A Content System on Claude — the Working Version

An 11-step poster for turning Claude into a content-writing system crossed our desk, and here is the unusual news: it's mostly right. Nine of eleven steps map to real, current mechanisms under roughly their real names. That earns it the full rebuild rather than a debunk — verified, corrected where it drifts, and with the one thing the poster lacks attached to every step: how to check it's working.

01 — The genre verdictCredit where verification finds it

This series has documented invented commands, dead product walkthroughs, and jailbreak bait. Balance demands the opposite verdict when it's earned: this specimen's steps reference the real memory setting, real Projects, real extended thinking (under the invented name "adaptive thinking"), and real Artifacts. The verification method only means something because it can return good news — this is what that looks like.

02 — The working systemEleven steps, verified and tightened

Foundation — do once

1. Turn memory on. Settings → Personalization → Memory. Real setting, correctly located by the poster. VerifyAsk "what do you remember about my brand voice?" in a fresh chat — the answer tells you if memory is doing anything yet.
2. One Project per content line. A newsletter Project, a LinkedIn Project — not one bucket for everything. The poster's "keep it clean or Claude loses the plot" is folklore phrasing for a true thing: focused context beats mixed context. VerifyIf a draft cites the wrong audience, your Projects are bleeding into each other — split them.
3. Upload the style guide — with counterexamples. Your voice notes, three best posts, and crucially two bad ones labeled as such. The poster gets this exactly right: show good AND bad. VerifyAsk Claude to describe your voice back in one paragraph. If you wince, fix the guide, not the drafts.

Drafting — every piece

1. Frame with role–task–context–output. The poster's "RTCO structure" is a house name for the durable prompt shape (see I-10 — this pattern survives every model generation). VerifyIf output surprises you in kind (not quality), the task line was vague.
2. Think hard only where thinking pays. The poster's "adaptive thinking" is really extended thinking — and its advice is correct: use it for strategy and structure, skip it for captions. VerifyCompare one strategy task with and without. If you can't tell the difference, stop paying the latency.
3. Critique before you write. Ask for weak angles in the brief first, then the draft. Order matters; the poster has the order right. VerifyCount revision rounds per post for a week. Critique-first should cut them visibly.
4. Draft into Artifacts for anything multi-part — carousels, post sets, checklists — one prompt, clean copyable outputs. VerifyIf you're copy-pasting fragments out of chat prose, you skipped this step.

The loop — weekly

1. Repurpose winners only. Top post in, three formats out — carousel, short text, story — anchored to the original's audience pain points. Same idea, different format, no drift. VerifyRepurposed pieces should reference the source post's core line. If they wander, re-anchor.
2. Run the feedback pass. "What patterns did you notice in my feedback this week?" — then save the useful answers to memory. The poster calls the result a "taste map" that builds "automatically"; the honest version is that it compounds only if you actually run the pass. VerifyOnce a month, ask for your taste profile. It should be getting more specific, not staying generic.
3. Pre-publish gate: hook, value, CTA. A standing checklist prompt, scored separately, before anything goes live. Weak hook? Fix it. No CTA? Add one. VerifyThe gate is working when it occasionally kills a post. A gate that always says yes is decoration.
4. Date your system. The step the poster doesn't have: write the date on your setup notes. Product mechanics move; your system should be re-verified quarterly — this sheet carries its own check date for the same reason. VerifyIf any step names a button you can no longer find, the system is due its quarterly pass.

03 — The ledgerEleven claims, checked 2026-08-05