtitles for new markets.
• Fast cutdowns and resizes for every placement.
• Quick concept mock-ups to test an idea before investing in a shoot.
• Background and setting variations for real product photography.

None of these replaces a creator, a strategist or an editor with taste. They multiply them. The best AI creative is the kind nobody notices, because they were too busy watching the ad.