i tried running only AI ads for 30 days (it printed lol)

Mark opens by sharing his March results: $2.2 million in sales from just over $1 million in ad spend, all driven by AI-generated creatives. He frames the core shift AI has brought to advertising through what he calls the old 'pick two' tradeoff — quality, speed, or low cost — arguing that AI now allows advertisers to achieve all three simultaneously. While this lowers the barrier to entry dramatically, it also causes an exponential increase in competitor ad volume, making the landscape far more competitive.

To address this, Mark argues that the true competitive moat is no longer creative volume or technical quality — both of which are now commoditized — but rather the thinking and research behind each creative. He introduces the concept of finding a 'winning message' rather than a winning ad, drawing on direct response marketing history and referencing author Michael Masterson. His thesis is that a strong sales message can be adapted across hundreds of ad formats, while competitors merely copying surface-level creatives will consistently underperform.

Mark then outlines what he calls 'AI non-negotiables.' The first is that copywriting remains the foundation, and that psychographic research — which he says should consume 60–80% of a marketer's time — must precede any creative production. He references a 'foundational docs SOP' he built using AI to extract copy ingredients like unique mechanisms, discovery stories, and levels of awareness. The second non-negotiable is funnel quality as a creative multiplier, using a 'rising tide raises all ships' analogy to show that improving conversion rate and average order value raises the performance floor and ceiling of all creatives simultaneously. He argues funnel diversity is more impactful than creative diversity.

On the technology side, Mark breaks down his specific AI stack. For images, he relies almost exclusively on Flux Dev (Nano Banana Pro via Hailuo/Higgsfield) with Canva's Magic Layer for post-editing. For video, he uses Kling 3.0 as his primary model for quality-cost balance, Veo 3.1 for speed and image-to-video workflows, and Seedance 2.0 exclusively for remaking already-proven top performers due to its high cost. He emphasizes always doing image generation first in Flux, then passing to image-to-video models.

Mark then analyzes three real ad examples: an AI UGC video using a first-person 'I was just like you but worse' framework targeting a solution-aware thyroid/eye disease market; a native static image ad using visual contrast and provocative copy ('Please stop buying oral B12 for neuropathy'); and an AI animation ad for a liquid collagen product, which he notes performs well despite — or because of — its deliberately simple, non-realistic aesthetic. He concludes that AI realism is no longer the goal, that animation and stylized formats drawn from TV commercial history can outperform hyper-realistic content, and reiterates that AI is a vehicle and multiplier, not a strategy in itself.