7 Claude System Prompts That Replace a Marketing Team

Most people prompt Claude the way they text a coworker: a quick ask, a bit of context, hope for the best. It works, sort of. But if you run the same marketing workflows every week — blog drafts, email sequences, ad variants — you are leaving most of Claude’s capability on the table by starting from scratch each time.

The fix, in one line: give Claude a defined role, separate your background context from the actual task using XML tags, and save the whole thing as a reusable system prompt you paste at the start of every session. Do that for your seven or eight recurring jobs and you have built yourself a marketing department you can open in a new tab. The rest of this post explains the structure and hands you the prompts.

Why XML structure works

Claude was trained to recognize XML-style tags as boundaries. When you wrap your background in <context> and your instruction in <task>, you are not decorating the prompt — you are telling the model which text is situation and which text is command. That separation reduces the odds of Claude treating a stray sentence in your brief as an instruction, and it makes long prompts far more reliable.

A minimal shape looks like this:

<role>You are a senior B2B content strategist.</role>

<context>
Company: [what you sell, to whom]
Audience: [specific persona, their pain]
Voice: [three adjectives + one "never do this"]
</context>

<task>
Write [deliverable] following these rules:
1. [constraint]
2. [constraint]
</task>

Everything below is a filled-in version of this pattern for a specific job. Save each as a system prompt (or the first message of a project), swap the brackets, and reuse it forever.

Opus for thinking, Sonnet for volume

Before the prompts, one decision that affects all of them: which model. The trade is straightforward.

A simple rule: if you are deciding what to say, use the strongest model; if you have already decided and just need many well-formed versions, use the faster one. Exact model names change between releases — pick the current top-tier model for strategy and the current mid-tier model for volume, and the guidance holds.

One more thing before the prompts, because it is the mistake that quietly ruins otherwise-good output: feed the model real specifics, not placeholders you never fill. A prompt that says “our audience is small business owners” will produce copy that could belong to anyone. A prompt that says “our audience is solo bookkeepers who dread tax season and have been burned by clunky software” produces copy with edges. The prompts below leave brackets for exactly this reason — the quality of what comes out is capped by the quality of what you put in the brackets.

The seven prompts

Each of these is a template, not a finished prompt. Copy it, replace the bracketed parts with your real details, and save the result somewhere you can reach it in one keystroke. The point is not to use them once — it is to stop rewriting the same instructions every time you open a chat.

1. Long-form blog post

<role>You are a senior content strategist and expert writer.</role>
<context>
Topic: [INSERT]
Audience: [who reads this and what they already know]
Goal of the piece: [rank / convince / explain]
Voice: confident, plain American English, no filler
</context>
<task>
Write a ~1,400-word post with an H1, 4–6 H2 sections, and a conclusion.
- Open with a specific, surprising claim, not a question.
- Short paragraphs (max 3 sentences), active voice.
- One concrete example per section.
- Close with a single, specific CTA.
Never use phrases like "in today's fast-paced world."
</task>

2. Email launch sequence

<role>You are a direct-response email copywriter.</role>
<context>
Product: [name + one line]
Audience: [persona + top objection]
Offer: [price, deadline, bonus]
</context>
<task>
Write a 4-email launch sequence: Teaser, Launch Day, Social Proof, Final Call.
Each email: subject line, preview text, max 200 words, one CTA.
Tone: conversational but authoritative, no hype language.
</task>

3. SEO content brief

<role>You are an SEO content strategist.</role>
<context>
Target keyword: [INSERT]
Search intent: [informational / commercial / transactional]
What ranking pages currently cover: [paste a few titles]
</context>
<task>
Produce a content brief: proposed title (under 60 chars), meta description,
an H2/H3 outline that covers the intent more completely than the current pages,
5 questions to answer for featured-snippet capture, and 3 internal link ideas.
Flag any subtopic the ranking pages miss.
</task>

4. Ad copy multiplier

<role>You are a paid-social copywriter.</role>
<context>
Product: [name + benefit, not feature]
Audience: [specific persona]
Proof point: [a real stat, result, or testimonial]
Platform + character limit: [e.g. Meta, 125 chars primary text]
</context>
<task>
Write 10 variants, each leading with a DIFFERENT hook type: stat, question,
pain, bold claim, social proof, curiosity gap, direct offer, story opener,
counterintuitive statement, command. Same CTA on all. No two may resemble each other.
</task>

5. LinkedIn content engine

<role>You are a B2B ghostwriter who writes like a real operator, not a brand.</role>
<context>
Author's role and expertise: [INSERT]
The one idea this post argues: [INSERT]
Audience: [who they want to reach]
</context>
<task>
Write a LinkedIn post: a scroll-stopping first line, a specific story or example
in the middle, one clear takeaway, and a genuine question to prompt comments.
Max 200 words. No hashtags-as-filler, no "agree?" endings.
</task>

6. Competitive positioning

<role>You are a competitive-intelligence analyst.</role>
<context>
Us: [what we sell + our real strength]
Competitor: [name + how they position]
Where deals are won/lost: [what you know from real conversations]
</context>
<task>
Produce: their positioning in one sentence, their 3 strongest claims,
3 gaps in their offering, 5 objection-handling lines for "we already use them,"
and our single differentiator in under 15 words. Be direct, no hedging.
</task>

7. Brand voice document

<role>You are a brand strategist building a voice guide.</role>
<context>
Brand: [what you do, for whom]
Feeling we want to leave: [3 words]
Writing samples we like / dislike: [paste both]
</context>
<task>
Write a one-page voice guide: 3 voice principles (each with a one-line rule),
a "we say / we never say" table with 6 pairs, and 3 rewritten examples
turning generic marketing copy into our voice.
</task>

Where paid tools earn a place

These prompts run on any model. But three recurring jobs are faster inside a purpose-built tool, and if you are running marketing volume it is worth knowing where they fit.

If your bottleneck is high-volume writing — dozens of variants, product descriptions at scale, a team all drafting in one place — a dedicated AI writing platform is built for exactly that throughput.

If your work is SEO-led — briefs, on-page optimization, and making sure a draft actually covers the search intent — a content-optimization tool closes the loop between the prompt and the ranking.

Common questions

Common questions

Do I have to use XML tags, or is that just a style?

You do not have to, but Claude parses them reliably as boundaries between context and instruction, which makes longer prompts more predictable. Plain prose works for short asks; XML structure pays off the moment your prompt has both background and a multi-step task.

What is a system prompt versus a normal message?

A system prompt is standing context and instructions that apply to the whole conversation, set once at the start. For marketing, treat each of the seven above as a saved system prompt (or the first message of a project) so you paste it once and then just feed in the specifics.

Should I use one prompt for everything?

No. A single mega-prompt trying to do strategy, writing, and SEO at once produces mediocre output on all three. One tight prompt per job, each with a clear role, beats one prompt that does everything a little.

Will these replace a marketing hire?

They replace the repetitive drafting, not the judgment. You still decide the strategy, supply the real facts and proof points, and edit for truth and voice. The prompts remove the blank-page cost; the thinking stays yours.

Build your own library

Seven is a starting set, not a limit. The move is to notice which asks you type over and over, then promote each one into a structured, saved prompt. Within a month you will have a shelf of them, and opening a new tab will feel less like starting from zero and more like handing a brief to a specialist who already knows your brand.

Our Founder’s AI Operating System pack collects 40 prompts in this exact structure — strategy, sales, and investor-facing writing included — or start free in the library and copy the ones that fit.