Three AI Agents — Expertise, Marketing, and SEO
Each agent checks content from its own perspective
Expert Agent
Checks facts and depth. 10 expertise criteria. Evaluates number accuracy, source relevance, and topic coverage completeness.
Marketer Agent
Evaluates persuasiveness and readability. 8 marketing criteria: headline strength, clarity of conclusions, calls to action.
SEO Agent
Checks GEO optimization. 12 SEO criteria: @block directives, Featured Snippets structure, keyword density, JSON-LD schema.
How the Three-Agent Debate Works
First Pass: Each Agent Independently
Three agents receive the article draft. Each outputs a rating based on their criteria. 30 criteria total.
Debate: Agents Discuss Disagreements
If ratings diverge >20% — agents exchange arguments. 2-3 rounds of debate until consensus.
Final Version: Average Rating ≥4.2/5
The article is considered ready when all agents give ≥4.0. The final version gets JSON-LD markup and @block directives. Time: 40 minutes for a 180KB+ article.
6 Quality Metrics for AI Content
How article quality is measured using the 7-stage methodology
Article Size
Bytes of HTML. 18-24 sections with tables, statistics, and JSON-LD.
Average Rating
Across 30 criteria from three agents. Minimum threshold: 4.0.
Evaluation Criteria
10 expertise + 8 marketing + 12 SEO/GEO.
Per Article
7 stages from prompt to final HTML with JSON-LD.
Time Savings
40 minutes vs 16 hours manually. + quality 4.2/5.
JSON-LD Types
Article, Person, Organization, FAQPage, HowTo — auto.
24 GEO Prompt Blocks: Structure
Each block solves its own task in AI optimization
Blocks 1-8: Preparation
Audience and competitor analysis. Keyword research with LSI. Entity map building for JSON-LD. Identifying the main Featured Snippet. Collecting 6-8 sources with data.
Blocks 9-16: Creation
Writing the answer-first block. Structuring H2/H3 for Featured Snippets. Comparison tables. 6-8 statistics with sources. E-E-A-T signals.
Blocks 17-24: Verification
Three-agent debate. Checking 30 criteria. JSON-LD validation. @block directive marking. Rating ≥4.2/5. Final output: .md with raw_html + full Schema.org.
Evolution of AI Content: From Prompt to Long-Read
5 key stages of methodology development
2022 — ChatGPT and First Prompts
ChatGPT launched. Content makers experiment with prompts. Quality is unstable: no structure, no fact-checking, no SEO optimization.
20222023 — Prompt Engineering
Structured prompts emerge: role, context, format. Quality improves but remains unstable. No fact-checking system. Risks of hallucination remain.
20232024 — Multi-Agent Systems
The idea of AI agents checking each other. Expert + Editor + SEO. The first 7-8 stage pipelines. Quality reaches 3.8/5.
20242025 — GEO Prompts and E-E-A-T
24 GEO blocks incorporating JSON-LD, @block directives, E-E-A-T signals. AI search engines (SearchGPT, Perplexity) require structured content. Quality reaches 4.1/5.
20252026 — GitHub CMS: 7 Stages + Debate + 24 Blocks
Full pipeline: 7 stages, 3 agents, 30 criteria, 24 GEO prompt blocks. Articles 180KB+ with full Schema.org. Quality stable at 4.2/5. 40 minutes per article.
2026Three-Agent Debate: How AI Checks AI
Expert Agent checks facts. Marketer Agent evaluates persuasiveness. SEO Agent checks GEO optimization. If ratings diverge >20% — agents exchange arguments. 2-3 debate rounds until consensus. Result: content rated 4.2/5.
evaluation criteria
debate rounds
4 Steps to Adopt the Methodology
From setup to the first 180KB+ article
Configure Agents
Define roles for Expert, Marketer, SEO. 30 criteria. 10 min
Run 7 Stages
Analysis → research → debate → structure → writing. 30 min
Verify ≥4.2/5
Three-agent debate. 2-3 rounds. Quality stable. 10 min
Output .md + JSON-LD
Article 180KB+ with raw_html. npm run build. 60 sec
Testimonials: 7-Stage Methodology in Real Projects
Results from using the three-agent debate
Alexey K.
CEO B2B Platform
"15 articles using the 7-stage methodology — average rating 4.3/5. Expert Agent found 12 errors. Marketer improved CTA — +23% conversion. The three-agent debate really works."
Marina S.
Tech Director EdTech
"The 24-block GEO prompt is a systematic approach. Before, ChatGPT produced unstable quality. Now 180KB+ articles with JSON-LD and @block directives."
Dmitry V.
DevOps Engineer
"40 minutes for a 180KB+ article is fantastic. The pipeline is automated: prompts → agents → .md → build. 96% time savings. Articles pass Google E-E-A-T check."
FAQ on AI Content
Common questions about the 7-stage methodology
Result: 180KB+ Articles with Full Schema.org
JSON-LD: 10+ types automatically
Article, Person, Organization, FAQPage, HowTo — from Frontmatter at build time.
E-E-A-T signals built in
author, certifications, sources — AI sees the source authority.
@block directives for Featured Snippets
answer-first, howto, faq — AI knows exactly what to cite.
6 Benefits of the 7-Stage Methodology
Why the three-agent debate produces better content
Quality 4.2/5 — 30 criteria
Triple check: facts, marketing, SEO.
40 min vs 16 hours — 96% savings
7 stages are automated.
24 blocks — full GEO coverage
From keywords to Featured Snippets.
180KB+ — long-reads that rank
18-24 sections with statistics and tables.
JSON-LD auto — 10+ types
Without plugins, without manual work.
Pipeline: prompts → agents → .md → build
From idea to production in 40 minutes.
Start Creating AI Content in 40 Minutes — Quality 4.2/5
Clone GitHub CMS — configure 3 agents — and every article will go through 7 verification stages. 24 GEO prompt blocks guarantee completeness and AI visibility.
Free · MIT License · 7 stages · 3 agents · 4.2/5 quality
Article from Section 3: Content / Markdown. Created using prompt template article-4.txt (HOME-4 style). AI Content — 7 stages and 24-block GEO prompt for 180KB+ articles.