AI Newsletter Writer — Turn Your Weekly Notes Into a Published Newsletter in 8 Minutes
Built by QuantumGlint ·
About this build
A Claude powered workflow that takes your raw weekly notes, bookmarks, thoughts and ideas dumped into a single document and transforms them into a fully written, formatted and ready to publish newsletter — complete with intro hook, main sections, a curated links section and a closing call to action — all in your own vo
What was built
I had been wanting to write a weekly newsletter for about 18 months. Started and stopped 4 times. The problem was never having things to say — I read constantly and have opinions about everything. The problem was the activation energy of turning raw thoughts into something publishable felt too high on a Friday afternoon. This build fixed that entirely. Here is exactly how the workflow runs: Step 1 — Note capture throughout the week Throughout the week I dump everything into a single Notion page. Links I found interesting, thoughts I had, things that frustrated me, things I learned, observations about my industry. No structure required. Just a running brain dump. This takes zero extra effort because I was already doing this kind of thinking — I just was not capturing it anywhere useful. Step 2 — Weekly trigger via n8n Every Friday at 2pm n8n reads the current week's Notion page and passes the raw content to Claude. Step 3 — Voice calibration and newsletter generation via Claude The Claude prompt does something most newsletter AI tools skip — it starts with a voice calibration step before writing anything. The prompt includes 3 examples of previous editions I wrote manually plus a description of my tone — direct, slightly opinionated, no corporate speak, occasional dry humour. Claude uses these as style anchors before generating anything new. The newsletter gets written in six sections: Opening hook — one paragraph that starts with the most interesting or provocative thing from the week's notes. Not a summary of what is coming. A hook that makes someone want to keep reading. Main section one — the biggest idea or observation from the week, expanded into 3 to 4 paragraphs with my take on why it matters. Main section two — a secondary topic or counterpoint to the main section. This is where the newsletter gets interesting rather than one-note. Quick takes — 3 to 5 bullet points of shorter observations that did not fit a full section. These are the most shareable part of every edition. Links worth your time — 3 curated links from the week's bookmarks with a one sentence note on why each one is worth reading. Not a dump of everything I saved — Claude selects the 3 most relevant to the newsletter's theme. Closing — one paragraph that either connects back to the opening or sets up something to think about before next week. Ends with a question or a call to action. Step 4 — Delivery as draft The completed newsletter gets sent to my Gmail as a draft and optionally pushed directly to Beehiiv as an unpublished draft ready to review. I spend about 8 minutes reading through it, making small edits where my voice drifted, and hitting publish. What I changed after the first 4 editions: The opening hook was too safe in early versions — Claude defaulted to a summary opener even when instructed otherwise. Fixed by adding an explicit example of a bad opener and a good opener directly in the prompt so it had a concrete contrast to work from. Also added an instruction that the newsletter should have exactly one moment where a reader might disagree with something. Without this instruction every edition was too agreeable and forgettable. Results after 14 editions: Open rate sitting at 43 percent which is above the industry average for my category. Three editions have been shared by readers without me asking. The consistency of publishing every week has done more for the audience than any individual edition — something only possible because the workflow removed the weekly friction completely. The voice calibration step is the thing I would not remove under any circumstances. Without it every edition sounds slightly off in a way that is hard to identify. With it the output reads like something I actually wrote. Happy to share the full Claude prompt and n8n workflow in the comments for anyone who wants to run it.