Skip to content
Back to Resources
ClaudeCopywritingBeginnerGuide

Claude CoWork for Copywriters

A practical guide to using Claude as your AI co-worker for client voice mapping, multi-variant drafting, and approval workflows — from setup to daily use.

Claude CoWork for Copywriters

What is Claude CoWork?

Claude CoWork is the practice of using Claude as a persistent, context-loaded collaborator inside your copywriting workflow — not a content generator you paste prompts into and hope for the best. Copywriter postings are down 34% year over year. Autonomous tools like Ralph Wiggum Marketer exist specifically to cut you out of the loop. The question is not whether AI will write copy — it already does. The question is whether your judgment, client relationships, and voice control make you irreplaceable.

Claude-native prompts. The prompts in this guide use Claude's XML tag structure (<context>, <instructions>, <format>, <avoid>) for precise, consistent output. These tags reduce ambiguity and improve reproducibility. They work in other models but are optimized for Claude.

The copywriters who stay employed are the ones clients trust to be in the room — because they own the voice doc, manage the approval chain, and know when a Cialdini framing is wrong for a Stage 1 awareness audience. Claude handles throughput. You handle judgment. This guide shows you how to set up that division of labor from day one.

This guide covers setup, five workflows that protect and extend your client value, and the anti-patterns that will make you look like you are just running a wrapper.

Install the Copywriter Plugin

This guide works on three Claude surfaces. The plugin is the fastest path on two of them. Pick whichever you use:

If you're on Cowork (desktop or mobile app)

Claude Cowork is Anthropic's agentic workspace — Claude completes work autonomously and returns finished deliverables. The Copywriter plugin packages the workflows below as native skills and slash commands.

  1. Open the Cowork plugin directory in your desktop app.
  2. Filter by Cowork, search for "Copywriter", and click Install.
  3. The plugin's slash commands and ambient skills are now available in any Cowork task.

If you don't see the plugin in the directory yet, install via custom marketplace: paste https://github.com/alexclowe/awesome-claude-cowork-plugins in your Cowork plugin settings.

If you're on Claude Code (CLI)

Install from your terminal:

claude plugin add alexclowe/awesome-claude-cowork-plugins/copywriter

The plugin's slash commands and skills load on next session.

If you're on Claude.ai (web chat only)

Plugins aren't directly installable on the web chat surface. You have two options:

  1. Use the prompts in this guide directly in a Claude Project (covered in the next section). Same outputs, more typing.
  2. Upload the plugin's skills as a zip via Settings → Features → Custom Skills (Pro/Max/Team/Enterprise plans). Higher friction; only worth it if you want the auto-activating skills, not the slash commands.

What the plugin gives you (any surface)

Slash command What it does
/brief-intake Capture client tone, brand voice, target audience, success metrics, and constraints into a structured brief
/draft-batch Generate 3–5 variations of copy with different angles, lengths, and tones for client review
/tone-check Analyze copy against the client's brand voice guide and flag deviations with phrase-level evidence
/approval-workflow Track client feedback and version history, generate diff summaries, prepare sign-off package

Auto-activating skills (no command needed — Claude applies them when relevant):

  • Client Voice Mapping — Extract and replicate tone from sample copy into a reusable Voice Doc across projects
  • Conversion Psychology — Persuasion hooks, urgency framing, objection handling, CTA optimization grounded in Cialdini and Sutherland

The plugin works standalone for one-off tasks. Pair it with the surface-specific setup below for persistent context across every task — that combination is the full Claude CoWork setup.

Setting Up Claude for Copywriting Work

Surface note: The Project setup below is for claude.ai web users. Cowork users have their own task-context mechanism (set context once when starting a Cowork task). Claude Code users get the plugin's ambient skills automatically — no Project setup needed. The workflows themselves are surface-agnostic — paste the prompts wherever you're working. Step 1: Create a Client Project for each client. In Claude, go to Projects and create one per client — not one master project for all clients. Voice bleeds if you mix them.

Step 2: Set your custom instructions. In Project settings, add:

You are my copywriting assistant for [Client Name]. Here is my context:

<client-profile>
- Client: [Name, industry, product/service]
- Audience: [Primary persona — role, pain, awareness stage per Schwartz]
- Voice: [3-5 adjectives with brief definitions, e.g., "Direct: no jargon, no hedging"]
- Tone ranges: [e.g., formal for compliance docs, conversational for social]
- Banned words/phrases: [list them]
- Regulatory context: [FTC Endorsement Guides / FCA / SEC / FINRA / HIPAA / none]
</client-profile>

<rules>
- All copy must match the voice profile above — flag any draft that drifts
- Required disclosures must appear in every applicable piece (list them here)
- Never fabricate testimonials, statistics, or endorsements
- Output is always a draft — label every piece "DRAFT — PENDING CLIENT REVIEW"
</rules>

Step 3: Upload the Voice Doc. Once you complete Workflow 1 below, upload the resulting Voice Doc to this Project. Claude will reference it for every piece going forward.

Step 4: Upload reference samples. Add 3–5 approved pieces of client copy that represent the voice at its best. Claude uses these as calibration anchors, not templates.

Step 5: Pin a session opener. Start every session with: "We're working on [Client]. Confirm the Voice Doc is loaded and state the awareness stage we're writing for today." This prevents generic output and forces Claude to confirm context before generating anything.

Five High-Leverage Workflows

1. Capture a Client's Voice into a Reusable Voice Doc

A Voice Doc is a structured reference that defines a client's voice with enough precision that Claude can apply it reliably — and that you can hand off to any collaborator, human or AI.

<context>
Client: [Name and brief description]. I have: [intake transcript / brand guidelines PDF / existing approved copy samples — describe what you're uploading]. The client described their voice as: [paste their own words if available].
</context>

<instructions>
Analyze the materials and produce a Voice Doc with these sections: Brand Positioning (1 sentence), Voice Adjectives (5 terms, each with a 1-sentence definition and a do/don't example), Tone Ranges (formal vs. casual — when each applies), Sentence-level rules (length, rhythm, punctuation preferences), Vocabulary (preferred terms, banned terms, industry jargon to use/avoid), Awareness stage default (using Schwartz's five stages — where does the primary audience usually enter?).
</instructions>

<format>
Markdown. Each section under an H2. Rules as bullet points. Do/don't examples inline under each adjective.
</format>

<avoid>
Generic descriptors like "friendly" or "professional" without concrete definition. More than 600 words. Inventing characteristics not supported by the source materials.
</avoid>

2. Generate Angle Variations for One Brief

A single brief can be approached from multiple angles — different hooks, awareness stages, and psychological framings. This workflow produces 3–5 distinct variations so you and the client can choose direction before you invest in a full draft.

<context>
Brief: [Paste the brief or summarize: product, offer, CTA, channel, word count]. Target audience awareness stage (Schwartz): [Unaware / Problem Aware / Solution Aware / Product Aware / Most Aware]. Voice Doc: [reference uploaded doc or paste key rules].
</context>

<instructions>
Write 3–5 angle variations. Each variation gets: a one-line angle label (e.g., "Authority + Social Proof" or "Stage 2 Problem Agitation"), a headline, and an opening paragraph only — not a full draft. Map each angle to a Cialdini principle (authority, social proof, scarcity, reciprocity, liking, commitment/consistency) and note which Schwartz awareness stage it is optimized for.
</instructions>

<format>
Numbered list. For each: Angle label, Cialdini principle, Awareness stage, Headline, Opening paragraph. No full drafts — variations only.
</format>

<avoid>
Producing full drafts at this stage. Using the same hook mechanism across multiple variations. Drifting from the Voice Doc on any variation.
</avoid>

3. Tone-Check a Draft Against the Voice Doc

Before you send a draft to a client, run it through a structured tone check. This surfaces drift early — before it becomes a revision round.

<context>
Client: [Name]. Voice Doc is uploaded to this Project. Draft to review: [paste the full draft].
</context>

<instructions>
Audit the draft against the Voice Doc. For each Voice Adjective in the doc, assess: does this draft reflect it? If yes, cite one specific sentence as evidence. If no, cite one sentence that violates it and suggest a specific revision. Then flag: any banned words or phrases used, any sentences that exceed the preferred length/rhythm rules, any place where the awareness stage assumption seems off.
</instructions>

<format>
Table with columns: Voice Attribute | Pass/Fail | Evidence or Violation | Suggested Fix. Summary paragraph with overall assessment and top two priority revisions.
</format>

<avoid>
Vague feedback like "consider simplifying." Every flag must cite a specific sentence from the draft and a specific fix.
</avoid>

4. Run a Structured Approval Workflow

Client approvals fail when versions are unclear, feedback is undocumented, and the sign-off is a Slack message no one can find. This workflow produces a clean approval package.

<context>
Project: [Client, piece name, channel]. Version history: [list previous versions and key changes, e.g., "V1: original draft. V2: shortened headline per client feedback 4/20."]. Current version: [paste V-current]. Outstanding feedback to incorporate: [paste raw client notes if any].
</context>

<instructions>
Produce an approval package containing: (1) a change log table showing V1 → V-current deltas, (2) the current draft with tracked-change annotations for anything changed since last client review, (3) a sign-off request summary (3 sentences: what changed, why, and what you need the client to confirm), (4) a "scope boundary" note listing anything the client raised that was explicitly out of scope for this revision.
</instructions>

<format>
Markdown. Change log as a table. Draft with inline annotations using [CHANGED: reason] format. Sign-off summary as a short paragraph. Scope boundary as a bullet list.
</format>

<avoid>
Burying scope boundary items. Presenting V-current without the change context. Any language that implies approval has already been given.
</avoid>

5. Repurpose Approved Copy into Platform-Native Variations

Approved copy is a locked asset. Once signed off, you can extract maximum value by adapting it to each platform's native format, length, and convention — without reinventing the brief.

<context>
Approved copy: [paste the signed-off piece]. Original channel: [e.g., landing page hero]. Target platforms: web, email subject line + preview text, X/Twitter thread, LinkedIn post. Voice Doc: [reference uploaded or paste key rules]. Regulatory constraints: [e.g., FTC Endorsement Guides apply — no implied guarantees; or FINRA — include required disclosures].
</context>

<instructions>
Adapt the approved copy for each platform below. Preserve the core message and voice. Apply platform-native conventions: Web excerpt (100–150 words, SEO-friendly H1 + meta description), Email (subject line 40–50 chars + preview 85–100 chars — avoid spam triggers), X thread (3–5 tweets, first tweet hooks without being clickbait, thread ends with CTA), LinkedIn post (200–300 words, opens with a hook not a preamble, ends with a question or CTA). Include required disclosures on every variation where they apply.
</instructions>

<format>
One section per platform, labeled with H2. Each variation labeled clearly. Disclosures in bold at the end of any variation requiring them.
</format>

<avoid>
Stripping claims that require FTC/FINRA/FCA disclosure rather than adding the disclosure. Cross-contaminating platform conventions (a LinkedIn post should not read like a tweet thread). Introducing new claims not present in the approved source.
</avoid>

What This Looks Like in Your Week

Monday. New client onboarded on Friday. You sit down with the intake transcript, brand guidelines, and two approved ads the client liked. You run Workflow 1 and have a Voice Doc in 30 minutes. You upload it to the client's Claude Project. Every piece this week starts from that doc.

Tuesday. Brief in. Campaign email, mid-funnel audience, 300 words. Instead of free-writing, you run Workflow 2 to generate five angle variations — two problem-agitation, two social proof, one authority play. You send the five headlines and opening paragraphs to the client for direction before writing a word of the full draft.

Wednesday. Client picks the social proof angle. You draft the full email, then run Workflow 3 against the Voice Doc before sending. One voice attribute fails — the rhythm is too formal for this client's conversational baseline. You fix three sentences. The draft goes to the client clean.

Thursday. Revision round. Client sends notes in a Google Doc comment thread. You run Workflow 4 to build the approval package: change log, annotated V2, sign-off summary. You also include a scope boundary note — they asked for a full page rewrite in the comments; you document that this is V2 of the email only. Scope creep is documented before it happens.

Friday. Client approves V2. You run Workflow 5 to repurpose the email into a LinkedIn post, X thread, and web excerpt. Required FTC disclosures go into every variation automatically because you specified it in the prompt. You deliver four assets from one approved piece in under an hour.

What to Avoid

Working without a Voice Doc. Every piece you write without one is a coin flip on whether the client recognizes their own voice in the output. The Voice Doc is what you are selling — judgment about their brand. If Claude can generate copy without it, a cheaper tool can too.

Sending angle variations as drafts. Variations are direction-setting tools, not deliverables. Sending five full drafts trains the client to expect unlimited iterations for a fixed fee and buries the judgment that makes variations useful.

Using Claude's output as the compliance review. Claude will flag common FTC and FINRA patterns if you prompt it, but it does not know your client's specific disclosures, state-level requirements, or current regulatory guidance. Get counsel or your client's compliance team to sign off on regulated copy. You catch the obvious issues; they catch the rest.

Blending client voices across Projects. If you run multiple clients from one Claude Project, voices contaminate. Client A's casual directness bleeds into Client B's formal authority. One project per client is not optional.

Skipping the approval package. A Slack message that says "looks good" is not sign-off. The approval package in Workflow 4 creates a paper trail, documents scope, and protects you when a client claims they "never approved that."

Resources

Get weekly AI prompts for Copywriting professionals

Join professionals already saving hours every week. Free. No spam.