wp-auto-seo-pipeline

# WordPress Auto SEO Pipeline Architecture ## Overview This skill defines the SEOSONA standard for building automated content generation pipelines (like plugins or external orchestrators) that target WordPress or any CMS. It takes the strong software-engineering concepts from traditional auto-spinners but strips out the fatal SEO flaws that cause "Helpful Content Update" (HCU) penalties. ## 1. Core Pipeline (The Good) A robust automated system MUST separate generation from post-processing to avoid LLM hallucinations (like broken JSON-LD or invalid HTML). **Phase 1: Profile & Context Assembly** - Extract Search Intent, Target Keyword, LSI/Semantic Keywords. - Load Brand Guidelines (Tone, Restrictions, CTAs). **Phase 2: LLM Generation (Multi-Model Approach)** - Do not lock into a single model (e.g., Vertex AI only). - Use **OpenAI o1** or **Gemini 1.5 Pro** for Data Extraction and Structuring. - Use **Claude 3.5 Sonnet** for the actual prose and Copywriting (most natural cadence). **Phase 3: Post-Processing (Deterministic)** - **Schema (JSON-LD)**: Inject schema programmatically via a script using the LLM's structured output variables. Do not ask the LLM to write raw JSON-LD inline with the article text. - **HTML Sanitization**: Run a strict regex/HTML parser to fix broken tags `<h2>`, `<ul>`. - **Background Worker**: Execute via Queue (Redis/BullMQ or WP-Cron) to prevent HTTP 504 timeouts. --- ## 2. E-E-A-T & Quality Hardening (The Fixes) Traditional auto-spinners fail because they generate "Me-Too Content" (regurgitating TOP 3). You MUST implement the following nodes in your pipeline: ### 2.1 Information Gap Analysis (Anti "Me-Too" Content) Before drafting the outline, the system must scrape the Top 3 competitors and run a Gap Analysis: - *What questions did the Top 3 fail to answer?* - *What unique angle can our Brand provide?* The LLM must be explicitly instructed to include these "Information Gains" rather than just copying the competitor's heading structure. ### 2.2 SME Data Injection (Experience & Expertise) Do not generate YMYL (Your Money or Your Life) content purely from internet scraped data. - **RAG Requirement**: The Pipeline must accept an input vector (PDFs, internal data, expert interview transcripts). - The LLM prompt must enforce: *"Cite the provided internal expert data exactly. Do not invent statistics."* ### 2.3 Semantic Internal Linking (Anti-Spam) Do not use "Exact Match Keyword" auto-linking plugins. - **Method**: Use Vector Embeddings to find semantically related published posts. - **Anchor Text**: Use an LLM to select a natural, 3-5 word phrase to hyperlink, avoiding exact-match over-optimization penalties. ### 2.4 Media Policy - **Avoid AI Slop**: Do not auto-generate featured images with DALL-E/Midjourney for YMYL content. It reduces trust. - **Method**: Pull from a curated, brand-approved Media Library or use premium API stock integration (with real human subjects).