提示词库
每条提示词只解决一件事，明确说明需要你提供什么，并被要求在数据不可得时明说，而不是给一个估算值。运行之前先读示例产出。
提示词库以英文发布。提示词本身可以要求模型用任何语言作答。
Rewrite a page passage by passage so an assistant can quote a correct, self-contained answer from it, plus a straight read on whether page edits alone can change anything.
Edit an AI first draft into publishable prose by cutting throat-clearing and machine rhythm, and flagging every claim that needs a fact instead of inventing one.
Write an X vs Y comparison page from dated, observed competitor facts, including the cases where the competitor is the better choice and a ledger of what could not be verified.
Sort an existing set of pages into keep, update, consolidate or retire, with a reason, a next action, and a named destination for every URL that comes down.
Decide what to change on a page that has lost or never gained traffic - one diagnosis, the change list it implies, and an explicit list of what to leave alone.
Draft an FAQ section from questions readers have actually been recorded asking, plus FAQPage JSON-LD whose answers match the visible text word for word.
Produce an internal link plan for one page: anchor text, the sentence each link belongs in, and a flag for any pages that overlap so heavily they should be merged instead.
Draft landing page copy that answers one search query in the first screen and then earns one action, using only the facts you supply.
Sort a keyword list by what the searcher wants and what page type each implies, reading intent from the results that currently rank and marking it unverified where nobody looked.
Turn a topic and its search intent into a section-by-section outline where every heading states what it proves, with unequal word budgets and an explicit cut list.
Turn a content brief into a full first draft that marks where first-hand evidence is needed instead of inventing statistics, customer names, or study citations.
Turn one target query into the brief a writer can work from - the page's argument, the questions it must answer, the evidence each section needs, and what is out of scope.
Group a raw keyword list into topic clusters you can actually build pages around, with one page intent per cluster.
Read the pages actually ranking for a query, get the requirements a page must meet to belong there, and a straight verdict on whether your site should target it at all.
Write title tags and meta descriptions for a batch of pages, checked for length, uniqueness across the batch, and accuracy to what each page actually contains.
Turn a subject into a hierarchy of pillar and supporting pages with internal links, an explicit out-of-scope list, and a page count that fits what your team can publish.
关于本库
不需要。本页所有提示词都按发布状态直接可读、可复制。只有当你希望由 GenGrowth 的 Agent 代跑这项工作时，账号才有意义。
提示词是你粘贴给模型、换回一份产出的一条指令；Skill 是一套 Agent 反复执行的完整流程，通常把好几条提示词的工作量当作其中的步骤。提示词面向一次任务，Skill 面向一条持续运转的闭环。
可以，而且应该改。变量标出的是每次都会变的部分；执行步骤和质量检查是值得保留的部分——正是它们在阻止模型编造数据。
当底层做法发生变化，或者某次真实运行暴露出提示词本应拦住的问题时，我们会修订它。每页都标注了最后更新时间。
GenGrowth