Autonomous SEO agent — $11/year in AI costs

SEO is the perfect job for an agent: high-value, recurring, rule-driven, and mind-numbing for a human. This one runs on a VPS via cron and handles two production sites — lamose.com and oxone.ca — end to end.

The weekly loop

• Audit (Sundays) — pulls Google Search Console data (clicks, impressions, positions, CTR), crawls both sites for technical issues: broken links, slow pages, missing tags, thin product descriptions
• Optimize — rewrites meta titles, descriptions, and image alt text, and injects JSON-LD structured data directly through the Shopify Admin API
• Publish — writes three blog posts a week targeting keyword gaps found in the GSC data
• Report — emails me a summary of what changed and what moved

No dashboard to check, no tickets to file. The site just gets incrementally better every week.

The $560 → $11 swap

The agent originally ran on Claude. The work — metadata rewrites, alt text, structured blog drafts from keyword briefs — turned out to be well within a much cheaper model’s ability, so I swapped the brain to DeepSeek. Annual AI cost dropped from ~$560 to ~$11 with no quality regression I could measure in rankings or click-through.

That swap taught me a principle I now apply everywhere: match the model to the task, not the task to your favorite model. Frontier models for judgment and strategy; commodity models for high-volume production work. Total system cost including the VPS is $30–40/month — less than one hour of an SEO consultant’s time.