Marco Patzelt
Back to Overview
February 9, 2026

What Is Agentic SEO? The Dev Guide, Not the $2k Platform

What is agentic SEO, really? Not the enterprise pitch—the actual engineering setup. Claude Code, Supabase, GSC. Real stack, real results, zero platform fees.

Search "agentic SEO" and you'll find two things: enterprise platforms charging $2,000/month and thought-leadership articles that explain the concept without showing a single implementation. Here's the developer version.

The Enterprise Definition vs. The Real One

Every SEO agency and their content team is writing about agentic SEO right now. Their definition: "AI agents that autonomously manage SEO tasks." Then they pitch you their platform.

Here's what agentic SEO actually is when you strip away the marketing: an AI agent with direct access to your database, your search console data, and your content—that can read performance data, identify problems, and fix them without you manually operating a CMS.

That's it. Not a dashboard. Not a SaaS product. Not a team of prompt engineers in a Slack channel. An AI agent connected to your infrastructure that operates on your data.

The "agentic" part means the agent acts autonomously. You don't prompt it for every task. You tell it "analyze my SEO and fix what's broken" and it reads GSC data, identifies CTR problems, rewrites meta titles, checks search intent alignment, and pushes changes to your database. You review the changes. The agent does the work.

Why Everyone's Talking About It Now

Three things converged in early 2026:

AI agents got good enough. Claude Code, specifically, can read files, execute commands, query APIs, and maintain context across a full project. Earlier AI tools could suggest changes. Agentic AI tools make changes.

SEO became data-heavy enough. Google Search Console gives you impressions, clicks, CTR, and position per query per page. That's thousands of data points on a blog with 30 articles. No human systematically analyzes all of that daily. An agent can.

The compounding math became obvious. Daily SEO optimization beats weekly. Weekly beats monthly. An agent that optimizes daily for 10 minutes outperforms a consultant who audits monthly for 3 hours. Not because it's smarter—because it's faster and more consistent.

The Two Versions of Agentic SEO

Version 1: Enterprise ($2,000+/month)

Platforms like Wordlift, Siteimprove, and seo.com are building "agentic SEO" features into their enterprise tools. These are dashboard products where AI agents run inside the platform, analyze your content, and suggest or implement changes through their interface.

The value prop: managed infrastructure, team collaboration, compliance features, integration with enterprise CMS platforms.

The reality: you're paying for the platform, not the agent. The AI underneath is the same models available to everyone. You're paying for the wrapper.

Version 2: DIY Developer Stack ($20/month)

My approach. I built the agentic SEO stack with four components:

  • Claude Code — the AI agent (requires Claude subscription)
  • Supabase — database-backed CMS with REST API
  • Google Search Console — performance data
  • Knowledge files — context about your voice, schema, and project

Total cost: the Claude subscription. Supabase has a free tier. GSC is free. The knowledge files are markdown documents you write once.

The tradeoff: you need to be technical enough to set up API connections and write knowledge files. There's no dashboard. There's no support team. It's your terminal, your agent, your data.

For solo developers and small teams, Version 2 is the obvious choice. You get the same agentic capabilities without the enterprise tax.

How Agentic SEO Actually Works (Step by Step)

Here's the concrete implementation, not the concept diagram:

1. The Agent Reads Your Performance Data

Claude Code connects to Google Search Console via API credentials stored as environment variables. It pulls impressions, clicks, CTR, and average position—per page, per query. This isn't a weekly export. The agent reads it live, every session.

2. The Agent Identifies Problems

With the data loaded, the agent cross-references against your published content in Supabase. It flags:

  • Bleeding pages: High impressions, low CTR. The page shows up in search results but nobody clicks. Usually a meta title or description problem.
  • Intent mismatches: Google shows the page for "setup guide" queries but the page is structured like a news article. The content doesn't match what the searcher wants.
  • Content gaps: Queries with significant impressions where you have no dedicated page. Google is trying to show you for these terms but has nothing relevant to serve.
  • Striking distance pages: Position 5-15 for valuable queries. Small optimizations can push these onto page 1.

3. The Agent Fixes Problems

This is where "agentic" differs from "assisted." An AI-assisted tool tells you what to fix. An agentic tool fixes it.

The agent rewrites meta titles to match the dominant search query. It adjusts descriptions to match search intent. It adds internal links to related content. It restructures articles to match the content format Google prefers for that query type. Then it pushes all changes directly to Supabase via the REST API.

Newsletter

Weekly insights on AI Architecture. No spam.

You review the changes. The agent did the analysis and the execution. Your job is quality control—making sure the tone is right and the technical details are accurate.

4. The Agent Creates New Content

For content gaps, the agent doesn't just flag the opportunity—it drafts the article. Using knowledge files that define your writing voice, database schema, and project context, it writes content that sounds like you, not like generic AI output.

The knowledge files are the secret weapon. I documented this in detail in my Claude Code architecture breakdown. Without them, you get ChatGPT-quality content. With them, you get content calibrated to a specific voice with specific expertise.

Agentic SEO vs. Traditional SEO vs. AI-Assisted SEO

Three distinct approaches:

Traditional SEOAI-Assisted SEOAgentic SEO
Data analysisManual (spreadsheets)AI suggests insightsAgent reads data autonomously
Content changesManual (CMS admin)AI suggests editsAgent pushes to database
Cycle timeWeekly/monthlyFaster, still manualDaily (10 minutes)
ScalingHire more peoplePrompt more oftenAgent handles more autonomously
CostTime-intensiveTool subscriptions$20/month + knowledge files
Expertise neededSEO knowledgeSEO + promptingSEO + engineering basics

The key insight: agentic SEO doesn't remove the need for SEO knowledge. It removes the manual labor of implementing SEO decisions. You still need to understand search intent, content quality, and what makes a good article. The agent handles the data crunching, the CMS updates, and the repetitive optimization work.

The Results You Can Expect

I'm going to be honest about what agentic SEO can and can't do.

What it does:

  • Turns daily SEO maintenance from 3 hours to 10 minutes
  • Catches optimization opportunities you'd miss manually (CTR drops, intent mismatches, content gaps)
  • Compounds results through daily iteration instead of weekly cycles
  • My numbers: 5 impressions/week → 68,000 impressions in 9 days after switching to this workflow

What it doesn't do:

  • Create expertise you don't have. If your articles have no original insight, agentic SEO won't fix that.
  • Generate search demand. It optimizes for queries that already exist. If nobody searches for your topic, agents can't change that.
  • Replace content review. The agent makes mistakes. Wrong tone, inaccurate details, bad takes. Every output needs human review.

How to Get Started

Step 1: Set up the infrastructure (1 afternoon)

  • Install Claude Code
  • Connect your CMS via API (Supabase, WordPress REST API, whatever you use)
  • Set up Google Search Console API access
  • Store credentials as environment variables

Step 2: Write knowledge files (1 weekend) This is the hard part and the most important part:

  • database_schema.md — your exact table structure
  • tone.md — your writing voice, rules, vocabulary
  • context.md — your background, expertise, what makes your content unique
  • internal_links.md — all published articles with URLs

Step 3: Start the daily routine (10 min/day) My daily Claude Code SEO workflow covers this step by step: morning GSC check, fix bleeding pages, identify content gaps, write or optimize.

Step 4: Monthly deep audit (optional) Use Agent Teams for a comprehensive audit—multiple AI specialists analyzing your content, data, intent, and competitors in parallel.

The Verdict

Agentic SEO is real. It's not a buzzword and it's not vaporware. It's an AI agent connected to your data that does SEO work autonomously.

The enterprise version costs $2,000/month. The developer version costs $20/month plus a weekend of writing knowledge files. Both use the same underlying AI models. The difference is the wrapper.

If you can set up an API connection and write markdown, you can run agentic SEO today. Start with the knowledge files. Connect Claude Code to your CMS and GSC. Run the daily 10-minute routine. The results compound from day one.

The only question is whether you're building the enterprise version or the developer version. I built the developer version. 68,000 impressions in 9 days later, I haven't looked at a single enterprise demo.

Newsletter

Weekly insights on AI Architecture

No spam. Unsubscribe anytime.

Frequently Asked Questions

An AI agent with direct access to your database, search console data, and content that autonomously reads performance data, identifies SEO problems, and fixes them. Not a dashboard or SaaS product—an agent connected to your infrastructure.

Enterprise platforms charge $2,000+/month. The DIY developer stack costs about $20/month (Claude subscription) plus free tools (Supabase free tier, Google Search Console). Same AI models, different wrapper.

AI-assisted SEO suggests changes that you implement manually. Agentic SEO acts autonomously—the agent reads data, identifies problems, fixes them, and pushes changes to your database. You review instead of execute.

Claude Code (AI agent), a database-backed CMS like Supabase or WordPress with API access, Google Search Console API credentials, and knowledge files (markdown documents defining your voice, schema, and context).

No. It removes the manual labor of implementing SEO decisions, not the need for SEO understanding. You still need to know search intent, content quality, and what makes good content. The agent handles data analysis and CMS updates.

Daily SEO maintenance drops from 3 hours to 10 minutes. Results compound through daily iteration. One example: 5 impressions/week to 68,000 impressions in 9 days. But it won't create expertise you lack or generate search demand that doesn't exist.

Technical infrastructure (API connections, Claude Code setup): one afternoon. Writing knowledge files (voice, schema, context, internal links): one weekend. The knowledge files are 80% of the setup effort and the most critical part.

Yes, if you have at least some Google Search Console data to work with. The agent needs real performance data to optimize against. Blogs with zero impressions should focus on creating content first, then switch to agentic optimization.

Let's
connect.

I am always open to exciting discussions about frontend architecture, performance, and modern web stacks.

Email me
Email me