It was 11pm on a Sunday in May when Iris, the ex-AFFiNE COO and Forbes Asia 30 Under 30 honoree, found themselves staring at Google Search Console for the seventeenth time that week. They were watching impressions tick up by single digits while their blog sat crippled after a domain migration wiped out four years of hard-earned domain authority. Something had to change—and it did. Thirty days later, their site went from roughly 600 monthly Google impressions to approximately 32,000. That's a 53x multiplier, achieved not through some viral content play but by handing the boring, daily work of SEO to an AI agent running on Claude Code with a custom skill installed.

The Setup: Starting From Zero

The blog in question—gingiris.tools, which covers growth playbooks for AI startups—was forced into a domain migration after a hosting incident. The move was clean technically, but the damage was brutal: 60+ solid articles, zero inbound links on the new domain, and no bandwidth to do daily SEO work manually because Iris was running another product simultaneously. Rather than grinding through the tedious checklist themselves, they decided to write an SOP—not for a human, but for an agent. The assumption: anything that couldn't be written down in concrete steps wasn't automatable. Everything else? Let the machine handle it.

How It Works: 25 Steps, Every Morning

The workflow runs each morning like clockwork. First, it queries Google Search Console API for yesterday's data and DataForSEO for SERP positions on tracked keywords. Then comes the diagnosis: detecting deltas in rankings, identifying which articles need internal links versus refreshes versus Schema markup, picking today's top three actions ranked by impact versus difficulty, executing git commits to the Jekyll repo, pushing changes (or queueing them if rate-limited), generating a daily report with page-1 keyword counts and GA4 metrics, and finally submitting URL Inspection requests for modified pages. Most days, Iris reads that morning report over coffee, replies 'looks good, continue' or 'skip action B, do C instead,' and that's it. The agent doesn't need motivation. It doesn't get bored. It just executes.

Week 1: Foundation Work (600 → 2,100 Impressions)

The first week was almost entirely technical cleanup—and it's where the real gains started. The agent discovered 31 issues that had been sitting unfixed for weeks: five articles with broken canonical_url tags, twelve missing hreflang annotations, eight articles with Schema.org typos, four zombie pages returning 404 but still in the sitemap, and two robots.txt entries blocking AI crawlers. Fixing these technical issues doesn't sound like a growth story, but three-quarters of that initial jump came entirely from making content actually crawlable. The articles were already good—they just weren't accessible to search engines.

Week 2: Schema + GEO (2,100 → 8,400 Impressions)

This is where Generative Engine Optimization started moving the needle. The agent added FAQPage Schema to twelve tutorial articles, HowTo Schema to eight step-by-step guides, Article Schema with author and datePublished across every post, and Organization plus WebSite markup to the homepage. These changes don't directly boost Google rankings—but they make content machine-readable enough that AI engines feel confident citing it. Three of the FAQ-tagged pages became eligible for featured snippets. One tutorial on open-source marketing started appearing in Google's AI Overview. Perplexity began surfacing articles when queried about Product Hunt tactics. The entire Schema batch took roughly 90 minutes of agent time with zero human involvement.

Week 3: Internal Linking + CTR Rewrites (8,400 → 18,200 Impressions)

The third week diagnosis was blunt: nine top-ranking articles had zero internal links pointing to any landing page. Traffic was arriving and bouncing immediately. The agent rewrote 23 article titles using a five-element CTR rubric—number, year, brackets, social proof, and 50-60 character length—and inserted internal links from those nine articles to the three most relevant conversion pages. It also identified a cannibalization issue where three articles competed for the same keyword and proposed which to canonicalize. Two got approved; both saw immediate position gains. Iris admits they had a Notion doc full of title rewrites they'd been meaning to do since February. The agent completed all 23 in a single day.

Week 4: Compounding Returns (18,200 → 32,000 Impressions)

The final week was quieter but more rewarding. One article hit position #3 for 'github star growth playbook'—a keyword with roughly 1.3K monthly searches and low competition, generating approximately 800 additional daily impressions from that single ranking alone. Google AI Overview cited the open-source marketing guide for a high-intent query, driving further qualified traffic. By day 30, Iris realized they hadn't written a single SEO-related git commit message themselves in nine days. Everything had been handled by the agent, signed under its own identity with natural commit messages indistinguishable from human-written ones.

Key Takeaways

  • The boring daily discipline of SEO is easier to automate than the strategic decisions—once you stop trying to make agents strategic and let them be operationally relentless, everything clicks into place
  • Schema.org markup remains wildly undervalued for AI search; adding FAQPage schema to twelve pages took 90 minutes but unlocked featured snippets and AI Overview citations worth far more in traffic
  • The daily report is the real productivity hack—seeing 'how many keywords on Google page 1 today?' every morning creates accountability that willpower alone can't sustain
  • The SOP itself becomes the competitive moat; without documented procedures, agents improvise and produce inconsistent results; with them, execution stays reliable and repeatable

The Bottom Line

This isn't about gaming Google's algorithm—it's about removing yourself from the loop on work that doesn't require human judgment. Iris went from 25 hours per week of SEO labor to three, and traffic grew 53x because the machine finally handled what humans keep procrastinating: the tedious daily discipline that actually moves rankings. The full SOP is now open-sourced as a Claude Code skill. If you've been ignoring your site's technical debt or sitting on a backlog of optimizations you keep meaning to get to, this is your sign to automate it—or at least audit what's actually crawlable right now.