How to use Claude for marketing (Projects-first workflow)

Use Claude for marketing as a Projects-first loop, not a blank chat and not autopilot publish. Load brand guidelines and examples into a Project, brief before you draft, run prompts as specs with required inputs, then human-QA anything customer-facing. Chat is for spikes; skills come after a prompt stops changing. For path choice by job, use the Claude for marketing map. For the full prompt pack, use Claude prompts for marketing.

When to use Claude / when not

Use Claude when you can hand it a real file set: brand voice, ICP notes, offer facts, 2-3 on-brand examples, and proof you can stand behind. Strong jobs: campaign briefs from messy notes, voice rewrites, competitor messaging tear-downs from pasted copy, single-section drafts.

Don't use Claude as live media buyer, bid engine, or legal/compliance signer. Don't invent metrics, testimonials, logos, or competitor weaknesses. Don't publish first-pass ads or emails without a human gate. Empty context ("just write marketing copy") produces generic output - that is a setup failure, not a model failure. Same rule for competitive claims: only tear down copy you can paste and label.

Channel depth (SEO systems, nurture maps, social calendars) belongs on later spokes - not this how-to. Start from the marketing map if today's job is a different fork.

Setup: Projects as default

Default surface for marketing work is a Project (Claude Projects), not a naked chat thread. Projects hold persistent brand context so you stop re-pasting the same guidelines every morning.

Docs to load (minimum viable pack):

Name the Project by client or brand - agencies: one Project per client, hard boundaries. Paste the agent block (below) as Project instructions once the docs are in. Keep banned claims and proof sources in the same Project so every run shares one truth set.

If you only have a messy Notion dump, paste it once as notes-raw.md, then run a positioning or offer cleanup before campaign work - do not feed raw dumps into customer-facing drafts. Chat still works for one-off spikes. Skills come later (Claude Skills). If you are still setting up Claude itself, see How to use Claude. Claude Code is optional automation for technical operators - not day-one marketing setup. Confirm Projects behavior on Anthropic's What are projects? the day you implement (Checked 2026-09-12). Product names change; this page is field setup, not a product clone.

Path router if you're still choosing the job: Claude for marketing.

Workflow (brief - draft - QA - skill)

Run this loop for every customer-facing deliverable. Skip a step and quality drops.

1. Define the job - One sentence. Example: "Campaign brief for Q4 webinar promo" or "Rewrite homepage hero to match voice examples." If you can't name one job, you're not ready to prompt.
2. Open the right Project - Brand docs already loaded. If context is missing, add it before drafting - don't paper over gaps with clever prompts.
3. Brief before draft - Collect required inputs (audience, offer, CTA, proof or "none", constraints). Thin inputs - mark [GAP]; do not invent fills.
4. Run one prompt as a spec - Paste a full spec (role · inputs · steps · rules · output). Fill every required field. Refuse empties.
5. Draft in the Project - Keep the thread on that single job. Scope creep (brief - full ad set - nurture map) is a new run, not the same chat.
6. Human QA - Kill invented claims; check voice against examples; verify CTA and proof. Label output draft until a human approves.
7. Save the winner - Store the brief + final copy + what worked in the Project or a shared doc.
8. Skill candidate - When the same prompt stops changing for 3+ runs, promote it toward a skill (stub below) instead of re-pasting forever.

That is the day-to-day system. Time-box QA: claim check, voice check against examples, CTA check - then ship or reject. Depth on prompt cards lives in Claude prompts for marketing; this page owns the loop.

Prompts for the loop (≥3)

Three workflow-tied specs - enough to run brief - rewrite - competitive angles. Not the full five-card pack. Copy the pack (positioning, section draft, extras) from Claude prompts for marketing. Full mirror of these three specs lives on the prompts guide.

1. Campaign brief

Role: Campaign lead. Inputs: goal · audience · offer · one CTA · proof · channels · timeline/budget (or "unknown") · voice notes. Output: Objective · Audience · Insight · 3 pillars · Offer · CTA · Proof · Channels · Risks. QA: one job · one CTA · proof sourced · channels match. When not: full ad sets, media plans, bids.

You are a campaign lead. Write an executable brief from inputs only.

REQUIRED INPUTS: Campaign goal (one sentence); Audience; Offer; Primary CTA (one); Proof (or "none"); Channel(s); Timeline/budget (or "unknown"); Brand voice notes/examples.

STEPS: (1) Restate the job in one line from the stated goal. (2) Draft Markdown: Objective | Audience | Insight | Message pillars (exactly 3) | Offer | CTA | Proof | Channels | Risks/do-nots. (3) Insight from inputs; if thin, mark [GAP] and ask one question. (4) Proof = provided only.

RULES: One primary CTA. No full ads, LPs, or email sequences. No invented metrics or competitor claims.

OUTPUT: brief + filled QA (single job, single CTA, proof sourced, channels match).

2. Brand-voice rewrite

Role: Brand editor. Inputs: draft · 2-3 on-brand examples · do/don't rules · audience · forbidden claims. Output: rewrite · 5 voice patterns · change notes · risks. QA: matches cadence · no new claims · forbidden absent. When not: zero examples; need legal sign-off.

You are a brand editor. Rewrite the draft to match provided examples  -  not generic "professional."

REQUIRED INPUTS: Draft copy; On-brand examples (2-3); Voice rules (do/don't); Audience; Forbidden claims.

STEPS: (1) Infer 5 voice patterns from examples (length, person, concreteness, humor, CTAs). (2) Rewrite applying patterns + rules. (3) Bullet what changed and why. (4) Remove/bracket unsupported or forbidden claims.

RULES: Add no metrics, testimonials, or capabilities absent from inputs. Don't flatten specific examples into corporate bland. Keep roughly the same length unless asked.

OUTPUT: rewritten copy | 5 patterns | change notes | residual risks.

3. Competitor messaging teardown - angles

Role: Messaging analyst on pasted copy only. Inputs: our offer · ICP · labeled competitor copy · real differentiators · do-not-claim. Output: theme table · 5 angles · evidence needed · do-not-claim. QA: no invented competitor facts · angles tied to differentiators · sources labeled. When not: no paste; smear requests.

You are a messaging analyst. Tear down competitor messaging from PASTED copy only, then propose angles.

REQUIRED INPUTS: Our offer; Our ICP; Competitor copy (paste; label each source); Our real differentiators; Claims we must not make.

STEPS: (1) Theme table: theme | example phrase | source label. (2) Contrast themes with our differentiators. (3) Five angles (one line + why for ICP). (4) Evidence still needed before any contrast claim. (5) Do-not-claim list (bans + anything needing unstated competitor facts).

RULES: Never invent competitor pricing, customers, weaknesses, or roadmap. Thin copy  -  stop at themes; no speculation. No smear.

OUTPUT: theme table | 5 angles | evidence needed | do-not-claim | QA self-check.

Full pack + optional extras - Claude prompts for marketing.

Agent block (copy-paste)

Paste as Project instructions after brand docs are loaded. Enforces inputs and blocks invented metrics.

You are the Project operator for marketing work in this Claude Project.

CONTEXT: Obey brand guidelines, ICP/offer notes, and example assets loaded here. If critical context is missing, ask before drafting.

FLOW:
1. Confirm the single job for this run (brief, rewrite, teardown, or one section draft).
2. Collect every required input; refuse empty critical fields.
3. Draft only from provided facts; mark [GAP] where inputs are thin.
4. Return draft + filled QA checklist.
5. End with open questions  -  never invent answers.

HARD REFUSALS: No invented metrics, testimonials, awards, logos, or competitor claims. Label all output as draft  -  never "approved" or "ready to publish." No live media buying, bid advice, or legal/compliance sign-off. No expanding one section into a full site, nurture map, or SEO strategy under this Project.

If the job belongs on another spoke, say so and stop.

Skill path

Two honest paths - neither is a downloadable zip yet:

1. Stub on this spoke: marketing-project-setup skill stub (outline only). Encodes Project docs checklist, job choice, then brief/draft/QA gate. Outline only.
2. Prompts runner (sibling): the marketing-prompt-pack stub under Claude prompts for marketing - once prompts stabilize, run them as a skill instead of re-pasting.

Install order for the first few marketing skills: What Claude marketing skills should I install first?. Full catalog depth: Claude marketing skills (planned). Until packaged, use the agent block above.

Stack at a glance

Field note, tested against Claude Projects workflows in September 2026: the loop holds when the Project already holds guidelines, examples, and claim bans. Verify surfaces on claude.com and Anthropic docs (Checked 2026-09-12).

Pitfalls

• Empty Project - guidelines missing - generic voice.
• Draft before brief - you get pretty copy with the wrong job.
• Prompt dump without QA - specs without a checklist still ship invented proof.
• Autopublish - first pass is draft; human owns customer-facing claims.
• Wrong spoke - SEO playbooks, email lifecycles, social systems - leave via the marketing map; channel URLs marked (planned) until live.

Quality bar: one job · inputs filled · no invented claims · human sign-off before publish.

FAQ

Do I need Claude Code to use Claude for marketing?

Usually no. Projects + prompts + human QA cover most marketing wins. Reach for Code only when automating repo or report workflows - after the Project loop is stable.

Projects vs Chat for marketing - which first?

Projects first when brand context repeats. Chat for one-off spikes. If you re-paste guidelines every day, you chose the wrong surface.

How is this different from the marketing map?

The map routes by job. This page is the day-to-day setup and loop. Don't expect fork depth here.

Where is the full prompt pack?

Claude prompts for marketing - five curated specs + agent block. This how-to only ships the three loop-tied prompts above.

When should I turn a prompt into a skill?

When the same spec stops changing for several real runs and inputs are stable. Start from the stubs; see What Claude marketing skills should I install first? for install order. Packaging catalog: Claude marketing skills (planned).

Where to go next

Growth hub (planned) (hub) · Claude for marketing (map) · Claude prompts for marketing (full pack) · Claude marketing skills (planned). Channel spokes when you need depth: content creation (planned) · email (planned) · social (planned) · SEO (planned). Run the loop; keep a human on publish.