The short answer
A creative that sells on Meta or TikTok comes from a clear angle (why someone would buy your product), a hook that stops the scroll in the first second and organized testing that tells you what works. AI doesn't replace that method: it speeds it up. It lets you produce lots of versions of each idea in little time, so you can test more angles and hooks with the same budget. What doesn't change is that the idea has to come from your customer, not from the tool.
Angle first, tool second
The most common mistake is opening an AI tool and asking for “an ad for my product.” You get something pretty and generic that doesn't sell. Before producing anything, define:
- Who you're talking to. Not “women aged 25 to 45,” but one person with a specific problem: “the one who trains early and has no time for breakfast.”
- What hurts or what they want. The problem your product solves or the desire it fulfills.
- What their objection is. “It's expensive,” “I don't know if it'll work for me,” “I already tried something similar.”
With that you build your angles: the different reasons to buy that you're going to test. The ones we use most:
- Problem and solution. You show the problem and your product solving it.
- Desire. What life looks like with your product.
- Objection. You tackle head on the doubt that stops the purchase.
- Comparison. Your product against the alternative your customer uses today.
- Use case. A specific moment of the day when your product comes in.
- Social proof. What real customers say, with their permission.
The hook: the first few seconds decide
In the feed, people decide in an instant whether to stay or keep scrolling. The hook is the first thing they see and hear. Some approaches that work:
- A question your customer asks themselves: “Why does your coffee upset your stomach?”
- A surprising statement or one that contradicts what everyone believes.
- The result first: start at the end and then show how you get there.
- Motion or visual contrast in the first frame: something opening, spilling, changing color.
- On-screen text that reads without sound, because lots of people watch muted.
For each angle, write several hooks. It's the cheapest part to test and the one that changes the result the most.
Where AI helps (and where it doesn't)
AI is great at multiplying production. We use it for:
- Statics with your product in different settings, without a photo shoot for each one.
- Variations of the same ad: another background, another line of copy, another format (square, vertical, story).
- Short videos and animations from product photos.
- Voiceovers and translations to adapt a piece to another market.
- Avatars that present the product, in a creator-talking-to-camera format.
Some things are worth watching:
- The product has to look like it really is. If AI changes the label, the color or the size, the ad promises something you won't deliver.
- No fake testimonials. An avatar can present the product, but it can't pretend to be a real customer sharing their experience.
- Check each platform's rules. TikTok requires labeling realistic AI-made content, and Meta has its own transparency rules. Check them before publishing.
- Judgment is still human. Someone who knows the brand has to choose what goes out and what doesn't.
Formats for each platform
Meta and TikTok reward similar things, but not the same ones:
- TikTok wants the ad to look like native content: vertical, fast-paced, with sound and people talking to camera. Anything that looks like traditional advertising gets skipped.
- Meta (Instagram and Facebook) allows more variety: reels, stories, statics, carousels and catalog ads. A good static with a clear idea still sells.
- On both, on-screen text, captions and a clear call to action help.
How to test without burning your budget
Producing lots of variations only pays off if you test them in an orderly way. Our method:
- Test from big to small. Angles first (which reason to buy works?), then hooks within the winning angle, and only then formats and details.
- Change one thing at a time. If you change the hook, the background and the copy all at once, you won't know what made the difference.
- Give each creative time and budget to prove itself before judging it. Turning it off after a few hours leads to the wrong conclusions.
- Look at the metrics in order. If nobody sticks around for the first few seconds, the problem is the hook. If they stay but don't click, it's the message or the offer. If they click but don't buy, look at the landing page.
- Keep a log. Which angle, which hook, which format and what result. Over time, that log is worth more than any single creative.
- Scale what works and refresh it. Creatives wear out: when a winner starts to drop, make variations of the same angle before looking for a new one.
The goal isn't the perfect ad, but a system that finds winners every month.
Frequently asked questions
Do AI-made ads work as well as filmed ones?
They can work just as well or better, as long as they start from a good angle and show the product as it really is. What doesn't work is generic content, whether it's made with AI or a camera. Often the best combination is a mix: real product footage plus AI-made variations.
How many creatives do I need to start testing?
Rather than a fixed number, it's best to start with a few different angles and a few hook variations for each. That way every piece gets enough budget to show whether it works.
Do I have to disclose that an ad was made with AI?
It depends on the platform and the type of content. TikTok requires labeling realistic AI-generated content and Meta has its own rules. Check each platform's current policies before publishing.