[논문 리뷰] PersonaCite: VoC-Grounded Interviewable Agentic Synthetic AI Personas for Verifiable User and Design Research
PersonaCite는 AI 페르소나를 실제 고객의 목소리 산출물에 바탕을 두고, 대화 중에 증거를 검색하며, 확인 가능한 디자인 연구를 위해 출처가 명시되고 회피를 인지한 응답을 제공하는 에이전틱 시스템이다.
LLM-based and agent-based synthetic personas are increasingly used in design and product decision-making, yet prior work shows that prompt-based personas often produce persuasive but unverifiable responses that obscure their evidentiary basis. We present PersonaCite, an agentic system that reframes AI personas as evidence-bounded research instruments through retrieval-augmented interaction. Unlike prior approaches that rely on prompt-based roleplaying, PersonaCite retrieves actual voice-of-customer artifacts during each conversation turn, constrains responses to retrieved evidence, explicitly abstains when evidence is missing, and provides response-level source attribution. Through semi-structured interviews and deployment study with 14 industry experts, we identify preliminary findings on perceived benefits, validity concerns, and design tensions, and propose Persona Provenance Cards as a documentation pattern for responsible AI persona use in human-centered design workflows.
연구 동기 및 목표
- Address the unreliability and unverifiability of prompt-based AI personas by grounding personas in empirical VoC data.
- Enable real-time retrieval of voice-of-customer artifacts during conversations to constrain and verify responses.
- Introduce mechanisms for explicit abstention when evidence is insufficient and for response-level source attribution.
- Demonstrate the approach through formative evaluation with industry experts and outline a documentation pattern for responsible AI persona use.
제안 방법
- Retrieval-augmented interaction that pulls VoC artifacts from multimodal data sources for each conversation turn.
- Constraint mechanism that limits LLM responses to only what retrieved evidence supports.
- Explicit abstention when evidence is missing, avoiding speculative claims.
- Response-level source attribution linking claims to underlying VoC artifacts.
- Agentic context engineering to provide the right evidence as context and mitigate brevity bias.
- Two interaction modes: persona interviews and reaction simulation with design stimuli.

실험 결과
연구 질문
- RQ1How can AI personas be grounded in real VoC data to improve verifiability and trust in interactive design research scenarios?
- RQ2What design mechanisms (abstention, source attribution, provenance documentation) improve transparency and reliability of AI personas in practice?
- RQ3Does retrieval-augmented grounding support faster and more responsible design exploration compared with traditional methods?
- RQ4What documentation patterns (Persona Provenance Cards) aid responsible deployment of AI personas in human-centered workflows?
주요 결과
- Grounding through real-time evidence retrieval and response constraints shifts persona grounding from creation-time to interaction-time.
- Abstention and source attribution improve perceived trustworthiness and support verification of persona claims.
- Reaction simulation enables rapid design exploration and hypothesis testing while complementing direct user engagement.
- Participants value transparency about data provenance and expect confidence scores and citations linking claims to VoC data.
- Grounded personas are seen as exploratory tools that complement, not replace, traditional user research.

이 논문 읽기
누빈트 뷰어에서 원문을 열고, 읽으면서 궁금한 곳을 바로 물어보세요.
누빈트에서 이 논문 읽기이 리뷰는 AI가 논문 원문을 바탕으로 만들었습니다. 인용하기 전에 원문을 확인해 주세요.