PosterMate: Audience-driven Collaborative Persona Agents for Poster Design

Paper · arXiv 2507.18572 · Published July 24, 2025
Personas PersonalityCo Writing CollaborationDesign Frameworks

PosterMate gathers feedback from each persona agent regarding poster components, and stimulates discussion with the help of a moderator to reach a conclusion. These agreed-upon edits can then be directly integrated into the poster design. Through our user study (𝑁 = 12), we identified the potential of PosterMate to capture overlooked viewpoints, while serving as an effective prototyping tool. Additionally, our controlled online evaluation (𝑁 = 100) revealed that the feedback from an individual persona agent is appropriate given its persona identity, and the discussion effectively synthesizes the different persona agents’ perspectives.

3.1 How can we effectively establish target audience roles using real-world materials from the existing workflow? To develop effective persona agents that enrich the poster design from the target audiences’ perspectives, it is critical to establish each target audience’s role. Prior work suggested that LLMs can be prompted to take on various backgrounds as a form of personas to simulate their behaviors [53]. In this process, each persona agent needs a defined identity, which encompasses a unique set of background, to create varying roles and perspectives on the target of the simulation—in this case, simulating target audiences of the marketing campaign. These elements can be formulated in the form of seed memory using persona description [53, 54]. However, manually crafting personas and using them as seed inputs for agents is labor-intensive and often demands thorough user research [45]. One potential solution to address this is to use marketing briefs (also known as design or creative briefs) as guidance. Marketing briefs are structured text documents that serve as the foundation for creative work [13] and are widely used in the commercial marketing sector to develop playbooks that generate actionable insights [3]. In general, they provide a broad description of the audience that they are targeting, along with an overview of a marketing campaign’s goals and marketing details (e.g., message to deliver, constraints) [58]. A concrete example is illustrated in Figure 13, which outlines the details of an oral care brand’s campaign—containing sections like the problem statements, target audiences, and goals (i.e., driving increased market share among young customers), which can serve as the foundation for the agents’ persona. However, since these descriptions are providing a highlevel overview of the target users and are not intended for crafting highly detailed profiles like personas, these audience summaries need to be refined and structured into personas that capture diverse audience segments and provide meaningful feedback.

we conducted a semi-structured, think-aloud interview study with designers. Participants were given materials that would serve as the foundation for building the agents intended to support design. They then engaged in a think-aloud process, providing insights on the end-to-end design procedure—from constructing these agents using the provided materials to the overall design process.

3.5 Design Goals

Following our exploration of prior works and our formative interview study, we set the design goals of a persona-driven collaborative assistant for poster design as follows:

DG1 Generate diverse persona agents constructed from research on real-world audience data (i.e., marketing brief) by leveraging its key attributes to enhance audience diversity, enabling the simulation of real-time audience feedback within the design tool.

DG2 Enable agents to provide contextualized feedback on poster components (i.e., text, image, consolidated design theme) within the canvas, while offering both an inspirational insight and an actionable suggestion (i.e., actual modification to be applied) for feedback.

DG3 Support discussions among the persona agents and the user to reconcile conflicting design feedback through articulation and by reaching a conclusion that maximizes satisfaction among all agents concerned.