06-reference

write with ai ogilvy writing rules claude skill

2026-06-07·reference·source: Write With AI·by Dickie Bush & Nicolas Cole
writing-craftclaude-skillscritic-agentcopywritingogilvy

"I Gave Claude David Ogilvy's Writing Rules And Built A Legendary AI Writing Coach" — Dickie Bush & Nicolas Cole

Why this is in the vault

This is a working example of someone packaging a closed, well-documented rule set (Ogilvy's 1982 staff memo on writing) into a Claude Skill that does severity-graded critique rather than ghost-rewriting. The skill-architecture choices they describe (one-line overall verdict, every violation tagged Critical/Moderate/Minor with the offending quote + why + a specific fix, plus an explicit "what's working" pass so good lines don't get edited away) are a near-exact analog to the critic-agent pattern RDCO already runs. Keeping it as a reference data point for both the newsletter-craft library and the critic-skill design library.

⚠️ Sponsorship

No third-party paid sponsor. This is house/self-promo: the article is a lead-magnet funnel. The free "David Ogilvy AI Digital Writing Coach" skill download sits behind a Substack redirect link, and the close pitches a waitlist for a paid "Start Writing Online 5-Day Sprint" launching the next day. Footer links push the authors' other properties (Ship 30 For 30, Premium Ghostwriting Academy, Typeshare, Ghostbase). Treat the "free skill" framing as top-of-funnel acquisition for the paid Write With AI newsletter, not neutral teaching. The technique content (the 10 rules) is genuine and extractable, which is why it was processed rather than skipped.

The core argument

The piece restates Ogilvy's ten writing rules from his memo and then frames them as the training corpus for an audit skill. Paraphrased:

  1. Read the Roman-Raphaelson book on writing, multiple times.
  2. Write the way you talk; don't hunt for a separate "writing voice" — record yourself, transcribe, start there.
  3. Keep words, sentences, and paragraphs short; read aloud to catch wordiness.
  4. Avoid jargon; write as if to an eighth-grader.
  5. Cap any subject at two pages; if it won't fit, simplify.
  6. Verify your quotations.
  7. Don't send same-day; sleep on it, read aloud the next morning, then edit.
  8. For anything important, get a colleague to improve it.
  9. Make the desired reader action crystal clear before sending.
  10. If you need action, have the conversation — writing is not a substitute for talking to the person.

Their build claim: most writing feedback ("tighten it up") is too vague to improve a writer, while the principles that produce sharp business prose are well-documented but scattered. The skill consolidates that into a deterministic audit. Notably they keep it audit-only on purpose — it surfaces problems and the human still does the rewriting, so the rules get internalized instead of outsourced.

Mapping against Ray Data Co

Two direct connections.

  1. Critic-skill architecture. Their severity-graded output (overall verdict + Critical/Moderate/Minor findings with offending quote, reason, fix, plus a "what's working" preservation pass, and an explicit no-rewrite rule) is the same shape as RDCO's existing fresh-eyes critic gates — draft-review, design-critic, verify-pdf-output, voice-match. The "audit only, don't rewrite, so the operator internalizes the rule" stance reinforces the verification-as-independent-worker discipline: the critic's job is to surface, the author's job is to fix. Worth borrowing the explicit "what's working" preservation pass into draft-review if it isn't already there — protecting good lines from over-editing is a real failure mode in iterative review.
  2. Sanity Check voice rules. Several Ogilvy rules are already RDCO doctrine but cleanly stated: write the way you talk (matches the brief-iMessage / conversational-voice preference), short words and read-aloud (the tangible-over-abstract draft-review check), crystal-clear single CTA, and "99% of books should be blog posts" (the unbundling instinct). These are usable as a checklist layer inside draft-review for newsletter drafts. The corpus-to-skill move itself is also a content-as-product reminder: a tight, attributable rule set is the cheapest possible skill to build and the easiest to trust.

The build pattern (take a closed authoritative rule set, encode it as a graded-severity audit skill) generalizes beyond writing — it's the same recipe RDCO uses for the data-quality audit (audit-model) and could use for any domain with a canonical checklist.

Related