EZ SEOS
Documentation menu

The SEO agent

Grounded SEO agent with Do Next, project memory, editable skills, and full tool reach across your project data.

The SEO agent is a chat interface with live access to your project's tools. It uses your site and project data without inventing metrics it did not pull. When you ask what to work on, it leads with the Do Next visibility queue, then opportunities, CTR, and GEO tools as needed. It keeps per-project memory, follows editable skills, and can be the same capability surface you reach from Claude Code or Codex via MCP. Marketing overview: SEO agent (/features/seo-agent).

What can the agent do?

It can reach a broad suite of platform tools for the project you are in. That is not a claim that every tool loads into every turn. Grouped by area:

  • Keywords and SERP
  • Rank tracking and history
  • Domain, competitors, and Search Truth
  • Backlinks and prospects
  • Audits, originality, and page content
  • Visibility Engine: Do Next, Site Changes, brand lookup, prompt explorer, citation loss, access, 404s, Proof
  • Opportunities lanes
  • CTR and exposure
  • Local (including GBP Q&A through the agent, not a full UI panel)
  • Analytics, GSC, and IndexNow
  • Page Glance
  • Reports, digests, and export
  • Memory and skills

What can you ask it?

Examples that match how people actually talk to it:

  • What should I fix next for SEO and AI visibility?
  • Why don't third-party estimates match Search Console?
  • Where am I losing AI citations?
  • What does Do Next want me to ship this week?
  • How is this client looking in the prompt explorer?

What does it lead with when you ask what to work on?

Do Next first. That is the visibility queue. Then opportunities, CTR gaps, and GEO actions as the question needs them. Not opportunities alone as the whole prioritization story. Depth: GEO (/docs/geo), opportunities (/docs/opportunities), CTR gaps (/docs/ctr-gaps).

Does it make up numbers?

No. It does not invent volume, difficulty, traffic, rankings, or backlinks. When analytics are under a sampling threshold, you get a warning instead of a fake certainty.

Memory and skills

Memory is persistent per project. Forget on request. Project context loads with the project so the agent is not guessing which site you mean. Skills ship pre-seeded and editable. Add your own. Enable, disable, or reset. External clients and installable skills: MCP (/docs/mcp).

How does its judgment stay grounded?

It checks redirects and non-canonicals before it hands out optimize advice. Recent edits get a lag rule (about 2 to 4 weeks) so it does not judge too early. Diagnostics can say do nothing when that is the honest answer.

Approvals and outputs

Paid scans need human approve. Answers come back in Markdown you can copy or download.

Same tools outside the app

MCP connects Claude Code, Codex, or any MCP client to the same project-scoped tools so clients do not mix. Full brief: MCP (/docs/mcp). Also useful nearby: credits (/docs/credits), projects (/docs/projects), features overview (/docs/features), FAQ (/docs/faq).