For the study, the researchers were granted year-long access to the creative department of RTL Nederland, where advertising professionals and graphic designers develop campaigns and promotional materials for television programmes, streaming platforms, and brands. The team observed how AI tools were used in daily practice, interviewed staff members, and analysed 207 AI-generated images, including the associated prompts, responses, and creative decisions.

One of the central conclusions is that AI leads to what the researchers call “process collapse”: the creative process effectively folds in on itself. Steps such as brainstorming, sketching, visualising, and producing increasingly merge together through AI. With a tool like Midjourney, a single creative professional can conceive, test, and develop ideas directly, without the immediate involvement of colleagues, designers, or other contributors. This accelerates the process but puts collective discussion and course correction under pressure.

The researchers describe generative AI in their publication as a “spirited technology”: a technology that does more than what is asked of it and adds unexpected elements. Hafermalz compares generative AI to a “spirited horse.” In her view, it moves creators forward at remarkable speed and can surprise them, but it also deviates from instructions and adds details the creator did not ask for. This can be frustrating, but it can also generate new ideas. That unpredictability is precisely what makes AI creatively interesting: Midjourney can introduce unexpected details, lighting effects, or compositions that set a creator on a new path. This only functions well, however, when creatives possess sufficient expertise to judge which outcomes are valuable and which need adjustment.

A further significant finding concerns what the researchers term “early aesthetic closure.” Because images generated by Midjourney often already appear polished and nearly finished, clients and commissioners are quicker to treat them as the expected final result. Jana Retkowsky of Erasmus University explains that striking AI images can lock in an idea too early. They help present a concept convincingly, but at the same time limit what remains possible afterwards.

The researchers emphasise that generative AI does not simply take over creative work. Behind a convincing AI image there is still human judgement: someone who recognises what works, what is off, and what is feasible in practice. That expertise risks becoming less visible, because from the outside it can appear as though the technology did the work. Hafermalz addresses a clear message to organisations: those who underestimate the maker behind the AI image risk undervaluing creative talent and making poor decisions as a result.