From Brief to Landing Page: 11 Copywriting and Marketing Prompts That Actually Ship Content

From Brief to Landing Page: 11 Copywriting and Marketing Prompts That Actually Ship Content

Most prompt lists you find online are written by people who have never had to ship a landing page by Friday. They give you "write a blog post about X" and call it a day. That is not how content work happens. Real marketing work looks like this: a vague brief from a stakeholder, a tone-of-voice doc nobody reads, three audience segments, and a deadline. The prompts below are the ones that survive that reality.

I've been running these in daily production work — B2B SaaS, e-commerce, and agency projects. Each one is written to be pasted as-is, with the placeholders swapped for your context. Where a prompt depends on a known technique (like chain-of-thought or role prompting), I name it so you can adapt it instead of cargo-culting it. For background on why role and structure matter, OpenAI's own prompt engineering guide (platform.openai.com/docs/guides/prompt-engineering) is still the cleanest reference, and Anthropic's prompt library (docs.anthropic.com) has good examples of the same patterns applied to long-form writing.

One rule before we start: a prompt is not a magic spell, it is a specification. The more constraints you give — audience, format, banned words, length, examples — the less you rewrite afterward. Every prompt below is built around that principle.

1. The Brief Compressor: turn a messy stakeholder message into a real brief

Task it solves: a product manager sends you a wall of text in Slack. You need a structured creative brief before you write a single word.

Prompt:

You are a senior content strategist. Below is a raw, unstructured request from a stakeholder.
Convert it into a creative brief with these sections:
1. Business goal (one sentence)
2. Target audience (role, pain, current alternative)
3. Core message (one sentence, no jargon)
4. Proof points available (list only what is stated; mark gaps as MISSING)
5. Desired action (what the reader should do)
6. Constraints (length, channel, deadline, compliance)
7. Open questions to ask the stakeholder (max 3)

Rules: do not invent facts. If something is not in the source text, write MISSING.
Raw request:
"""{{STAKEHOLDER_MESSAGE}}"""

Example input: "hey can we do something for the new pricing page, sales keeps saying people don't get the new tiers, maybe a blog post or an email, we launch next month"

Expected output: a brief where the business goal is "reduce pre-sales confusion about the new pricing tiers," the audience is "evaluating buyers who bounced from the pricing page," and the open questions are about which tier is actually confusing people.

Why it works: this is role prompting plus a hard anti-hallucination rule. Marking gaps as MISSING is the single most useful trick in marketing prompting — it stops the model from quietly filling in numbers you'll later have to defend in a review.

2. The Tone-of-Voice Extractor: build a style guide from your own best content

Task it solves: you have 20 published articles that sound right, but no written tone-of-voice doc. Instead of inventing rules, extract them.

Prompt:

Below are 5 samples of our published content. Analyze them and produce a tone-of-voice guide:
- Voice in 3 adjectives, each justified with a quote from the samples
- Sentence length pattern (short/medium/long, with examples)
- Person and address (we/you vs. I/you), with evidence
- Vocabulary we use and vocabulary we avoid
- Formatting habits (headers, lists, emoji, exclamation marks)
- 5 "do" and 5 "don't" rules rewritten from the evidence
Do not describe generic best practices. Only what is visible in the samples.
Samples:
"""{{PASTE_3_TO_5_ARTICLES}}"""

Example output line: "We avoid exclamation marks entirely — zero occurrences across 4,200 words — so 'Great news!' breaks our voice."

Why it works: it inverts the usual process. Instead of writing a style guide and hoping writers follow it, you derive the guide from content that already converted. Paste the output into your team wiki and it becomes enforceable.

3. The Landing Page Skeleton: problem → mechanism → proof → offer

Task it solves: writing landing copy from scratch, where the hardest part is the order of arguments, not the words.

Prompt:

You are a direct-response copywriter. Build a landing page skeleton for the product below.
Structure it in 4 blocks: Problem, Mechanism, Proof, Offer.
For each block give: a headline (max 12 words), 2-3 supporting bullets, and one sentence of reasoning for why this argument comes here.
Then add: 3 alternative hero headlines using different angles (pain, outcome, curiosity).
Product: {{PRODUCT}}
Audience: {{AUDIENCE}}
Primary objection we hear: {{OBJECTION}}
Banned words: {{e.g. revolutionary, seamless, cutting-edge}}

Expected result: a page outline you can hand to a designer the same day, plus headline variants for A/B testing.

Why it works: direct-response structure (problem, mechanism, proof, offer) is a known framework, so the model has strong priors for it. Asking for the reasoning per block forces it to justify placement — which is where you catch weak logic before it reaches the designer.

4. The Objection Miner: pull real objections out of reviews and support tickets

Task it solves: you need copy that answers real fears, not imagined ones.

Prompt:

You are a customer research analyst. Below are raw customer messages (reviews, tickets, chat logs).
1. Extract every distinct objection or hesitation, grouped by theme.
2. For each theme, quote the original text (verbatim, max 20 words).
3. Rate frequency: how many messages mention it.
4. Suggest one headline and one FAQ answer that neutralizes each theme.
Do not paraphrase quotes — copy them exactly.
Messages:
"""{{PASTE_RAW_MESSAGES}}"""

Example output: Theme "hard to set up" — 6 mentions — quote: "took me two evenings to connect it to my CRM." Suggested FAQ: "How long does setup actually take?"

Why it works: verbatim quotes are the raw material for testimonials and FAQ sections, and they are the only part of AI-assisted copywriting that is genuinely unfakeable. Frequency counts turn "I think people worry about setup" into a data-backed decision about what goes above the fold.

5. The Email Chain Architect: a 5-email sequence with one job per email

Task it solves: onboarding or nurture sequences where most drafts repeat the same pitch.

Prompt:

Design a 5-email sequence for {{GOAL}} (e.g. trial-to-paid conversion).
For each email provide:
- Day number and send trigger
- One single job for this email (only one)
- Subject line (max 45 characters) + 2 alternatives
- Body outline: opening line, argument, proof, CTA
- The reason this email exists in the sequence and what breaks if you remove it
Rules: no email may repeat the CTA of the previous one. Vary the angle: utility, social proof, objection, urgency, and a plain-text personal note.

Expected result: a sequence where email 1 delivers value, email 2 proves it with a case, email 3 handles the objection, email 4 creates a deadline, and email 5 sounds like a human.

Why it works: "one job per email" is a discipline most humans skip, and models skip it too unless forced. The "what breaks if you remove it" line is the cheapest way to catch filler emails before they get scheduled.

6. The Tone Shifter: rewrite one text for five channels without losing the message

Task it solves: one core message, five channels — LinkedIn, email, landing page, ad, and push notification.

Prompt:

Take the source text below and produce 5 versions:
1. LinkedIn post (max 150 words, hook in first line, no hashtags)
2. Cold email (max 90 words, no flattery opener)
3. Landing page paragraph (max 60 words, benefit-led)
4. Paid ad (max 30 words, one concrete claim)
5. Push notification (max 12 words)
Keep the core claim identical in all five. Change only length, register, and CTA.
After each version, state in one line what you changed and why.
Source: """{{CORE_MESSAGE}}"""

Expected result: five ready-to-publish variants that all say the same thing, plus a short changelog you can paste into your content calendar.

Why it works: the "keep the core claim identical" constraint is the whole point. Without it, models drift — by variant four you're advertising something you don't sell. The per-version changelog makes the drift visible.

7. The Headline Multiplier: 20 variants across 5 psychological angles

Task it solves: you have a decent headline and need test variants that differ in angle, not just wording.

Prompt:

Generate 20 headlines for the offer below, grouped by angle:
- Pain (what it costs to ignore the problem)
- Outcome (what life looks like after)
- Curiosity (an open loop the reader wants closed)
- Social proof (who else did this)
- Contrarian (challenge a common belief)
Constraints: max 12 words each, no clickbait lies, no exclamation marks, include the number or concrete noun where possible.
Mark the 3 you'd test first and explain why in one line each.
Offer: {{OFFER}}
Audience: {{AUDIENCE}}

Expected result: a test matrix rather than a list. Grouping by angle is what makes the set useful — you can test one angle per week instead of random word swaps.

Why it works: copy testing fails when variants are too similar. Forcing five distinct psychological angles gives you real experimental separation, which is the difference between a headline test and a coin flip.

8. The SEO Cluster Planner: one pillar, ten supporting intents

Task it solves: keyword lists are easy to get and hard to structure. This turns a flat list into a content cluster.

Prompt:

You are an SEO content strategist. Below is a keyword list for {{TOPIC}}.
1. Identify the pillar keyword (highest intent, broadest).
2. Group the rest into clusters by search intent: informational, commercial, transactional, navigational.
3. For each cluster, propose article titles mapped to a specific intent.
4. Note internal linking: which articles should link to the pillar and which to each other.
5. Flag keywords that look like duplicates or cannibalization risks.
Do not invent search volumes. If you don't know a number, say "volume unknown."
Keywords: {{PASTE_KEYWORD_LIST}}

Example output: pillar "CRM integration," cluster "how to connect X to Y" (informational, links up to pillar), cluster "best CRM integration tools" (commercial, links across).

Why it works: cannibalization detection is the part humans forget and models are good at, because it's a pure text-similarity problem. The "volume unknown" rule keeps you from building a strategy on invented numbers.

9. The Ad Creative Factory: 10 concepts with visual notes

Task it solves: you need ad concepts, not just headlines — including what the image should show.

Prompt:

Generate 10 ad concepts for {{PRODUCT}} targeting {{AUDIENCE}} on {{PLATFORM}}.
For each concept give:
- Headline (max 8 words)
- Primary text (max 40 words)
- Visual direction: what the image or video frame shows, in one sentence
- The emotion it targets
- Why it fits this platform's format specifically
Avoid: stock-photo clichés, generic smiling people, abstract gradients.

Example output: Concept 4 — Headline: "Setup took 11 minutes. Here's the timer." — Visual: a phone screen recording with a visible stopwatch overlay.

Why it works: asking for visual direction turns a text prompt into a creative brief. The banned-cliché list is doing real work here — without it you get the same three stock-photo descriptions every time.

10. The Case Study Builder: from raw notes to a structured story

Task it solves: you have interview notes and metrics scattered in a doc, and the case study keeps getting postponed.

Prompt:

Turn the raw notes below into a case study with this structure:
1. Context: who the client is and what they were doing before
2. Problem: the specific bottleneck, with numbers if present
3. Approach: what was actually done, in chronological order
4. Results: only metrics present in the notes; mark everything else as "not measured"
5. Quote: pull one usable client quote verbatim, or write "no quote available"
6. One-sentence takeaway for a reader in a similar situation
Length: 500-700 words. Do not add adjectives the notes don't support.
Notes: """{{RAW_NOTES}}"""

Expected result: a publishable draft where every claim traces back to your notes — which is exactly what E-E-A-T reviewers and legal teams want to see.

Why it works: case studies rot because writers pad them. The "not measured" rule and the verbatim-quote requirement make padding impossible: the model has nowhere to hide a fabricated 40% improvement.

11. The Pre-Publish Auditor: catch weak copy before a human does

Task it solves: the final pass before publishing, when you're too close to the text to see its problems.

Prompt:

Act as a skeptical senior editor. Review the draft below and report:
1. Claims that are not supported by evidence in the text (list each)
2. Sentences longer than 25 words (quote them)
3. Banned words present: {{BANNED_LIST}}
4. Places where the reader's next action is unclear
5. Three specific cuts that would make it tighter
6. One thing the draft does well — be specific, not polite
Do not rewrite the text. Only diagnose.
Draft: """{{DRAFT}}"""

Expected result: a punch list you can work through in 15 minutes instead of re-reading the draft five times.

Why it works: the "do not rewrite, only diagnose" instruction is critical — otherwise the model silently replaces your voice with its own. Diagnosis keeps you in control of the final text, which is the only version that should ever carry your byline.

Stacking the prompts

These eleven prompts are not meant to be used in isolation. The real leverage comes from chaining them: Brief Compressor → Objection Miner → Landing Page Skeleton → Tone Shifter → Pre-Publish Auditor is a complete pipeline from a Slack message to a published page. In practice, teams that run this kind of chain report the biggest time savings not in drafting (where AI is obvious) but in the review loop — because the auditor catches the same issues a senior editor would, before the draft reaches them.

A note on honesty: AI will not replace your judgment about what to say, only the labor of saying it. The prompts above are deliberately full of constraints — MISSING, not measured, verbatim quotes, banned words — because those constraints are what keep the output defensible. Use them as specifications, keep your own editorial veto, and you get the speed of a content team without the headcount. Start with prompt #1 today: take the messiest stakeholder message you received this week and run it through. You'll have a brief in two minutes and a much better Friday.