Proxona: Supporting Creators' Sensemaking and Ideation with LLM-Powered Audience Personas

TL;DR

Proxona tackles the gap between quantitative audience analytics and the nuanced motivations driving creator success by converting comments into data-driven audience personas organized as dimensions and values. Using an LLM-powered pipeline, clustering, and retrieval-augmented chat, it enables creators to converse with simulated audience segments, test storyline ideas, and gather plausible feedback during early content development. Technical evaluations show high relevance and low hallucination rates for generated personas, while a user study with 11 creators indicates improved sensemaking, diverse audience discovery, and more confident, audience-informed ideation. The work demonstrates a practical, co-creative framework for integrating LLM-based personas into the creative workflow, with implications for scalable audience targeting and early-stage content strategy.

Abstract

A content creator's success depends on understanding their audience, but existing tools fail to provide in-depth insights and actionable feedback necessary for effectively targeting their audience. We present Proxona, an LLM-powered system that transforms static audience comments into interactive, multi-dimensional personas, allowing creators to engage with them to gain insights, gather simulated feedback, and refine content. Proxona distills audience traits from comments, into dimensions (categories) and values (attributes), then clusters them into interactive personas representing audience segments. Technical evaluations show that Proxona generates diverse dimensions and values, enabling the creation of personas that sufficiently reflect the audience and support data grounded conversation. User evaluation with 11 creators confirmed that Proxona helped creators discover hidden audiences, gain persona-informed insights on early-stage content, and allowed them to confidently employ strategies when iteratively creating storylines. Proxona introduces a novel creator-audience interaction framework and fosters a persona-driven, co-creative process.

Figures (10)

• Figure 1: Exploration page where creators can explore and interact with audience personas. (A) Persona list showing multiple audience personas. (B) Persona cards with details of persona profile. (C) Conversation space to freely chat with personas. Creators can ask questions to personas and request their opinions. (D) Dimensions and values list showing diverse audience attributes. (E) Customization options for creating new personas by selecting or generating different dimension-value combinations.
• Figure 2: Creation page where creators write a video storyline, proceed with conversations about their plot and request feedback on their written content. (A) Text editor where creators can draft their video storylines. (B) Conversation space, where creators ask personas for thoughts on the storyline. (C) Feedback feature allows creators to get [evaluation] or [suggestions] on particular sections of the text from a particular persona. (D) Example feedback provided by a chosen persona (Diane).
• Figure 3: Our pipeline generates audience personas with GPT-4 and k-means clustering. Our pipeline first builds audience summaries and transcript summaries and constructs a dimension & values list with GPT-4. With comments, the pipeline predicts the audience characteristics of each comment based on the dimension & values list. Using pre-trained BERT embeddings and k-means clustering, the 200 comments are clustered in k groups of predicted audiences. In the end, our pipeline generates a persona profile that consists of a job, a short persona description, and recent experiences of generated personas with GPT-4.
• Figure 4: For interacting with personas, our pipeline retrieves relevant video transcripts and context data using LangChain and RetrievalQA, enabling the generation of context-aware responses. It also provides evaluative critiques or actionable suggestions, ensuring that the feedback is relevant and grounded in real insights.
• Figure 5: User study procedure including tasks using the Docs and Proxona, surveys, and post-interview. The study lasted for approximately 2 hours.