Thematic jokes are central to stand-up comedy, sitcoms, and public speaking, where contexts and punchlines rely on fresh material - news, anecdotes, and cultural references that resonate with the audience. Recent advances in Large Language Models (LLMs) have enabled interactive joke generation through conversational interfaces. Although LLMs enable interactive joke generation, ordinary conversational interfaces seldom give creators enough agency, control, or timely access to such source material for constructing context and punchlines. We designed Jokeasy, a search-enabled prototype system that integrates a dual-role LLM agent acting as both a material scout and a prototype writer to support human-AI collaboration in thematic joke writing. Jokeasy provides a visual canvas in which retrieved web content is organized into editable inspiration blocks and developed through a multistage workflow. A qualitative study with 13 hobbyists and 5 expert participants (including professional comedians and HCI/AI specialists) showed that weaving real-time web material into this structured workflow enriches ideation and preserves author agency, while also revealing needs for finer search control, tighter chat-canvas integration, and more flexible visual editing. These insights refine our understanding of AI-assisted humour writing and guide future creative-writing tools.

翻译：主题笑话是单口喜剧、情景喜剧和公开演讲的核心，其语境和笑点依赖于与观众产生共鸣的新鲜素材——新闻、轶事和文化参考。大型语言模型（LLM）的最新进展使得通过对话界面进行交互式笑话生成成为可能。尽管LLM能够实现交互式笑话生成，但普通的对话界面很少能为创作者提供足够的自主权、控制力或及时获取此类原始素材以构建语境和笑点。我们设计了Jokeasy，一个具备搜索功能的原型系统，它集成了一个双重角色的LLM智能体，既作为素材侦察员，又作为原型写手，以支持主题笑话写作中的人机协作。Jokeasy提供了一个可视化画布，其中检索到的网络内容被组织成可编辑的灵感模块，并通过多阶段工作流进行开发。一项涉及13名爱好者和5名专家参与者（包括专业喜剧演员及人机交互/人工智能专家）的定性研究表明，将实时网络素材融入这一结构化工作流能够丰富构思过程并保持作者自主权，同时也揭示了对更精细的搜索控制、更紧密的聊天-画布集成以及更灵活的可视化编辑的需求。这些见解深化了我们对AI辅助幽默写作的理解，并为未来的创意写作工具提供了指导。