reddit · SideProject · 09-15 13:50 55 分 ← 返回列表

I scanned 10,000 pages for generic AI writing tells. Sites over 200 pages are performing the worst for it.

Hi everyone, I am the owner of SiteTell . A little bit of context around it so the rest of the post makes sense. It is a SaaS tool that scans your whole website and checks to see if it sounds like AI wrote it in a generic way. I made it because I was getting frustrated with seeing the same AI phrasings, structures, tells etc... everywhere and thought maybe there's a solution here. It uses a curated ruleset to check against and return all the tells. So far we've scanned 103 sites, 10,126 page

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：AI 批量生成的内容因句式与结构高度同质化（套话、冗余结论段等），被 Google 判定为无价值内容而失去 SEO 排名，站长需要一个可以全站扫描并定位这些「通用 AI 写作痕迹」的工具。

目标用户：SEO 内容站站长、内容营销/代运营团队、用 LLM 批量产文的独立开发者与 SEO 代理商

证据：I made it because I was getting frustrated with seeing the same AI phrasings, structures, tells etc... everywhere and thought maybe there's a solution here. / So far we've scanned 103 sites, 10,126 pages and over 23,000 flagged instances. / 58% of items flagged so far have been structural issues. The most prevalent, on 66% of sites, is a redundant closing paragraph.

变现：订阅制 SaaS：免费扫描少量页面，按站点数/月度扫描页数分级收费（如 $19/$49/$99）；也可提供一次性全站审计报告 + 修改建议的付费包，或做成 WordPress/CMS 插件按站收费。

MVP：2 周内做一个极简版：输入域名 → 抓取 sitemap 与主要页面 → 用正则 + 规则集检测四类痕迹（冗余结尾段、模板化开头、公式化过渡句、空泛总结句）→ 输出全站评分与问题页面清单，免费看 1 页，付费解锁全站报告 CSV。抓取用现成的 Firecrawl/Playwright，检测层纯规则无需训练模型。

竞品：Originality.ai、GPTZero、Copyleaks AI Detector、Surfer SEO AI Detector、getSiteTell（原帖作者的同类产品本身）

摘要：作者用自研 SaaS 工具 SiteTell 扫描了 103 个网站、10,126 个页面，标记出 23,000+ 处「通用 AI 写作痕迹」，其中 58% 属于结构性问题，66% 的站点存在冗余结论段。其核心论点是 Google 并不惩罚 AI 写作本身，而是惩罚可被复制到百处的同质化无价值内容。帖子目前只展示使用量与检测数据，没有任何付费用户或收入数据，因此付费意愿仍需验证。