# SEO Agent Workflow Guide

Use this as a practical checklist before trusting an AI agent with SEO work.

## 1. Start With A Skill

A skill is a repeatable method, not a loose prompt.

Every SEO skill should define:

- When to use it
- When not to use it
- Required inputs
- Step-by-step procedure
- Output format
- Common mistakes
- Verification checklist
- Examples of good output

Good first skills:

- Competitor page comparison
- Local SEO proof check
- Content gap review
- Internal-link opportunity finder
- Google Business Profile checklist
- Content brief builder

## 2. Use The Skill Manually First

Do not turn a weak skill into an agent.

Run the skill inside real work until you know:

- Which inputs are always missing
- Which steps the model skips
- Which examples improve output
- Which checks catch mistakes
- Which parts still need human judgment

## 3. Add Memory And Lessons

The agent needs durable memory.

Use memory for:

- Stable brand rules
- Service areas
- Client context
- Known tool quirks
- Safe defaults

Use lessons for:

- Mistakes Harman corrected
- Repeated failures
- Rules that prevent bad output
- Things the agent must never forget

## 4. Wrap Skills Into An Agent

A production SEO agent needs files around the skill:

- `AGENT.md` for purpose, scope, workflow, safety, and final output
- `MEMORY.md` for durable facts
- `LESSONS.md` for mistake prevention
- `run-log.csv` for each meaningful run
- `checklists/` for preflight, safety, and verification
- `templates/` for reports, briefs, and handoffs
- `rubrics/` for quality scoring
- `skills/` for task-specific procedures

If the agent has only a prompt, it is not production-ready.

## 5. Build The Verification Loop

Before anything ships, the agent should verify:

- Primary keyword and supporting keywords are present
- No page is cannibalizing another page
- Internal links work
- Claims are backed by proof
- Screenshots and examples match the claim
- Schema is valid
- The copy sounds human
- The final page answers the reader's real question

## 6. Keep Human Approval

The agent prepares the work. A person still approves:

- Which page should exist
- Which claim is safe to make
- Which screenshot is fair proof
- Whether the page sounds like the business
- Whether the final recommendation is worth acting on

That is the useful split: AI handles repeat prep; humans protect judgment.