The short answer: Claude is the strongest single tool a marketing team can plug into a campaign workflow in 2026. Anthropic’s Claude Sonnet 5 (released June 30, 2026) ships with a 1M-token context window, an “adaptive thinking” reasoning mode, Projects that hold your brand bible, and Skills that package repeatable workflows. The 10 prompts below turn that capability into ten campaign deliverables - strategy doc, persona, creative brief, channel plan, email sequence, social calendar, landing page, paid ad copy, post-campaign retro, and brand voice bible - with concrete 2026 marketing data baked into the system prompts.
HubSpot’s 2026 State of Marketing Report found that 80% of marketers use AI for content creation, 75% for media production, and 61% say marketing is going through its biggest disruption in 20 years because of AI. The marketers pulling ahead are not the ones writing more prompts. They are the ones who built a system so their prompts produce consistent, on-brand, compliant output every time.
That’s what this guide is.
“Today, more content is generated by AI than by humans. But it’s mostly average. Consumers seek human-created content, and will tune out brand and AI-generated content.” Kieran Flanagan, SVP of Marketing, AI & GTM, HubSpot - 2026 State of Marketing Report
Why Claude (and Which Version) Shines for Marketing in 2026
Definition: Claude is Anthropic’s family of large language models. The current generation available to marketers in July 2026 includes Claude Sonnet 5 (the default model on Free and Pro plans), Claude Opus 4.8 (for complex agentic work), Claude Haiku 4.5 (cheapest, fastest), and the new Claude Fable 5 (most capable, expensive, generally available since June 9, 2026).
Three things make Claude different from ChatGPT, Gemini, or Copilot for marketing work specifically:
1. A 1M-token context window on every Sonnet 5, Opus 4.8, and Fable 5 model. That is roughly 555,000 words or 2.5 million Unicode characters in a single conversation (Anthropic Models Overview, July 2026). You can paste a year of customer interviews, an entire competitor corpus, and a 60-page brand bible into one project, and the model will hold the relationships across all of it. For a campaign brief that needs to reconcile customer language with positioning and channel constraints, that single-window capability is the difference between “drafting from notes” and “drafting from the full truth.”
2. Adaptive thinking and an explicit effort parameter. Sonnet 5, Opus 4.8, and Fable 5 all ship with adaptive thinking, where the model decides how long to think before answering. On Opus 4.8 and Sonnet 5, you can tune the effort parameter (low, medium, high, xhigh) to trade cost for depth (Anthropic Sonnet 5 launch, June 30, 2026). Marketing strategy prompts get high. Subject-line A/B variants get low.
3. Skills, Projects, Artifacts, and Memory - features purpose-built for repeat marketing work.
- Projects are persistent workspaces. Drop your brand bible, persona docs, and prior retro notes into a Project, and every new chat in that Project has them baked into the system prompt.
- Skills are packaged prompt + tool workflows that you can invoke with
/skill-name. Skills are also deployable org-wide on Team and Enterprise plans (Anthropic Pricing, July 2026). - Artifacts are side-panel documents Claude creates alongside the chat (HTML, code, structured docs) - perfect for handing a campaign brief back to your team.
- Memory persists preferences across conversations so Claude remembers your brand voice, your house style, and your banned phrases.
The version you should pick depends on what you are doing:
| Model | Context | Input $ / MTok | Output $ / MTok | Best for marketing |
|---|---|---|---|---|
| Claude Haiku 4.5 | 200K | $1 | $5 | High-volume A/B copy variants, subject lines, social captions |
| Claude Sonnet 5 | 1M | $2 intro (then $3) through Aug 31, 2026 | $10 intro (then $15) | Default for briefs, sequences, landing pages, calendars |
| Claude Opus 4.8 | 1M | $5 | $25 | Complex multi-document strategy, full-campaign synthesis, audits |
| Claude Fable 5 | 1M | $10 | $50 | Long-running agentic workflows, research at scale, browser/terminal work |
(Sources: Anthropic Models Overview, Anthropic Pricing, July 2026.)
The pricing change matters. Sonnet 5 launched on June 30, 2026 at $2 / $10 per million tokens through August 31, 2026, then steps to $3 / $15. If you are running a big campaign Q3 push, lock in the intro pricing by deploying before September.
At-a-Glance: 10 Prompts by Campaign Stage
Every prompt below maps to a real campaign stage, ships with a sample output, a channel variation, an A/B variant, and a compliance flag. The “Claude feature used” column tells you which Anthropic feature to lean on.
| # | Prompt | Campaign Stage | Output Type | Claude Feature Used | When NOT to Use |
|---|---|---|---|---|---|
| 1 | Campaign Strategy Doc | Strategy | 8-section markdown brief | Project + Opus 4.8 | Budget-only approvals (use finance tool) |
| 2 | Audience Persona | Strategy | JSON persona + empathy map | Artifacts | B2B accounts (use Clay / Apollo instead) |
| 3 | Creative Brief | Creative | One-page brief in Artifact | Skills | Brand-system refresh (use agency partner) |
| 4 | Channel Plan | Channel | Allocation table by stage | Project | Real-time bidding (use Ads platforms) |
| 5 | Email Sequence | Channel | 5-email series with subject variants | Memory | Transactional email (use ESP templates) |
| 6 | Social Calendar | Channel | 30-day calendar | Skills | Crisis comms (route to comms lead) |
| 7 | Landing Page | Channel | Full-page copy + variants | Artifacts | Multi-step funnels (split per step) |
| 8 | Paid Ad Copy | Channel | Channel-specific copy sets | Adaptive thinking high |
Sensitive categories (health, finance, kids - see compliance) |
| 9 | Post-Campaign Retro | Measurement | Performance debrief doc | Project + Opus 4.8 | Campaigns < $1K spend (skip the formality) |
| 10 | Brand Voice Bible | Strategy | Reusable voice + guardrails | Skills (deployable org-wide) | Logo / visual identity (use design tool) |
The categories Anthropic recommends for repeatable marketing work - system-prompt patterns, brand voice, structured outputs - are baked into each prompt below. The prompts also follow Anthropic’s documented best practices for prompting Claude in 2026: clear success criteria, role + context + task, examples, XML tags for structure, and an explicit output format.
Anthropic’s Brand-Voice / System-Prompt Patterns (Use Them First)
Before you drop into the 10 prompts, install this scaffolding. Anthropic’s prompt engineering docs (Claude Prompting Best Practices, 2026) consistently recommend four patterns. None of the 10 prompts below work as well without them.
Pattern 1: Role + Context + Task + Format. Open every marketing prompt with role, ground it in context, give it a task, and constrain the format. Example skeleton:
<role>You are a senior direct-response copywriter for [BRAND].</role>
<context>[paste brand voice bible, audience, offer, constraints]</context>
<task>Write [deliverable] for [audience segment] that achieves [goal].</task>
<format>Output as markdown with H2 sections: Hook, Body, CTA. Use bullets
under 12 words. Match tone: confident, plain-spoken, no jargon.</format>Pattern 2: Few-shot examples for tone. Claude anchors tone harder from examples than from adjectives. Include two or three short excerpts that are exactly the voice you want, then write “match the tone of the examples above.”
Pattern 3: XML tags for structure. Anthropic explicitly recommends XML tags over JSON for prompt structure. Use <context>, <example>, <instructions>, <output>. Claude was trained to treat these as boundaries.
Pattern 4: Negative constraints. Tell Claude what not to do. “Do not use exclamation marks. Do not use ‘unlock’, ‘harness’, ‘delve’, ‘ever-evolving’. Do not invent statistics. If a claim is unverified, write ‘[VERIFY]’ next to it.”
HubSpot’s 2026 research is blunt: “Most marketers today use generative AI as a starting point” - only 7% publish AI content with no edits, and 56% make significant edits before publishing (HubSpot blog, May 2026). The negative-constraint pattern is the cheapest way to cut the editing load.
1. Campaign Strategy Doc
Stage: Strategy
Best model: Opus 4.8, effort: high
Claude feature: Project with brand bible + persona + last-quarter retro
The Prompt
<role>You are a senior marketing strategist who has shipped 50+ product
launch campaigns for B2B SaaS companies between $10M and $500M ARR.</role>
<context>
Brand: [BRAND]. Product: [PRODUCT]. Audience: [PRIMARY PERSONA].
Stage: [launch / expansion / retention]. Quarter goal: [REVENUE OR
PIPELINE TARGET]. Competitive set: [LIST 3-5 COMPETITORS].
Last quarter results (paste from analytics):
- Pipeline generated: $
- CAC: $
- Top channel: [CHANNEL] at X% of pipeline
- Worst channel: [CHANNEL] at Y% of pipeline
</context>
<task>
Build a one-page campaign strategy doc for the next 90 days.
The doc must reconcile the brand bible, the persona, and the prior
quarter's data - not ignore any of them. Pick one primary message,
one supporting message, and one channel to deprioritize.</task>
<format>
Markdown with these H2 sections, each under 120 words:
## Objective & Guardrails
## Primary Message
## Supporting Message
## Audience Tension (the specific belief we need to move)
## Channel Mix (table: Channel | % of Budget | Role | KPI)
## Channel to Deprioritize (and why, with data)
## Risks & Mitigations
## 90-Day Milestones (table: Week | Milestone | Owner)</format>
<instructions>
- Pull facts from the context. Do not invent statistics.
- If a number is missing, write "[VERIFY]".
- Do not use these words: "leverage", "synergy", "delve", "unlock",
"game-changer", "ever-evolving".
- Match the tone of the brand bible pasted above.
</instructions>Sample Output (abbreviated)
Objective & Guardrails
Drive $4.2M in new ARR pipeline over 90 days at a CAC ceiling of $11,400. Guardrails: no claims about AI replacing human work (brand bible §4.2), no testimonials without signed release (legal review at brand@…).
Primary Message
“Stop routing your customer success tickets through six tools.”
Audience Tension
CS leaders believe consolidated platforms mean migration pain they can’t afford in Q4. Move them by leading with the 14-day pilot.
Channel Variation
For an agency client, swap <role> to “fractional CMO at a 20-person B2B agency” and add a sub-section to the format: ## Client Approval Path (legal, brand, exec sponsor).
A/B Variant
Change <task> to also generate a counter-positioning doc: the strategy a competitor would run against you. Use Sonnet 5 with effort: high.
Compliance Tip
Claims about specific dollar results, ranking, or awards must be backed by internal data - flag with [VERIFY] so a human checks before it ships.
2. Audience Persona
Stage: Strategy Best model: Sonnet 5 Claude feature: Artifacts (Claude returns a styled HTML persona card)
The Prompt
<role>You are a research-trained marketer who specializes in Jobs To Be
Done interviews and audience segmentation.</role>
<context>
Product: [PRODUCT]. Price point: $[X]. Sales cycle: [DAYS/WEEKS].
Past buyer interviews (paste 3-5 excerpts verbatim):
"[QUOTE 1]"
"[QUOTE 2]"
Top support ticket themes from last 90 days (paste list):
- [TICKET THEME 1] (X% of tickets)
- [TICKET THEME 2]
</context>
<task>
Synthesize the interviews and tickets into one primary persona.
Surface the literal language customers use - paraphrase kills resonance.
Identify the job the customer is hiring [PRODUCT] to do, the
competing alternatives they considered, and the moment they decided
to switch.</task>
<format>
Output as a Claude Artifact (HTML) with these sections:
## Snapshot (name, role, age, company size, single sentence summary)
## Job To Be Done (one sentence)
## Day-In-The-Life (3 bullets, each ≤15 words)
## Pains (5 bullets, each a real customer quote)
## Gains (3 bullets)
## Triggers (the 3 events that push them to evaluate new tools)
## Alternatives Considered (table: Alternative | Why they rejected it)
## Channels They Trust (table: Channel | Why they trust it)</format>
<instructions>
- Use the customer's exact words from the interviews, not paraphrases.
- Mark any inference you added (not a quote) with [INFERENCE].
- Do not invent demographic data. If age or company size isn't in the
interviews, write "Unknown - verify with sales."
</instructions>Sample Output (snippet)
Snapshot
“Maya, the operations director at a 60-person logistics company, is the person who gets paged when shipment ETA emails fail.”
Day-In-The-Life
- Juggles three Slack channels before 9am
- Has a standing 1:1 where she apologizes for routing delays
- Spends Friday afternoons writing runbooks nobody reads
Channel Variation
For B2C e-commerce, replace “interviews” with “Amazon reviews (paste 30-50)” and “support tickets” with “return reasons.” Output a buyer persona plus a non-buyer persona.
A/B Variant
Generate a “near-buyer” persona - someone close to buying but blocked - alongside the primary persona. Both appear in one Artifact.
Compliance Tip
Customer quotes must be either verbatim from the source you provided or marked [INFERENCE]. Never invent quotes. The FTC’s Keep Your AI Claims in Check guidance (Feb 2023, still active 2026) treats invented customer language as a deceptive practice.
3. Creative Brief
Stage: Creative
Best model: Sonnet 5
Claude feature: Skills (save as /campaign-brief Skill for the team)
The Prompt
<role>You are a creative director with 15 years at Droga5, Anomaly,
and independent shops. You write briefs that copywriters and art
directors can actually run with.</role>
<context>
Campaign name: [NAME]. Background: [1-2 SENTENCES OF BUSINESS
CONTEXT]. Target: [PERSONA FROM PROMPT #2]. Competitors we are
explicitly not: [LIST]. Single-minded proposition we want to land:
"[THE ONE THING]". Brand voice: [paste 2-3 lines from voice bible].
Mandatory inclusions: [LOGO TREATMENT / LEGAL FOOTER / CLAIMS].
</context>
<task>
Write a one-page creative brief a creative team can pick up tomorrow.
Make the strategy line land the audience tension in a single sentence.
The brief should be tight enough that an art director can sketch
thumbnails in 30 minutes.</task>
<format>
Markdown, exactly one page when printed. H2 sections:
## The Strategy (one sentence, ≤25 words)
## Why Now (≤40 words, with the trigger event)
## Audience Truth (the thing they believe today that we need to move)
## The Single-Minded Promise (≤15 words)
## Reason To Believe (3 bullets, each a real proof point)
## Tone (3 adjectives + 3 words to avoid)
## Deliverables (table: Asset | Spec | Quantity)
## Mandatories (logo, legal, footer, claims)</format>
<instructions>
- The strategy line must be one sentence, period at the end.
- Reason To Believe items must come from context - if a proof point
isn't there, write "[PROVIDE PROOF]".
- Do not use the words "compelling", "premium", "elevated", "best-in-class".
</instructions>Sample Output (excerpt)
The Strategy
Logistics directors trust the platform that gets a notification right the first time.
Tone
Direct. Calm. Earned.
Avoid: hype, exclamation marks, “supercharge,” “AI-powered,” “next-gen.”
Channel Variation
For OOH (out-of-home), replace ## Deliverables with billboard and transit specs (e.g., “20-sheet poster, 7-sheet, transit king/queen, NJ Transit horizontal”). For TikTok, replace with “9:16 vertical, ≤21 seconds, captioned, sound-on and sound-off variants.”
A/B Variant
After the brief, ask Claude to write the “anti-brief” - the brief a competitor would write to undermine your strategy. It’s a forcing function to test whether your strategy line holds.
Compliance Tip
“Reason To Believe” claims must be source-backed. The FTC pursued Operation AI Comply in September 2024 specifically against companies making unsupported AI and performance claims (FTC press release, Sept 2024). The bar did not loosen in 2026. If you can’t prove it, don’t write it.
4. Channel Plan
Stage: Channel
Best model: Opus 4.8 with adaptive thinking high
Claude feature: Project (paste last quarter’s channel data)
The Prompt
<role>You are a performance marketing lead who has managed $20M/yr
across paid search, paid social, programmatic, email, SEO, and
partnerships.</role>
<context>
Total budget: $[X]. Primary KPI: [PIPELINE / REVENUE / TRIALS].
Audience: [PERSONA FROM PROMPT #2]. Customer LTV: $[X]. Target CAC:
$[X]. Last quarter channel performance (paste table from your
analytics):
| Channel | Spend | Pipeline | CAC | MQL->SQL rate |
| --- | --- | --- | --- | --- |
| Google Search | $X | $Y | $Z | A% |
| LinkedIn | ... | ... | ... | ... |
| Meta | ... | ... | ... | ... |
| Email | ... | ... | ... | ... |
| SEO | ... | ... | ... | ... |
| Partnerships | ... | ... | ... | ... |
</context>
<task>
Build a 90-day channel plan that meets the KPI at or below target CAC.
Decide where to add, where to hold, and where to cut. Justify each
decision with the prior quarter data.</task>
<format>
Markdown with these H2 sections:
## Headline Allocation (table: Channel | Budget | % | Role | KPI Target)
## Why This Mix (≤80 words, with 3 numbers from context)
## Channel-by-Channel Plan
(for each channel, 3-4 bullets: targeting, creative angle,
experiment, kill criteria)
## What We Are NOT Doing (and why)
## 30/60/90-Day Spend Curve
## Risk Plan (what we cut first if pipeline lags by 25%)</format>
<instructions>
- Every allocation decision must cite a number from context.
- If context is silent on a channel, write "[NEEDS DATA]" - do not invent.
- Do not propose channels outside the existing set without naming the
test budget required to evaluate them.
</instructions>Sample Output (excerpt)
Why This Mix
LinkedIn produced 38% of pipeline at $9,200 CAC last quarter - 19% under target. Hold at 45% of budget, increase scale on lookalike of closed-won accounts. Meta delivered reach but MQL→SQL at 4%, well below the 9% benchmark. Cut from $X to 0.6X and retarget only.
Channel Variation
For a launch with no prior data, replace the channel performance table with a category benchmark from HubSpot’s 2026 report (e.g., 80% of marketers use AI for content creation; paid search CPC for SaaS averages $X in 2026) and run the plan as a hypothesis.
A/B Variant
Generate two plans - a “defend CAC” plan and a “maximize pipeline” plan - and present the tradeoff in a 3-column Artifact.
Compliance Tip
LinkedIn, Meta, and TikTok all restrict claims in financial services, health, and political verticals. Each platform publishes its own ad policy library; pull the current version into the Project before generating. As of 2026, Meta’s policy still prohibits “before-and-after” weight-loss claims in health ads without substantiation.
5. Email Sequence
Stage: Channel Best model: Sonnet 5 Claude feature: Memory (so Claude remembers your house voice across the sequence)
The Prompt
<role>You are a lifecycle email marketer who has shipped 200+ nurture
sequences for SaaS, DTC, and B2B services.</role>
<context>
Product: [PRODUCT]. Trigger event: [WHAT JUST HAPPENED - e.g., free
trial signup, abandoned cart, demo no-show]. Audience state: [WHAT
THEY BELIEVE RIGHT NOW]. Brand voice: [paste 2-3 lines]. House style:
[salutation rules, signature block, image policy, link tracking
conventions]. Past high-performing subject lines (paste 3-5):
- "[SUBJECT A]" - 64% open
- "[SUBJECT B]" - 58% open
Pain points to address: [LIST 3].
</context>
<task>
Write a 5-email sequence that moves the reader from [STARTING BELIEF]
to [DESIRED ACTION]. Each email must have a clear single purpose.
Sequence must respect the unsubscribe expectation - do not bait.</task>
<format>
Markdown, one H2 per email. For each email:
## Email [N]: [Purpose]
**Send timing:** [trigger offset, e.g., T+0h, T+24h]
**Subject line:** [primary]
**Subject line A/B:** [variant]
**Preview text:** [≤90 chars]
**Body:** [full copy, ≤180 words for promotional, ≤300 for educational]
**CTA:** [one primary, one secondary if relevant]
**Plain-text version note:** [any plain-text-only treatment]
Then a final H2:
## Sequence Health Metrics (table: Email | Goal Metric | Target)</format>
<instructions>
- Subject lines ≤50 characters. Preview text ≤90 characters.
- Match the tone of past high-performers pasted above.
- Do not use: "FREE!!!", "Act now", "Limited time" unless context
confirms a real scarcity event.
- Every claim that touches price, availability, or result must
appear in the house-style legal review footer block.
</instructions>Sample Output (snippet)
Email 1: Confirm + Set Expectation
Send timing: T+0h (immediate on signup) Subject line: Your [BRAND] workspace is ready Subject line A/B: Three things to set up first Body: Hi [first name],
You’re in. Your workspace is ready at [URL].
In the next 24 hours, here are three things I’d set up before you invite your team - the order matters.
- Connect your data source (5 min)
- Invite one teammate (1 min)
- Run your first [workflow name] (3 min)
If you get stuck, reply to this email. A human reads every reply.
- [Name]
CTA: Open my workspace
Channel Variation
For SMS, cap each message at 160 characters and rewrite the body to lead with the link and the CTA. For in-app, replace email format with a 3-step onboarding tooltip script.
A/B Variant
Generate a “warm” sequence (assumes the reader is curious) and a “cold” sequence (assumes skepticism) and A/B test the first send.
Compliance Tip
CAN-SPAM, CASL (Canada), and GDPR (EU) all require a working unsubscribe, a physical address, and accurate header information. Add a footer block to the Project so Claude appends it automatically. The EU AI Act’s Article 50 (in force for transparency obligations since February 2025) requires that EU users know when they are interacting with AI-generated content - relevant if any email includes AI-personalized copy (EU AI Act text, EUR-Lex 32024R1689).
6. Social Calendar
Stage: Channel
Best model: Sonnet 5 with effort: low for volume
Claude feature: Skills (deploy as /social-calendar Skill across the social team)
The Prompt
<role>You are a social media director who has managed brand channels
across LinkedIn, Instagram, TikTok, X, and YouTube Shorts for
consumer and B2B brands.</role>
<context>
Brand: [BRAND]. Voice: [paste 2 lines]. Campaign theme: [THEME].
Key dates in the next 30 days: [LIST 5-7, including product launches,
industry events, cultural moments]. Top-performing past posts (paste
5-10 with engagement):
- "[POST]" - [PLATFORM] - [X likes, Y comments, Z shares]
Pinned or pillar content that must be referenced: [LIST 2-3].
Avoid: [TOPICS THAT HAVE NOT PERFORMED].
</context>
<task>
Build a 30-day social calendar. Hit each platform at least 3 times
per week, except YouTube Shorts (1-2 per week). Anchor 1 post per
week to a key date or pillar. Vary format - do not post five text
quotes in a row.</task>
<format>
Markdown table with columns:
| Date | Platform | Pillar | Format | Copy | Hashtags | Visual/Asset | Owner |
Follow with:
## Posting Cadence Summary (table: Platform | Posts/week | Best Time)
## Three Pull-Through Ideas (repurpose winners into a new format)
## Risks (dates/hashtags that may misfire)</format>
<instructions>
- Keep copy under each platform's hard length (LinkedIn 3000,
Instagram 2200, X 280, TikTok 2200 caption).
- Hashtags: 3-5 on LinkedIn and Instagram, 1-2 on X, 3-5 on TikTok.
- Do not invent stats, awards, or partnerships.
- If you propose an "X% of marketers" stat, source it inline.
</instructions>Sample Output (excerpt)
Date Platform Pillar Format Copy Hashtags Jul 21 Launch Carousel 5-slide breakdown: how [BRAND] hit [REAL METRIC] #marketing #B2B Jul 22 TikTok Behind-the-scenes 15s video “How we ship a campaign in 9 days” #marketingtok
Channel Variation
For X/Twitter, prioritize threads (5-8 tweets) over single posts for engagement. For LinkedIn, lean into document carousels (PDF) - HubSpot’s 2026 State of Marketing report found that document posts drove 3x the engagement of standard text posts for B2B brands in 2025 (HubSpot, May 2026).
A/B Variant
Generate the same 30 days as a “pillar-led” plan (one big idea per week, multiple cuts) vs. a “volume-led” plan (more posts, lower stakes). Pick based on team capacity.
Compliance Tip
Influencer posts must follow FTC Endorsement Guides (16 CFR Part 255, current as of 2026): material connections must be disclosed, and “#ad” or “#sponsored” must be visible without clicking “see more.” LinkedIn, Meta, TikTok, and YouTube have additional platform-specific disclosure tools. Bake disclosure rules into the Skill so every generated post includes the right tag.
7. Landing Page
Stage: Channel
Best model: Sonnet 5 with effort: high
Claude feature: Artifacts (Claude produces the page as an HTML preview)
The Prompt
<role>You are a conversion rate optimization specialist with experience
on Unbounce, Webflow, and Instapage. You write above the fold that
earns the next scroll.</role>
<context>
Product: [PRODUCT]. Offer: [WHAT THE VISITOR GETS]. Audience state
on arrival: [WHAT THEY BELIEVE / WHAT BROUGHT THEM]. Primary action:
[SIGN UP / BOOK DEMO / DOWNLOAD]. Secondary action: [READ CASE
STUDY]. Top objection (from sales calls, paste 1-2 quotes):
"[OBJECTION QUOTE]". Proof points (paste 3-4 with sources):
- [PROOF 1 - name + source]
- [PROOF 2]
Brand voice: [paste 2 lines]. Page length budget: [200 / 600 / 1500
words].
</context>
<task>
Write the full landing page copy. Lead with the visitor's pain in
their own words (pull from objection quotes). Prove the offer before
you ask for the click. Include one hero, three benefit blocks, one
objection-handling block, and one final CTA.</task>
<format>
Output as a Claude Artifact (HTML) with these sections rendered as
semantic HTML:
## Hero (H1, sub-headline, primary CTA, secondary CTA, trust line)
## Proof Bar (3 logos or 1 stat with source)
## Benefit Block 1: [PAIN -> OUTCOME]
## Benefit Block 2: [PAIN -> OUTCOME]
## Benefit Block 3: [PAIN -> OUTCOME]
## Objection Handler (headline that names the objection, body that
resolves it)
## Final CTA (echo the hero CTA)
## FAQ (3 Q&A)
After the Artifact, also output:
## Copy Variants (table: Element | Variant A | Variant B)
## Mobile Considerations (≤80 words)</format>
<instructions>
- H1 ≤9 words. Sub-headline ≤18 words.
- Every proof point must cite a source in the same line.
- Match voice: [paste 2 lines].
- Avoid: "unlock", "harness", "elevate", "supercharge", "game-changer".
</instructions>Sample Output (snippet)
Hero
H1: Stop chasing shipment ETAs across five tabs. Sub: One workspace for the ops team that gets paged at 2am. Primary CTA: Start the 14-day pilot Secondary CTA: Watch the 90-second tour Trust line: Used by [LOGO1], [LOGO2], [LOGO3] - case studies with metrics inside.
Channel Variation
For paid traffic, replace the FAQ with a single “What happens after I click?” block. For organic traffic, expand the FAQ to 6-8 questions and add a comparison table at the bottom.
A/B Variant
Generate Variant A (pain-led, written above) and Variant B (outcome-led, opens with the result). The Variant B version assumes a warmer audience and uses less defensiveness in the objection handler.
Compliance Tip
Free-trial or money-back guarantees must be honored exactly as written. The FTC’s 2026 enforcement has been active on “risk-free” claims that aren’t actually risk-free. If your CTA says “no credit card required,” it cannot require one. The EU AI Act Article 50 also applies if the page uses AI-generated personalization that the visitor isn’t told about - for EU traffic, add a single line: “Personalized with AI - see our AI transparency note.”
8. Paid Ad Copy
Stage: Channel
Best model: Sonnet 5 with effort: high (for big-budget campaigns) or Haiku 4.5 (for variant volume)
Claude feature: Adaptive thinking
The Prompt
<role>You are a paid media copywriter who has scaled B2B and DTC
accounts past $1M/mo on Google, LinkedIn, and Meta. You understand
platform character limits, ad rank, and policy.</role>
<context>
Product: [PRODUCT]. Offer: [OFFER]. Landing page: [URL or full copy
from prompt #7]. Audience segment: [WHO, INCLUDING PAST SEARCH
HISTORY OR JOB TITLES]. Keywords to lean into (paste 10-20 from
Google Ads Keyword Planner). Keywords to exclude (paste negative
list). Competitive claims we can back: [LIST]. Claims we cannot
make: [LIST].
</context>
<task>
Build a copy set for [PLATFORM]. Generate 10-15 ad variants per
ad group. Vary the angle: pain, outcome, social proof, objection
handler, offer. Make every variant self-contained - never rely on
imagery to carry the meaning.</task>
<format>
Markdown, one H2 per ad group. For each ad:
## Ad Group: [ANGLE NAME]
**Headlines** (rotate, ≤30 chars for Google, ≤70 for LinkedIn,
≤40 for Meta):
1. [H1]
2. [H2]
...
**Descriptions** (≤90 chars Google, ≤100 LinkedIn, ≤125 Meta):
1. [D1]
...
**Display URL path:** [path1/path2]
**CTA button:** [PLATFORM ALLOWED CTAs]
Then:
## Copy Coverage Matrix (table: Angle | # of variants | Channel format fit)
## Compliance Pre-Flight (table: Claim | Source | Status)</format>
<instructions>
- Do not use superlatives we cannot substantiate ("best", "#1", "fastest")
unless a third-party source is in context.
- Do not use urgency that is not real ("only 3 left" - only if true).
- Check each claim against the "claims we cannot make" list. If a
variant violates, delete it from the set.
- Stay under platform character limits including punctuation and spaces.
</instructions>Sample Output (excerpt - Google Ads RSA)
Ad Group: Pain angle - lost shipments
Headlines:
- Lost shipments cost more than tools
- Stop tracking ETAs in 5 tabs
- One workspace for ops teams Descriptions:
- [BRAND] consolidates shipment tracking, exceptions, and customer comms in one place. 14-day pilot.
- Operations teams at [LOGO1], [LOGO2] use [BRAND] to cut ETA misses by [REAL METRIC]%. Display URL: brand.com/pilot
Channel Variation
For LinkedIn, write single-image and document ad copy separately (LinkedIn Sponsored Content has stricter intro-text limits and tolerates longer B2B copy). For TikTok Spark Ads, generate a 15-second voiceover script alongside the on-screen text.
A/B Variant
Build a “rational” copy set (data, proof, specificity) vs. an “emotional” copy set (pain, relief, identity). HubSpot’s 2026 data shows B2B buyers respond 2.4x more to emotionally resonant copy in early-funnel paid placements (HubSpot State of Marketing 2026). Test both.
Compliance Tip
Google, Meta, LinkedIn, and TikTok each maintain an ad policy library updated in 2026. Common landmines: health/financial claims without substantiation, before-and-after imagery, weight loss, “get rich” promises, and AI-generated imagery of real-looking people (synthetic media disclosure required on Meta and TikTok as of 2026). Add the platform policy URLs to the Project.
9. Post-Campaign Retro
Stage: Measurement
Best model: Opus 4.8, effort: high
Claude feature: Project (paste all channel data, creative, and ICP docs)
The Prompt
<role>You are a marketing operations analyst who has run retros for
$50M+ in cumulative spend. You do not accept "we think" - you want
data, sources, and a clear next step.</role>
<context>
Campaign name: [NAME]. Period: [DATES]. Total spend: $[X]. Target
pipeline / revenue: $[X]. Actual: $[X]. Channel performance table
(paste from analytics):
| Channel | Spend | Pipeline | CAC | CTR | CVR | MQL->SQL |
| --- | --- | --- | --- | --- | --- | --- |
| ... |
Creative performance (paste top 5 ads by ROAS or pipeline). Audience
performance (paste segment-level data). Sales feedback (paste 5-10
quotes from AE call notes about this campaign).
</context>
<task>
Write a post-campaign retro the leadership team will actually read.
Compare actual to target. Identify the one decision that moved the
most pipeline, the one decision that cost the most, and the one test
to run next quarter. Cite the data for every claim.</task>
<format>
Markdown with these H2 sections:
## Headline (one sentence outcome)
## Scorecard (table: Metric | Target | Actual | Variance | Root cause)
## What Worked (3 bullets, each with the data behind it)
## What Didn't (3 bullets, each with the data behind it)
## Sales Voice (3 verbatim AE quotes, with what they tell us)
## Surprises (2 things we did not expect, with data)
## Next Quarter: One Test (full hypothesis: If [X], then [Y], measured by [Z])</format>
<instructions>
- Every claim must reference a number in the context table or paste.
- Do not generalize from a sample size under 5 unless flagged.
- "Worked" / "didn't" claims must include variance, not just direction.
</instructions>Sample Output (excerpt)
Scorecard
Metric Target Actual Variance Root cause Pipeline $4.2M $3.6M -14% LinkedIn scale-up hit audience cap by week 5 CAC $11,400 $10,900 +4% favorable N/A - improved despite volume miss What Worked
LinkedIn lookalike of closed-won accounts outperformed interest-based targeting by 38% on MQL→SQL (12% vs 8.7%, n=412 MQLs).
Channel Variation
For brand campaigns (vs. performance), replace the “CAC” and “Pipeline” columns with “Brand search lift,” “Direct traffic delta,” and “Unaided recall lift.” Source these from Google Brand Lift, Nielsen, or similar.
A/B Variant
Add a “what would have happened” counterfactual: if we had spent the entire $X on LinkedIn only, what would pipeline have been? Use the per-channel efficiency to model it.
Compliance Tip
Avoid retro language that overstates lift. “Campaign X increased pipeline 14%” implies causation; “Pipeline during campaign X was 14% above the prior 90-day baseline” is safer. The FTC has gone after marketers for attributing outcomes to campaigns that ran in parallel with other variables.
10. Brand Voice Bible
Stage: Strategy (build once, deploy everywhere)
Best model: Opus 4.8 with effort: high for the first draft, Sonnet 5 for ongoing maintenance
Claude feature: Skills - package the bible as a deployable Skill for your entire organization
This is the prompt that pays for the other nine.
The Prompt
<role>You are a brand strategist who has written voice bibles for
B2B and consumer brands. You care about rules a copywriter can apply
in 5 seconds, not adjectives that need a glossary.</role>
<context>
Brand: [BRAND]. Mission: [1 SENTENCE]. Audience: [WHO]. Anti-audience
(who we are NOT for): [WHO]. Three brand values (paste). Five
"voice exemplars" - short excerpts that are exactly the voice you
want (paste from existing copy, customer emails, or founder writing).
Five "voice violations" - excerpts that are close but wrong (paste).
List of competitor voices (one line each). Banned words (paste from
existing guidelines, if any).</context>
<task>
Build a brand voice bible a copywriter can apply without calling
you. The bible must be specific enough to settle an argument between
two writers in 30 seconds.</task>
<format>
Markdown with these H2 sections:
## Voice In One Sentence (≤15 words)
## Voice At A Glance (table: We are | We are NOT)
## The 4 Traits (each with: definition ≤20 words, "do this" example,
"not this" example)
## The 5 Voice Rules (numbered, each ≤30 words)
## Banned Words (alphabetical, with one-line replacement)
## Lexicon (table: Use this | Not this | Why)
## Grammar & Mechanics (3-5 bullets: serial comma, sentence length
cap, number formatting, capitalization)
## Visual Voice (2 bullets: what the voice looks like, what it doesn't)
## Application Examples (table: Asset | Approved example | Rejected
example)</format>
<instructions>
- Every "do this / not this" pair must be concrete sentences, not
adjectives.
- The Lexicon must replace at least 10 jargon or weak words.
- Banned words list must include every word in the "voice violations"
you pasted above.
</instructions>Sample Output (excerpt)
Voice At A Glance
We are We are NOT Direct, plain, a little dry Chatty, hyperbolic, breathless Specific by default Vague by default Earned confidence Bragging Banned Words (sample)
- leverage → use, rely on
- unlock → open, get, reach
- synergy → (delete; the word adds nothing)
- delve → go into, look at
- ever-evolving → always changing (or be specific about how it’s changing)
Channel Variation
For a global brand, add a section per locale: “Voice in German, Voice in Japanese, Voice in Brazilian Portuguese.” The rules stay the same; the lexicon shifts to local idiom.
A/B Variant
Generate the voice bible, then ask Claude to write the same bible from the perspective of the audience. “Describe how this brand sounds to a first-time visitor.” The gap between the two reveals overused language immediately.
Compliance Tip
If your voice bible contains claims (“trusted by 10,000 teams”), they need to be accurate and dated. The FTC’s Air AI enforcement action in August 2026 targeted unsupported business-growth and earnings claims directed at small businesses (FTC press release, August 2026). Voice bibles with stale numbers are a documentation risk in any FTC inquiry.
Once the bible is approved, save it as a Skill in your Claude workspace. On Team and Enterprise plans, deploy it org-wide via Organization wide skills deployment (Anthropic Pricing, July 2026). Every prompt from #1-#9 will then pull from the same source of truth.
Anthropic’s Prompt-Engineering Patterns Baked Into These 10 Prompts
Every prompt above follows the same five patterns Anthropic documents in its Prompting Best Practices for Claude (updated 2026):
- Clear success criteria - every prompt names what “good” looks like (a one-page brief, a 30-day calendar, a 5-email sequence).
- Role + context + task + format - the four-part skeleton Claude performs best with.
- Examples for tone - pasted exemplar copy anchors voice harder than adjectives.
- XML tags for structure -
<role>,<context>,<task>,<format>,<instructions>. Claude was trained to respect these as boundaries. - Negative constraints - explicit “do not use this word” lists cut the editing burden. HubSpot’s 2026 data is clear: marketers edit AI output heavily. The fastest way to cut editing time is to constrain the input.
A sixth pattern - assigning Claude a model-appropriate effort level - is new in 2026. Set effort: low for variant volume, effort: high for synthesis, and effort: xhigh for multi-document audits. The cost-per-call shifts, but the quality-per-dollar often improves because Claude stops thinking past the answer.
Compliance: What You Can’t Let Claude (or Anyone) Skip in 2026
Three compliance layers are non-negotiable for marketing work in 2026. The 10 prompts above flag the relevant ones inline, but a quick map:
1. FTC truth-in-advertising + AI-washing enforcement.
The FTC has run Operation AI Comply since September 2024 (FTC press release), targeting companies making unsupported AI and product claims. The August 2026 Air AI lawsuit extended that into B2B marketing tools targeting small businesses with deceptive growth and earnings claims. The bar is not “AI says it’s true.” The bar is “you can substantiate it.” Bake that into every prompt’s instruction block.
2. Platform ad policies.
Each platform’s ad policy is updated multiple times a year. Google, Meta, LinkedIn, and TikTok all restrict claims in financial services, health, weight management, political, and synthetic media categories. AI-generated imagery of realistic people must be disclosed on Meta and TikTok as of 2026. Pull the current policy into the Project before generating ads.
3. EU AI Act transparency.
The EU AI Act was published in the Official Journal on July 12, 2024, and the Article 50 transparency obligations have applied since February 2, 2025 (EU AI Act, EUR-Lex 32024R1689). If your marketing content is personalized with AI and reaches EU users, you need to tell them. Add a one-line disclosure: “Personalized with AI - see our transparency note.” It’s cheap and it’s the law.
The other rules (CAN-SPAM, CASL, GDPR for email) were already enforced and remain so. Nothing in the AI Act weakens them.
Choosing the Right Claude Model for Each Prompt (Decision Aid)
Quick-reference matrix based on the Anthropic Models Overview and Sonnet 5 launch blog, July 2026:
| Prompt | Recommended Model | Effort | Why |
|---|---|---|---|
| #1 Strategy Doc | Opus 4.8 | high | Multi-document synthesis, highest reasoning |
| #2 Persona | Sonnet 5 | medium | Strong, fast, no Opus-level reasoning needed |
| #3 Creative Brief | Sonnet 5 | medium | Synthesis of context into a tight format |
| #4 Channel Plan | Opus 4.8 | high | Data reconciliation, allocation tradeoffs |
| #5 Email Sequence | Sonnet 5 | medium | Voice consistency across 5 emails |
| #6 Social Calendar | Sonnet 5 (volume) or Haiku 4.5 | low | Volume and pattern, not depth |
| #7 Landing Page | Sonnet 5 | high | Long-form copy needs good structure |
| #8 Paid Ad Copy | Sonnet 5 (big spend) / Haiku 4.5 (variant volume) | high / low | Mix depth and volume |
| #9 Retro | Opus 4.8 | high | Multi-source analysis |
| #10 Brand Voice Bible | Opus 4.8 first draft; Sonnet 5 for maintenance | high / medium | Foundational, referenced everywhere |
For most marketing teams, Sonnet 5 is the right default for 70% of work - it has a 1M-token window, adaptive thinking, and the intro pricing through August 31, 2026 is the cheapest 1M-context model Anthropic has ever shipped.
FAQ
Which Claude model should I use for marketing campaigns in 2026? Claude Sonnet 5 is the default for Free and Pro plans and handles 70% of marketing work. It runs at $2 input / $10 output per million tokens through August 31, 2026 (then $3/$15), with a 1M-token context window and adaptive thinking. Use Claude Opus 4.8 for strategy docs, channel plans, and retros that need multi-document synthesis. Use Claude Haiku 4.5 for high-volume variant generation.
Does Claude actually beat ChatGPT for marketing campaigns? It depends on what you are optimizing. Claude wins on long-context tasks (over 200K tokens) because it holds the entire input in one window. ChatGPT wins on tool breadth (image gen, web search, code interpreter in the same chat) and a broader plugin ecosystem. For campaign work that needs brand voice consistency across long documents, Claude’s Projects + Skills + Memory stack is the cleaner system in 2026.
How many tokens does the average marketing campaign need? A single Project holding brand bible + persona + 3 prior retros + last quarter’s analytics lands at 80,000–250,000 tokens. That’s well inside the 1M window on Sonnet 5, Opus 4.8, and Fable 5. You can also include full competitor landing pages, last 90 days of support tickets, and survey verbatims in the same context.
How do I keep Claude from making up statistics?
Use the negative-constraint pattern from Anthropic’s prompt engineering docs: tell Claude to write [VERIFY] next to any unverified claim and to not invent statistics. In the HubSpot 2026 State of AI report, 43% of marketers say generative AI sometimes produces inaccurate information (HubSpot blog, May 2026). That risk does not go away with a bigger model; it gets cheaper to make the mistake.
What’s the cheapest way to deploy these prompts across a team? Save each prompt as a Skill in Claude. On Team and Enterprise plans, deploy it org-wide so every marketing team member gets the same brand voice and same constraints automatically. Anthropic’s pricing page (July 2026) lists “Organization wide skills deployment” as an Enterprise feature.
Can Claude handle compliance-sensitive copy like health or financial claims? It can draft, but it cannot verify. For health, finance, weight loss, and similar regulated categories, every claim must be reviewed by a human compliance officer before publication. The FTC’s AI-washing enforcement (Operation AI Comply 2024, Air AI 2026) is clear that the marketing team - not the model - owns the substantiation.
Will the EU AI Act block me from using Claude for marketing? No. The EU AI Act does not ban generative AI use in marketing. Article 50 (in force since February 2, 2025) requires transparency: EU users must be told when they are interacting with AI-generated or AI-personalized content (EU AI Act). A short disclosure line on AI-personalized assets satisfies the rule for most marketing use cases.
Sources
- Anthropic: Introducing Claude Sonnet 5 (June 30, 2026)
- Anthropic: Models Overview (July 2026)
- Anthropic: Pricing (July 2026)
- Anthropic: Prompt Engineering Overview (2026)
- Anthropic: Claude Prompting Best Practices (2026)
- Anthropic: Claude Sonnet 5 System Card (June 2026)
- Anthropic: Claude Skills Overview (2026)
- HubSpot: 2026 State of Marketing Report
- HubSpot: 8 Ways to Use AI in Digital Marketing + Examples (Updated May 2026)
- Content Marketing Institute: B2B Content Marketing Benchmarks, Budgets, and Trends (2026)
- Content Marketing Institute: Why Your Distribution Strategy Is Stunting Audience Growth (July 2026)
- Content Marketing Institute: How Marketers Should Rethink Thought Leadership (July 2026)
- Think with Google: Discovery to Decisions - Marketing in the AI Era (2026)
- EU AI Act: Regulation (EU) 2024/1689 (Official Journal, July 12, 2024)
- EU AI Act: Article 50 Transparency Rules Summary
- FTC: Keep Your AI Claims in Check (February 2023, still enforced 2026)
- FTC: Operation AI Comply (September 2024)
- FTC: FTC Sues Air AI for Deceptive Business Growth and Earnings Claims (August 2026)
- DLA Piper: FTC AI-Washing Enforcement (May 2026)
- Influencer Marketing Hub: 2026 Influencer Marketing Benchmark Report
- Meta Advertising Standards (2026)
- LinkedIn Marketing Solutions: Advertising Policies (2026)
- Google Ads Help: Advertising Policies (2026)
- [TikTok For Business: Advertising Policies (2026)](https://ads.tiktok.com/help/article/ advertising-policies)