SEO SOP: Build an End-to-End “Publisher” Agent

Zak Ali, GM of Finder US, runs his entire content pipeline from the terminal via an AI “publisher” agent — freeing his team for the work AI can’t do. Here’s the SOP to build your own.

Objective

Stand up a single AI agent that takes a topic from keyword research all the way to a published CMS page, run from your terminal — then reinvest the reclaimed time into original research, video, and community.

Key Steps

1. Set up the terminal environment. Run an agent in the terminal (e.g. Claude Code) and connect an SEO-data MCP (Ahrefs / OpenSEO) plus your CMS credentials.
2. Build a “second brain.” In Obsidian markdown, capture your process, decision log, brand + voice profile, and ICP — the agent reads it every run and it keeps you model-agnostic.
3. Seed & research keywords. Start from a real conversation, then validate/expand via the MCP.
4. Build the customer profile + brief + outline.
5. Competitor + deep research. Gather well-cited supporting research.
6. Draft → human edit → internal linking. Keep a human on the brief and the edit.
7. Publish to the CMS from the terminal.
8. Add a judgment ledger. Log the agent’s decisions; close the loop when one misses so the system compounds.
9. Reinvest the time into YouTube, guest posts, and community.

Cautionary Notes

• Don’t drift into scaled-content abuse — reclaim time for quality & information gain, don’t spray pages.
• Keep a human on the strategy layer (the brief and the edit).
• Agents break in new ways a human delegate wouldn’t — document edge cases.

Tips for Efficiency

• Make it model-agnostic — context lives in the second brain, not one model.
• Repeatability is scalability — force the agent to log every step.
• Once one agent works, run several in parallel.

Sources & Relevant Episodes

Zak Ali, General Manager, Finder US — on the Unscripted SEO podcast. Full interview: Search Everywhere Optimization with Zak Ali. “You don’t even have to leave your terminal to do your entire job.”