The pipeline
Twelve stages. One keyword in. One ranking post out.
Every stage is resumable. If a fact-check fails on draft 47, the system retries from that stage — not from the keyword.
Keyword research
DataForSEO + live SERP scraping. Volume, CPC, difficulty, intent, PAA, autocomplete, related terms, top-10 results — all real, all fresh.
Semantic clustering
Embed every keyword with text-embedding-3-large. HDBSCAN clusters into topics. Each cluster proposes a primary keyword + supporting set + suggested angle.
Brief generation
Word count target, tone, entities, link targets, search intent, FAQs, citations to include — all derived from the live SERP, not guessed.
Section drafting
H2-by-H2 with shared context. Never one-shot 3000 words — that's how you get filler. Each section is written to its specific role in the brief.
Fact-check pass
Every quantitative claim is verified before it ships. Citations included. Hallucinations get filtered.
SEO pass
Entity coverage, density, readability, schema markup, meta title + description, image alt text, slug.
Humanization pass
Targets ~50% AI-detection score without losing meaning. Optional — skip it if you don't care about detector noise.
Image generation
Featured image (Flux), in-content images, Unsplash fallback. All sourced through R2 with proper attribution and alt text.
QA + plagiarism
Originality.ai plagiarism check, AI score, broken-link sweep. If anything fails, the article goes back for a re-pass instead of shipping broken.
Pre-flight site check
Refuses to publish if quality is below your site's tier. Established domains get a stricter gate than fresh ones.
Internal linking
Per-project pgvector index. New posts link into the graph. Older posts get updated to link forward to the new one. Anchor text is real.
Publish anywhere
Native publishers for WordPress, Shopify, Webflow, Ghost, Wix, Notion, Framer — and a generic webhook that connects to anything else.