“We don't understand why winning ads win — and our human criteria is not very good, but actually the LLMs they're not very good either. It's more like: mass-produce and see what sticks.”

AI for creative analysis is still primitive — we don't know why winning ads win

Despite excitement around AI creative tools, Thomas Petit argues that analysis of why ads work remains the weakest link. Feeding 200 losing and 3 winning creatives to an LLM and asking for patterns yields some output, but those patterns rarely reliably predict the next winner. The honest conclusion is that creative testing is still largely a volume game — produce many variants quickly, identify winners empirically, double down, and keep the exploration pipeline full.