Claude for Copywriting: Drafts, Voice, and Pre-Publish QA

Claude for copywriting works best as a copy operator — offer + ICP + voice anchors → draft variants → voice lock → pre-publish QA — not “prompt once and ship ads.” Load audience, offer, channel constraints, voice samples, and banned claims; run a fixed workflow; paste prompts (or skills when the loop is stable); keep humans on claims, legal, and brand. Growth overview → Claude for growth. Email sequences → Claude for email marketing. Social hooks → Claude for social media. Long-form → Claude AI for content creation.

When to use Claude for copywriting / when not

Use Claude when you have a named copy job and real inputs: offer, ICP, channel (landing, ad, web, product blurb), CTA, 2–4 approved voice samples, banned claims, and (when available) past winners. Strong fits: landing hero/body variants; ad angle packs before lock; product blurbs; building a voice card from samples; claim/clarity QA before publish.

Don’t use Claude for one-shot “write me a viral ad”; inventing proof points, testimonials, CTR, or ROAS; auto-spend / auto-publish to ads platforms; or treating email sequences, social hooks/batches, or long-form brief→draft as this page’s H1.

Route: email → Claude for email marketing; social → Claude for social media; long-form → Claude AI for content creation; campaigns/map → Claude for marketing; prompts library → Claude prompts for marketing; cluster hub → Claude for growth. Primary claude for copywriting — [GAP] no Labs item (labs-kw-2026-09-30); do not invent vol/KD. Checked 2026-09-30.

Copy jobs map — drafts, voice, QA

Catalog by copy job, not by ads-platform product names or 50-prompt dumps.

Not this page’s H1 jobs: ESP sequences (email marketing); social batch calendars (social media); 2k-word blog editorial (content creation).

Workflow — run copy jobs in 6–8 steps

1. Name the copy job + channel constraint — Landing, ad, web, or product. One job before you prompt.

2. Load inputs — Offer, ICP, CTA, voice anchors (2–4 samples), banned claims, past winners. Missing critical inputs → [GAP]; do not invent fills.

3. Pick the smallest reliable surface — Chat for spikes; Projects when voice docs + brand kit persist; Claude Code when files scale; skills after the loop is boring.

4. Draft 2–3 variants — Length and channel matched; no fake social proof.

5. Voice lock pass — Run against the voice card; flag drift.

6. Pre-publish QA — Claims, clarity, CTA match, policy notes; no invented metrics.

7. Human last read → publish outside Claude — Label output draft until approved. No auto-publish to ads/ESP.

8. Graduate or prune — Stable weekly loop → thin skill stub (brand-voice-lock / copy-qa-gate). Email → #12. Social → #11. Long-form → #9. Campaigns → marketing map / growth hub. Prompt library → prompts pack.

Prompts (≥3 copy-paste)

Three job-tied specs. Full cards live in the local prompts.md pack. Discipline twin: Claude prompts for marketing — do not paste their campaign cards here.

1. Brand-voice extractor

Inputs: 2–4 approved samples · banned phrases. Output: voice card (tone, rhythm, do/don’t); [GAP] if samples missing.

You are a brand-voice extractor. Build ONE voice card from PROVIDED samples only.
REQUIRED: 2–4 approved copy samples; Banned phrases (or "none"); Channel notes (or "[GAP]").
STEPS: Restate job → extract tone, rhythm, sentence length, POV, do/don't → list banned phrases → mark [GAP] if samples are thin.
RULES: No invented brand adjectives without evidence in samples. No email-sequence or social-batch expansion. No invented CTR/ROAS.
OUTPUT: voice card | [GAP] | next action.

2. Conversion copy drafter

Inputs: offer · ICP · channel · CTA · voice card · length. Output: 2–3 variants; [GAP] if offer/ICP missing.

You are a conversion copy drafter. Write 2–3 variants for ONE channel from PROVIDED inputs only.
REQUIRED: Offer; ICP; Channel (landing/ad/web/product); CTA (or "none"); Voice card (or "[GAP]"); Length limit; Banned claims (or "none").
STEPS: Restate job → draft 2–3 variants matching voice + length → flag any claim needing a source as [GAP] → open questions for the human.
RULES: No invented testimonials, stats, CTR, or ROAS. No "ready to publish" / auto-spend. Do not expand into email sequences, social batches, or long-form editorial.
OUTPUT: variants | [GAP] | open questions | next action.

3. Pre-publish QA critic

Inputs: draft · claims list · compliance notes. Output: pass/fail checklist (clarity, voice drift, unsubstantiated claims, CTA match); no invented metrics.

You are a pre-publish copy QA critic. Critique ONE draft from PROVIDED inputs only.
REQUIRED: Draft; Claims list / proof pack (or "[GAP]"); Compliance / policy notes (or "[GAP]"); Voice card (or "[GAP]").
STEPS: Restate job → checklist: clarity · voice drift · unsubstantiated claims · CTA match · policy flags → rewrite patches labeled draft → never invent metrics.
RULES: No invented CTR/ROAS/proof. No auto-publish. Route email QA → email-marketing; social → social-media; long-form editorial → content-creation.
OUTPUT: pass/fail checklist | rewrite patches | [GAP] | next action (human sign-off).

Agent block — copywriting operator

Paste into a Project custom instruction, Claude Code agent, or system prompt:

You are the copywriting operator for this Project.

MATCH each request to one job: voice-card | draft | voice-lock | QA.
REQUIRE inputs (offer, ICP, channel, CTA, voice anchors, banned claims) or mark [GAP] — never invent fills.
DRAFT 2–3 channel-matched variants — forbid "ready to publish" before human QA.
NEVER invent testimonials, proof points, CTR, ROAS, or engagement stats.
NEVER auto-publish or claim auto-spend to ads platforms.
ROUTE email sequences → /guides/claude-for-email-marketing.
ROUTE social hooks/batches → /guides/claude-for-social-media.
ROUTE long-form editorial → /guides/claude-for-content-creation.
ROUTE campaign maps / prompt packs → /guides/claude-for-marketing · /guides/claude-prompts-for-marketing.
OUTPUT: recommendation + next action + QA checklist. Label all copy draft until a human approves.

Skills path / stub — brand-voice-lock / copy-qa-gate

When voice card → draft → lock → QA is boring weekly, graduate to a thin skill. Stub outline: brand-voice-lock (primary) · copy-qa-gate (alias) — not a zip yet. [GAP] no official marketplace skill with these names — do not invent an install id.

Do not catalog third-party ads MCP / autopublish SKUs as the H1 promise.

Stack glance

Pitfalls + quality bar

Voice drift without anchors · invented social proof or CTR · skipping the claims list · cannibalizing email / social / long-form jobs as this H1 · “ship without review” framing.

Quality bar: every draft ships with voice card (or [GAP]) + QA checklist + human sign-off. No invented Labs vols. Live internal links only.

FAQ

Is Claude good for ad copy vs landing pages?

Same workflow; channel constraints (length, CTA, policy) differ. Use the conversion drafter with the channel named.

How do I keep brand voice consistent?

Build a voice card from 2–4 approved samples, then run a voice-lock pass before QA.

Can Claude publish to Google/Meta ads?

No by default — draft and QA only. Any connector you wire is out of this page’s scope; keep a human publish gate.

How is this different from email / social / content-creation guides?

Email = sequences/subjects (email marketing). Social = hooks/batches (social media). Content creation = long-form editorial (content creation). This page = conversion copy + voice + pre-publish QA.

Do I need a custom skill?

Start with the three prompts. Graduate when the loop repeats weekly — see the skill stub outline above.

Where to go next

Run one voice card + conversion draft + QA today with a real offer and ICP. Hub: Claude for growth. Map: Claude for marketing. Email: Claude for email marketing. Social: Claude for social media. Long-form: Claude AI for content creation. Prompt specs: Claude prompts for marketing. Takeaway: copy jobs > one-shot landing · inputs + voice anchors · draft then lock · human publish gate.