The Meta Ads landscape has fundamentally shifted in 2026, from audience targeting to creative targeting. The algorithm needs less and less manual audience input and more and more creative variations to test with. That's a huge opportunity for brands that understand it.
Why volume beats perfection
Meta's machine learning works best when it has many options to experiment with. One perfect video ad you run for three weeks loses to twenty variations you refresh weekly. The platform continuously tests which creative converts best for which user. The more variations you offer, the more precisely Meta can optimise. We test 100-150 creatives a month per account on average.
The win rate is typically around 5-8%. That sounds low, but those 5-8 winners then scale incredibly hard. One creative running 6x ROAS can carry an entire account. The game is: test fast, kill fast, and scale winners. With Meta Ads as the base we build this system for all our clients.
AI as a creative production machine
This is where AI comes in. Tools like Runway, Kling, Pika and Midjourney make it possible to produce creative variations in a fraction of the time. Think: product placements in new settings, background variations, UGC-style videos, and even full ad scripts generated and produced by AI.
We combine AI-generated content with real UGC and studio content. The AI creatives serve as test material, cheap and fast to make. What converts is then validated with higher production value. This hybrid model cuts your cost per creative by 60-70% while tripling your test volume.
The practical workflow
Our workflow: every week we produce a batch of 10-15 creative variations per account. Of those, 40% is AI-generated, 40% UGC, and 20% existing material in new formats. After 3-4 days we analyse the results and scale the winners. After 7 days we refresh the underperformers. This continuous testing process is why our clients consistently sit above 3x ROAS. Have a look at the results at Ares Gym Clothing to see what this delivers.