AI agents are slowly but surely becoming a reality, so the digital marketing industry needs to rethink its assumptions about what constitutes attractive website design and advertising. That’s because emotional appeals won’t work on the models.
A new preprint paper titled "Are AI Agents Interacting with Online Ads?" says that when augmented with browsing capabilities or multimodal interfaces, AI agents see websites and digital ads differently from people.
According to the study’s authors, Professor with the Digital Media Lab at the University of Applied Sciences Upper Austria, Andreas Stӧckl, and researcher Joel Nitu, the models ignore visual cues and emotional appeals, focusing on structured data like price, availability, and specifications.
They analyzed interaction patterns, click behavior, and decision-making strategies through experiments with multimodal language models such as OpenAI GPT-4o, Anthropic Claude, and Google Gemini 2.0 Flash.
After the agents were directed to perform hotel search and booking tasks autonomously, the researchers quickly saw that the models interacted with websites in a very machine-like way. For instance, they didn’t react to image-based banners showing a call-to-action, making the paid promotion essentially useless.
“Our findings reveal that AI agents neither ignore nor systematically avoid advertisements but instead favor certain features, particularly keywords and structured data,” says the paper.
“These insights have significant implications for the future design of advertising strategies in AI-dominated digital environments.”
The implications of the findings are not insignificant for both publishers and advertisers. Clearly, ad personalization will have to be adapted to suit the AI agents in terms of design and placement, Stöckl told The Register.
“Although models will engage with visual ads, they generally favor textual elements that align with the user’s query,” explained the paper. In other words, AI agents don’t browse – they execute commands.
Your email address will not be published. Required fields are markedmarked