AI-made ads can beat human designs, when brand fit is built in

Researchers from INSEAD and the University of Wisconsin–Madison tested an AI system that generated and refined visual ads for a real Instagram campaign. Published in Marketing Science in 2026, the study found the AI-generated ads achieved a 0.98% click-through rate, against 0.65% for designs made by human designers and 0.78% for an AI system that optimised for looks alone [1].

Eighteen months later, without retraining, the AI set still performed slightly better (3.38% vs 3.24%) and was more consistent [1].

The important caveat

It worked because the system optimised for performance and brand alignment together. Simply making things more attractive wasn’t enough. It was also one company and one campaign, so the authors are careful about generalising [1].

What we take from it

• Use AI to explore, not to decide the brand. Let it generate and test variations inside clear brand rules.

• Write the rules down. Tone, colours, what you’d never show. AI can only stay on-brand if ‘on-brand’ is defined.

• Keep testing. The gains came from learning over time, not from a single perfect output.

• Be open about it. Gartner found half of US consumers prefer brands that avoid GenAI in customer-facing content [2], so quality and honesty both matter.

Sources

1. AI-generated ads outperform human designers in live campaign, advantage holds 18 months later, Tech Xplore (August 2026)

2. Half of consumers prefer brands that don’t use generative AI, Gartner finds, Retail Dive (March 2026)

If buyers are asking AI tools about your category, the first fix is a clear, consistent description of what you do and who it’s for. That’s what our Positioning Sprint settles in two weeks.