The three machines
- Brief library. Notion database where every brief has the keyword, search intent, target word count, target reader, and the H2 outline.
- Drafter. Claude with a system prompt tuned to your brand voice and a banned phrase list.
- QC layer. A second Claude pass that reviews the draft against a checklist before a human touches it.
Step by step setup
Build the brief database in Notion
Properties: Slug, Primary keyword, Search intent, Target word count, Status, Author, Publish date, Hero diagram. Status moves from Brief, Drafted, QC, Edited, Published.
Wire the brief to Claude
Use Claude Projects with the brand voice doc and three best in class examples attached. The brief becomes the user message. Claude returns a draft plus a short rationale on key choices.
Run a QC pass
Same Project, different prompt. Ask Claude to review the draft against a named checklist. Output is a list of issues, not a rewrite. Humans address the issues.
Human edit and publish
Editor reviews, applies cuts, fixes claims, ships. Average edit time on a 1500 word post is 30 to 60 minutes if the QC layer worked.
The QC checklist
- Primary keyword in H1 and meta title.
- Three banned phrases checked, none present.
- FAQ section with at least three Q and A pairs.
- One comparison table or one diagram.
- No unattributed statistics. Every claim has a source.
- Reading level Grade 8 or 9 by an automated check.
- One clear answer in the first 60 words.
Throughput math
With this setup a single editor plus a part time strategist can ship 8 to 16 posts per month at editor ready quality. Without AI, the same team would ship 4 to 6 with significantly more burnout. The unlock is the QC layer that catches drift before the human reviewer.
Quick answers
- Why Claude over ChatGPT for the drafting layer?
- Claude tends to follow constraint lists more strictly and writes more naturally on long form. ChatGPT is better for short copy and structured data. For SEO long form, Claude wins on average.
- Can I run this with GPT 4 instead?
- Yes. Same architecture. GPT 4 family handles the drafter and QC roles. Tune the system prompt for the model.
- What about MDX or Markdown output?
- Tell the drafter to output Markdown with proper heading hierarchy. Most CMS importers handle Markdown cleanly. For Notion, paste the Markdown into a code block, then convert.
- Will Google penalize AI content?
- Google penalizes low quality content regardless of how it was made. AI content that follows the QC checklist and ships through editor review has not seen blanket penalties through 2025 or 2026.
- Should the editor be a human or AI?
- Human. Always. The editor is who keeps the brand voice consistent and catches the claims that need fact checking. AI is the drafter and the QC pass, not the editor of record.