What 166K Clicks Taught Me About Using AI for SEO

In a three-month experiment, a static website produced 166,000 Google clicks from 1.39 million impressions, with an 11.9% click-through rate and an average position of 7.5. However, this traffic peaked and then declined, showcasing that AI does not guarantee automatic, lasting growth. The key to success lies in employing AI within a quantifiable SEO workflow and reverting when data contradicts the initial findings.

The author's workflow consists of five steps. First, Keyword Planner is utilized for discovery, not final decisions. Broad product terms are exported, and search volume and competition are analyzed to identify possible clusters. However, a high-volume keyword does not automatically mean a new page should be created. The author asks five questions to determine if the keyword is a good fit: does it match their product, does its intent differ from their existing pages, does an existing URL already rank for it, can they create something genuinely useful for the searcher, and does it align with their SEO goals? This prevents the creation of thin pages for the sake of keyword research.

The second step involves checking keyword ownership using Google Search Console's Measurement Protocol (MCP). The data is pulled directly into the author's workflow, providing more valuable insights than just a keyword report. If an existing page already ranks well for a keyword, creating another page could lead to keyword cannibalization.

The author records their decision in five fields: query (target keyword), current URL (ranking page or none), evidence (clicks and position from a stable GSC window), intent (informational, local, comparison, or action), and decision (improve, create, or reject). If the intent already belongs to an existing page, the author focuses on improving that page instead of creating a new one.

The third step is using Claude's SEO-audit and keyword-research skills to review proposed changes. Claude can help identify duplicate keyword targets, check titles, descriptions, H1s, canonicals, indexability and internal links, thin or repetitive content, unsupported product and pricing claims, and possible performance regressions. However, Claude cannot replace human editorial judgment or evidence. The author rejects templated pages that only swap a keyword or city name.

Fourth, the author addresses issues flagged by Search Console instead of immediately rewriting the entire site when clicks fall. They separate recent Search Console data from settled data, decompose the loss by comparing equivalent date ranges, find which page lost clicks, pull the page's query-level changes, check its indexing, canonical and crawl status, review recent commits, and revert only when the evidence supports it. This approach prevents treating every ranking fluctuation as a technical emergency.

Finally, the author conducts an audit before and after deployment. Before making any SEO changes, they run the repository SEO audit. After deployment, they inspect the live HTML, ensuring proper titles, meta descriptions, canonicals, robots directives, H1s, and sitemap entries are in place. Only when the production page passes these checks does the author submit it through Search Console and begin its validation window.

The main takeaway is that AI was not asked to "write SEO content," but rather given access to structured evidence and an enforced release process: Keyword Planner → GSC ownership check → human intent decision → SEO audit → deploy → live verification → measure or revert. This process has helped the author's product website, Flingo, achieve better results with the help of AI.