HackerNews中文版

我在 LinkedIn 上进行了两年的冷启动。并非大规模操作——每天可能向潜在客户发送 10-15 条消息。问题从来都不是找不到要发消息的人。 LinkedIn 搜索功能运行良好。 Sales Navigator 也很好用。推荐也行得通。问题在于打开他们的个人资料，然后……发呆。 “好的，他们是一家 B 轮公司的工程副总裁。他们上周发帖讨论了技术债务。我怎样才能不听起来像他们收件箱里的其他人？” 删除。重写。检查是否太长。太短。太过推销。还是不够具体。四十七分钟后，我写了四条消息。真正的问题大多数外展工具都解决了错误的问题。它们提供：带有合并标签的模板（{{firstName}} 在 {{company}} 工作）违反 LinkedIn 服务条款的自动发送功能批量序列自动化但这并不是我卡住的地方。我知道要给谁发消息。我只是不知道该说什么，而且要花很长时间。 ChatGPT 也没帮上什么忙。除非我花 10 分钟根据他们的个人简介、我的产品详细信息和语气指南来精心设计完美的提示，否则它只会给我一些通用的垃圾信息。在那种情况下，我最好自己写。我构建的东西我制作了一个 Chrome 扩展程序，该程序： 1. 你只需设置一次——添加你的产品描述并为不同的声音创建“角色”（技术、咨询、休闲，等等） 2. 当你打开 LinkedIn 个人资料时，单击该扩展程序 3. 它会读取他们的个人资料，将信号与你的产品匹配，应用你的角色，并在大约 5 秒内生成一条消息 4. 你可以查看它，根据需要进行编辑，手动复制并粘贴到 LinkedIn 中没有自动化。没有自动发送。没有违反 LinkedIn 规定。只是更快地编写消息。技术方法该扩展程序会抓取可见的个人资料数据（没有 API，因为 LinkedIn 几年前就关闭了该功能）。将其发送到后端，该后端将：提取关键信号（角色、公司发展阶段、最近的活动）根据你的产品背景匹配这些信号应用特定于角色的提示工程返回一条带有严格规则的消息（没有流行语，没有“希望一切安好”，保持在 150 字以内）角色系统是最有趣的部分。你可以切换不同的声音，而不是使用一个通用的提示。 “前顾问”的角色听起来与“技术联合创始人”的角色不同——不同的句子结构，不同的参考点。我学到的东西最大的惊喜：人们不想要更多的自动化。他们希望保持控制，但消除空白页问题。每次我在早期的用户访谈中提到“自动发送”时，人们都会感到紧张。但“为我写，我来审查和发送”立即引起了人们的兴趣。第二个惊喜：质量标准比我预期的要高。一条消息 80% 很好，但有一行很奇怪？人们会重写整个消息。它需要达到 95% 的好，否则就没用了。目前状态目前处于私人测试阶段。大约有 40 人在使用它。平均消息生成时间为 6 秒。大多数人在发送前会编辑大约 20% 的消息。不确定这是否真的是一项业务，但它解决了我的问题。也许它也能解决你的问题。很乐意回答有关技术实施或我所见过的用户行为模式的问题。它被称为 Prospectee。如果你感兴趣，可以在 prospectee.io 上查看。

查看原文

I've been doing cold outreach on LinkedIn for two years. Not at scale — maybe 10-15 messages a day to potential customers.The problem was never finding people to message. LinkedIn search works fine. Sales Navigator exists. Referrals happen.The problem was opening their profile and... staring."Okay, they're a VP of Engineering at a Series B company. They posted about technical debt last week. How do I not sound like every other person in their inbox?"Delete. Rewrite. Check if it's too long. Too short. Too salesy. Not specific enough.Forty-seven minutes later, I'd written four messages.The real issueMost outreach tools solve the wrong problem. They give you:Templates with merge tags ({{firstName}} works at {{company}}) Auto-send features that violate LinkedIn's ToS Bulk sequence automationBut that's not where I was stuck. I knew WHO to message. I just couldn't figure out WHAT to say without it taking forever.ChatGPT didn't help much either. It would give me generic garbage unless I spent 10 minutes crafting the perfect prompt with their bio, my product details, and tone guidelines. At that point, I might as well write it myself.What I built I made a Chrome extension that:1. You set up once — add your product description and create "personas" for different voices (technical, consultative, casual, whatever) 2. When you open a LinkedIn profile, click the extension 3. It reads their profile, matches signals to your product, applies your persona, and generates a message in ~5 seconds 4. You review it, edit if needed, copy and paste it into LinkedIn manuallyNo automation. No auto-send. No LinkedIn violations. Just faster message writing.Technical approach The extension scrapes visible profile data (no API, since LinkedIn shut that down years ago). Sends it to a backend that:Extracts key signals (role, company stage, recent activity) Matches those against your product context Applies persona-specific prompt engineering Returns a message with strict rules (no buzzwords, no "I hope this finds you well", keep it under 150 words)The persona system was the interesting part. Instead of one generic prompt, you can switch between voices. A "former consultant" persona sounds different than a "technical cofounder" persona — different sentence structure, different reference points.What I learnedThe biggest surprise: people don't want more automation. They want to stay in control but eliminate the blank-page problem.Every time I mentioned "auto-send" in early user interviews, people got nervous. But "write it for me, I'll review and send" got immediate interest.Second surprise: the quality bar is higher than I expected. A message that's 80% good but has one weird line? People rewrite the whole thing. It needs to be 95% good or it's useless.Current stateIn private beta now. About 40 people using it. Average message generation time is 6 seconds. Most people edit about 20% of messages before sending.Not sure if this is a real business yet, but it solved my problem. Maybe it'll solve yours too.Happy to answer questions about the technical implementation or user behavior patterns I've seen.It's called Prospectee. You can check it out at prospectee.io if you're curious.

我开发了一个工具，不再让我在领英（LinkedIn）上浪费时间发开发信。