Pretesting with AI-generated static and video ads

Artificial Intelligence (AI) makes it feasible to generate highly finished ads for pretesting, before final, human production. In this ‘proof of concept’ study, AI was used to create static and multi-shot video ads. When compared with the real ads they were based on, there were no significant differences observed in ad-effectiveness (recall, recognition, brand choice, brand attitude, and ad liking) under the specific conditions tested. These conditions included a low-involvement context in which several real and AI-ads were viewed, where all the AI-generated ads were carefully curated and professionally screened, and were about factual information rather than emotional experiences. Labelling ads as AI-generated may be compulsory soon, but this research showed that under these specific conditions, the presence of AI-disclosure, prior to exposure, made no difference. Funding: The AI-ad generation and data collection were funded by MediaScience. MediaScience also partly funds S.B.’s salary as the MediaScience Research Professor at the Ehrenberg-Bass Institute. Authors: Steven Bellman (Ehrenberg-Bass Institute, Adelaide University, Adelaide, SA, Australia), Sid Ali Ourradi (MediaScience, Austin, TX, USA), and Duane Varan. Steven Bellman is the corresponding author. Ethics and Conflicts: D.V. and S.A.O. work for MediaScience, which developed the software used in the study. However, S.B. is employed by an institute receiving funding from multiple companies, including MediaScience’s competitors. The institute remains independent, with no influence over data analysis or interpretation from sponsors. S.B. independently analyzed data and wrote the article, representing his own views. The study was published in Scientific Reports (2026) under a Creative Commons Attribution 4.0 International License.