coefficient-portfolio

Content Engine: Multi-Agent SEO Blog Generation Pipeline

A five-agent AI pipeline that produces SEO/AEO-optimized comparison blog posts at scale. The system handles planning, competitive research, writing, editorial review, and publishing — generating 17 long-form comparison articles in a single day.

Architecture

Agent 1: Planner & Orchestrator
    → Determines blog format and positioning strategy
    → Creates comprehensive checklist for each phase
    ↓
Agent 2: Researcher
    → Competitive intelligence and proof point gathering
    → G2 ratings, reviews, customer testimonials
    → Screenshots via Browserbase
    ↓
Agent 3: Writer
    → 3,500-4,500 word blog composition
    → Integrates proof points naturally
    → Follows dynamic positioning strategy
    ↓
Agent 4: Editor
    → 6-phase quality review framework
    → Format, tone, positioning, proof points, balance
    → Publish or revise decision
    ↓ (if revise → back to Agent 3)
Agent 5: Publisher
    → Google Docs creation with professional formatting
    → Hyperlinks, image placeholders, sharing permissions

Pipeline Stages

Stage 1: Planning (Agent 1)

• Determines blog format:
  • Format 1: “Us vs Competitor” (direct comparison)
  • Format 2: “Competitor A vs Competitor B” (third-party positioning)
• Analyzes competitor focus (marketing/finance/operations) to set dynamic positioning
• Creates detailed checklist for research, writing, and review phases

Stage 2: Research (Agent 2)

• Researches competitor products, features, pricing, positioning
• Collects G2 ratings, review excerpts, customer testimonials
• Mines own website for case studies, templates, social proof
• Captures screenshots using Browserbase MCP
• Documents all proof points with sources
• Maintains balanced perspective (competitor strengths AND limitations)

Stage 3: Writing (Agent 3)

• Writes complete 3,500-4,500 word blog posts
• Maintains conversational, flowing tone (not choppy/bullet-point)
• Integrates proof points naturally throughout
• Applies correct positioning based on competitor type
• Varied sentence structure: 20% short, 50% medium, 30% long

Stage 4: Editorial Review (Agent 4)

Six-phase quality framework:

1. Format compliance — word count, headings, markdown structure
2. Tone & readability — conversational test, flow check, context richness
3. Criticism addressal — verifies 6 specific quality standards are met
4. Positioning accuracy — competitor type alignment, dynamic strategy applied
5. Proof point verification — sources, testimonials, G2 ratings present
6. Balance & credibility — fair competitor acknowledgment (3+ times)

Stage 5: Publishing (Agent 5)

• Creates Google Doc with blog title as filename
• Applies professional formatting (headings, bold, lists, links)
• Inserts image placeholders with descriptions
• Adds all hyperlinks on first mention of key terms
• Sets sharing permissions (anyone with link can comment)

Dynamic Positioning Strategy

Agent 1 determines competitor focus and positions accordingly:

Competitor Focus	Positioning Strategy
Marketing-focused	“Matches marketing capabilities PLUS cross-functional”
Finance-focused	Lead with finance superiority and cross-functional advantage
Operations-focused	Emphasize operational capabilities first
General-purpose	Cross-functional breadth as primary differentiator

Quality Standards (Addressing 6 Original Draft Issues)

1. NO choppy sentences — Varied structure (1-4 short, 5-10 medium, 15+ long per section)
2. Rich context — Explain WHY differences matter, not just WHAT they are
3. Conversational tone — Colleague over coffee, not corporate brochure
4. Balanced perspective — Acknowledge competitor strengths genuinely (3+ times)
5. Integrated proof points — Customer testimonials, G2 ratings, specific data throughout
6. Finance positioning — Finance systems/use cases mentioned FIRST, not last

Blog Formats

Format 1: Direct Comparison

• Introduction
• What Sets Them Apart
• Head-to-Head: Integrations, Features, UX, Pricing, Support, Security
• The Verdict

Format 2: Third-Party Comparison

• Introduction
• Competitor 1 Overview
• Competitor 2 Overview
• Head-to-Head: Data Sources, Destinations, Features, UX, Pricing, Performance
• Limitations Both Share
• Why Our Product Is Better (positioned as the alternative)
• The Verdict

Content Specifications

Spec	Requirement
Word count	3,500-4,500 words
Structure	H1 title, H2 sections, H3 subsections
Proof points	3-5 customer testimonials per post
G2 data	Specific ratings and review excerpts
Screenshots	5-9 images with alt text and captions
Links	First mention of key terms hyperlinked
Tone	Conversational, balanced, context-rich

Results

• 17 blogs completed (exceeded 15-blog target)
• ~62,000-76,000 total words across all posts
• 1-day production timeline (Oct 9-10, 2025)
• All quality gates passed through editorial review
• All published as Google Docs with shareable links
• 85-153+ screenshots planned across all posts
• 50+ named customer testimonials integrated

Stack

• Agents: Claude Code multi-agent (5 agents, sequential with revision loop)
• Research: Firecrawl MCP, Exa MCP (competitive intelligence)
• Screenshots: Browserbase MCP (browser automation)
• Publishing: Google Docs MCP (document creation + formatting)
• Data: Google Sheets MCP (authentication and data management)

Key Decisions

1. Sequential execution, one blog at a time — User approval after first blog before proceeding. Catches quality issues early instead of generating 17 posts that all need rework.
2. Dynamic positioning over static templates — Competitor type determines how we position ourselves. Marketing competitors get a different angle than finance competitors.
3. Editor as hard gate with revision loop — Agent 4 can send back to Agent 3 for revisions. No post publishes without passing all 6 quality phases.
4. Criticism-driven quality standards — The 6 quality criteria came from analyzing failures in original draft attempts. Each criterion addresses a specific, documented problem.
5. Knowledge base as shared context — Writing guidelines, messaging docs, and format templates ensure consistency across all 17 posts without re-prompting.

Visual Direction

Diagram: Five-stage pipeline with Agent 1-5 as sequential nodes. Show the revision loop between Agent 3 (Writer) and Agent 4 (Editor). Side panel shows the two blog formats as template thumbnails. Bottom section shows the 6 quality criteria as a checklist with pass/fail indicators. Overlay shows 17 blogs flowing through the pipeline with completion status.