Ask HN: 各位如何检查 ChatGPT 对你们产品的评价？
1 分•作者: gissurthor•5 个月前
我一直在测试 AI 模型对 ChatGPT、Claude、Perplexity 和 Gemini 等平台中关于 SaaS 产品的买家式提问的反应。
有几件事让我感到惊讶：
* 一个在 GitHub 上有 3.1 万颗星的产品，ChatGPT 基本上从未在其核心用例中提及。
* 一家资金充足的公司，在其所属类别的一般购买查询中毫无存在感。
* 产品被用竞争对手的功能集来描述，因为模型似乎了解该类别，但对该产品本身却知之甚少。
我一直注意到的最大问题是，大多数创始人从未真正检查过这些模型在买家对话中对他们的评价。
所以我很好奇这里其他人是如何处理这个问题的：
* 你们是手动测试吗？
* 你们有一套可重复使用的提示语吗？
* 发布新内容后，你们是否看到任何实际的改变？
* 你们是否找到了任何可靠的方法来判断模型出错的原因，而不仅仅是注意到它出错了？
查看原文
I’ve been testing how AI models respond to buyer-style questions about SaaS products across ChatGPT Claude Perplexity and Gemini<p>A few things surprised me<p>A product with 31K GitHub stars that ChatGPT basically never surfaced for its core use case<p>A well-funded company with zero presence on generic buying queries in its category<p>Products being described using a competitor’s feature set because the model seemed to know the category but not the product distinctly<p>The biggest thing I keep noticing is that most founders never actually check what these models say about them in buyer conversations<p>So I’m curious how other people here are approaching this<p>Do you test it manually
Do you have a repeatable prompt set
Have you seen anything actually move after publishing new content
And have you found any reliable way to tell why a model is getting your product wrong instead of just noticing that it is