Original Summary

Small SaaS, a few months old, single-digit clicks from Google per week. Like most founders I'd open Search Console once a month, see a line going roughly up, and close the tab. A while ago I started using Claude to write blog posts. It occurred to me that the agent writing my SEO content had never once seen whether the last post got a click. So I gave it the data: a Search Console export, my web analytics, and keyword volumes. I asked one question: "what should I work on?" What it found in one session: 1. A quarter of my impressions were one useless query. My impressions had doubled and I was happy about it. 25% came from a single search at position 30 with zero clicks. It was someone else's brand name. I now track impressions with that query excluded, because otherwise the chart lies to me. 2. I'd optimized a post for a keyword with 10 searches a month. I'd spent an evening rewriting it. Meanwhile, Google was already testing one of my landing pages for a cluster of terms worth ~720/month, and the page title didn't target any of them. 3. 13 new pages had zero impressions after a week. The cause: every URL in my sitemap had the same "last modified" date, because it was set to the build time. Every deploy told Google every page had changed, so Google stopped trusting the field. 4. One of my three Google visitors that month signed up. Tiny numbers, but the takeaway was that search traffic matching what you sell converts, and ChatGPT was already sending more visitors than Google. None of this was visible on the Search Console overview chart. All of it was sitting in the tables. The part I didn't expect Doing this by pasting CSV exports got old, so I built a small integration that lets the agent pull Search Console data directly, as part of the blogging tool I'm working on. While taking screenshots of it, I noticed the "top queries" list was in alphabetical order. The query with a quarter of all my impressions wasn't even in it. Google's API sorts results by clicks. On a young site almost everything has 0 clicks, so it's all a tie, and the ties came back alphabetically. I kept the first 50 rows, which meant the first 50 alphabetically, not the 50 that mattered. The agent had been getting the same list. It was confidently "analyzing my SEO" from an alphabetical sample, with no idea anything was missing. Lesson for anyone building AI features on top of an API: whatever you cut from the data is invisible to the model. It won't ask for what it can't see. If you want to try this yourself You don't need any tool for it. Export Search Console (last 28 days, queries + pages), paste it into Claude or ChatGPT, and ask: Which pages have lots of impressions but a CTR well below others at a similar position? Suggest better titles. Which queries do I rank 8–20 for? What's missing from the page that ranks? Which queries land on my homepage or an unrelated post? Group them

  • 情报分类：商业与市场研究
  • 分类依据：内容涉及商业、投资或市场动态
  • 信息来源：Reddit · SideProject
  • 发布时间：2026/10/17 05:18:53