AI SEO Content Production System

AI Workflow · Internal product for a digital agency · 2025 - 2026

Impact

~100x faster
From 2-3 working days per article to ~10 min. Enabled 1-3 articles/day per client at a few dollars in AI cost each
20+
Monthly recurring clients
60+
AI SEO projects delivered (team total)

Below: Search Console data from one client site over 12 months. Impressions grew from near-zero to 60K+ daily. The goal is content that appears in AI-generated search answers (Google AI Overviews, ChatGPT Browse), not just traditional Google rankings.

Google Search Console showing 4.22M impressions and 15.6K clicks over 12 months with clear upward trend

Problem

The agency wanted to offer AI-powered SEO content as a service. Each article required 2-3 working days of manual effort: topic research, competitor analysis, writing 1,200-2,000 words, sourcing banner images, formatting for WordPress, and publishing. At 1-3 articles per day per client, this volume was humanly impossible. The content team was the bottleneck, and adding headcount did not scale with the service price point.

The goal was speed and volume: get content out fast, track what gains traction in Search Console, then double down on high-performing topic clusters with manual optimization. The AI pipeline removed the production bottleneck so the team could focus on strategy instead of writing.

My Role

Three-person team: project lead (non-developer) who drove product direction, myself, and one colleague (both developers with PhD backgrounds). We co-built the entire system together. During prototyping, we each built independent Dify workflows, tested on our own company website, then merged the best parts into the production version. My focus area was hallucination prevention and content grounding.

There was also an advanced version with enhanced features (enhanced Q&A, additional CMS integrations to Shopify/Drupal/PayloadCMS) that I was not directly involved in building.

Key Decisions

Architecture

Scheduler (Python on Digital Ocean, production) |-- Reads article queue from Google Sheet |-- Manages status: pending > processing > created/failed |-- Calls Dify workflow API (SSE streaming) |-- Retry logic with dedup protection | v Dify Workflow Engine (multi-stage LLM orchestration) | |-- Stage 1: Input Processing | PDF URLs --> PDF-to-markdown tool | Web URLs --> URL-to-markdown tool | |-- Stage 2: Competitor Research (if no references provided) | LLM with online search --> find top-ranked articles | Fetch top result --> convert to markdown | |-- Stage 3: SEO Article Writing | Claude Sonnet via OpenRouter (reasoning + online mode) | Inputs: topic, brand voice, competitor content, internal links, CTA | Output: structured JSON (title, TOC, intro, body, conclusion, CTA) | |-- Stage 4: Infographic (optional, HTML only) | Claude Sonnet generates WordPress-compatible HTML infographic | |-- Stage 5: Image Pipeline | Claude Sonnet generates image prompt from article context | Image generation API --> banner image | GPT-4o generates SEO alt text | |-- Stage 6: Assembly + Validation | Combine all outputs into final JSON | HTML completeness check (unclosed tags) | v CMS Publishing |-- WordPress API (Application Password) |-- Shopify, Drupal, PayloadCMS (advanced version) Quality Evaluation |-- Sample ~20% of published articles |-- LLM-as-judge (different model) scores quality |-- Results feed back into prompt refinement

Transferable Pattern

Multi-stage LLM pipeline with human-in-the-loop quality control. Applicable to any business that needs AI-generated content at scale with brand consistency: product descriptions, marketing copy, customer communications, internal documentation.

Tech Stack

Dify Python Claude Sonnet GPT-4o OpenRouter Google Sheets API WordPress REST API Digital Ocean SSE Streaming