Most ChatGPT prompt lists for marketers are junk.
They’re either written by people who don’t do marketing, or they’re scraped from old screenshots and dumped into listicles for traffic. Either way, you paste them in, get a polite generic answer, close the tab, and go back to writing the email yourself.
This is a working prompt library. Thirty prompts, organized by function, with the small adjustments that make them actually useful. They work in ChatGPT and most of them work in Claude too. We compared the two side by side in our Claude vs ChatGPT for marketing post if you’re choosing.
The shorter version of this list, with seven prompts we actually run every week, is in our weekly AI marketing prompts post. The use-case framing is in our 12 ChatGPT marketing use cases post. This post is the longer reference version when you want a categorized library you can save.
For prompt engineering technique itself, the canonical references are OpenAI’s prompt engineering guide and Anthropic’s prompt engineering docs. Read those if you want to go deeper than copy-paste.
What separates a working prompt from a useless one
Three things make most marketing prompts work better than the defaults:
- A specific role and audience, not “you are a marketing expert.”
- A real example of the input data, not a placeholder.
- A clear constraint on the output format.
Every prompt below uses all three.
You should swap the bracketed fields with your actual context. Don’t paste the brackets in. Don’t paste fake data. The output is only as good as the input.
Research and customer voice (6 prompts)
1. Customer language extraction from reviews or interviews
You are helping a marketer at [company] understand how customers describe theproduct in their own words. Below is a batch of [reviews / interview snippets /support tickets]. Extract:- The 5 phrases customers use most often- The 3 emotional drivers behind the purchase- The top 2 objections or hesitations- 5 verbatim quotes worth using on the homepage or in adsSource material:[paste reviews/snippets here]Format the output as four labeled sections. Use customer language verbatimwhere possible. Do not paraphrase the quotes.
2. Win/loss themes from sales calls
You are reviewing notes from [N] recent sales calls for [company]. Identify:- Why prospects who bought said yes- Why prospects who didn't said no- The objection that keeps repeating- One thing the marketing site is probably missingNotes:[paste call notes]Be specific. Don't list generic objections like "price." Tell me exactly whatthey said and why it matters.
3. Competitor positioning teardown
You are auditing [competitor.com] for a marketing team building positioningfor [our company]. Visit the homepage, pricing page, and one feature page.Tell me:- The single sentence positioning they're claiming- Their three loudest proof points- The audience they're clearly targeting- The audience they're clearly not targeting- One sentence we could say that they cannotKeep it under 250 words. No filler.
4. ICP refresh from current customer list
We have [N] customers across [list of segments]. Below is a sample of recentclosed-won customers with company size, industry, and use case. Identify:- The 2 segments where we win the fastest- The 2 segments where the deal cycle drags- The smallest pattern that predicts a good-fit customerSample:[paste anonymized customer data]Tell me which segment to lean into for the next quarter and why.
5. Audience research from public sources
You are researching [audience] for [our product]. Without making anything up,summarize what is publicly known or commonly discussed about this audience:- Where they spend time online- Three problems they complain about- Three tools or alternatives they currently use- Two recurring beliefs marketers about this audience get wrongNote where you're confident vs. guessing. If you don't know, say so.
6. Survey question writer
You are designing a 7-question survey for [audience] about [topic]. The goalis [specific decision]. Write questions that:- Avoid leading or loaded phrasing- Mix one open-ended question with six closed-ended- Include one question that tests willingness to pay- Include one that surfaces alternatives they consideredOutput as a numbered list with answer scales where appropriate.
Content and messaging (6 prompts)
7. Long-form to short-form repurposer
Take the article below and produce:- One LinkedIn post (under 250 words, conversational, no emojis)- Three tweet/X-post hooks- One short-form video script (45-60 seconds, single camera, no props)- One newsletter intro (3 paragraphs, hook-context-CTA)Source:[paste article]Match the source's voice. Do not add new claims that aren't in the source.
8. Headline tester
Below is a draft [landing page / blog post / ad]. Generate 10 alternativeheadlines that:- Pass the "would a stranger click this on LinkedIn" test- Avoid clickbait- Make the specific promise of the page (no curiosity gaps)- Range from descriptive to provocativeDraft:[paste current headline + first 200 words of body]Rank them from most to least likely to convert and explain the top 3.
9. CTA rewriter
Below are [N] CTAs from across our site. Rewrite each to:- Be 2-5 words- Start with a verb- Match the visitor's intent at that point on the page- Avoid "Learn more" and "Get started"Current CTAs and their page contexts:[paste]Output as a table with the original, the rewrite, and a one-line rationale.
10. Brand voice extractor
Below are [N] published pieces from [brand]. Identify:- 5 patterns that show up across the writing (sentence rhythm, vocabulary, rhetorical moves)- 3 things this brand never says- 2 quirks a junior writer would copy wrong if they imitated this voiceSource:[paste 3-5 sample pieces]Output as a brand voice doc a new writer could use as a checklist.
11. Cold email rewriter
Below is a cold email draft. Rewrite it to:- Open with a specific observation about the recipient, not "I noticed your company is doing X"- Get to the actual ask in the second paragraph- Be under 90 words total- End with a question that's easy to answer in one sentenceOriginal:[paste draft]Recipient context: [role, company, what they likely care about]
12. Internal linking auditor
Below is a draft blog post and a list of our published articles with theirURLs. Suggest where to add internal links:- Identify 4-8 places in the draft where an internal link would be helpful to the reader (not stuffed)- For each, name the existing post to link to- Write the exact anchor textDraft:[paste]Existing posts:[paste sitemap or post list]
SEO (5 prompts)
13. Search intent classifier
For each keyword below, classify:- Search intent (informational, commercial, transactional, navigational)- The dominant content type ranking on the SERP (listicle, how-to, comparison, product page, video)- Whether our [company] should target it now (yes/no/later) and whyKeywords:[paste 10-30 keywords]Output as a table. Be honest about which we shouldn't target as a [DR X] site.
14. Outline from a target keyword
You are writing a blog outline for the keyword "[keyword]". The reader is[audience] and the goal is [conversion]. Look at the top 5 ranking pagesmentally and produce an outline that:- Covers what the SERP requires to rank- Includes 2 sections the SERP is missing- Has a working title that matches search intent- Suggests one chart, one table, or one diagram to addOutput as H2/H3 outline with 1-2 sentences per section.
15. Meta description rewriter
Below are [N] meta descriptions from our site. For each, rewrite to:- Stay under 155 characters- Include the focus keyword once, naturally- Include a benefit the click-through reader will care about- Avoid "Discover," "Unlock," "Learn how to"Current:[paste page URL, title, current meta]Output as: URL | Old | New | One-line rationale.
16. Topic cluster mapper
We have published [N] posts (list below). Group them into 3-5 topic clustersand identify:- The natural pillar post for each cluster (write a working title if it doesn't exist)- The 2-4 supporting posts in each cluster- The 3 internal links each pillar should add to its supporters- Any orphan posts that don't fit any clusterPosts:[paste post list with URLs]
17. SERP gap finder
For the keyword "[keyword]", the top 5 results are:[paste URLs and titles]Without making anything up, identify:- 3 questions all 5 articles answer- 2 questions only some answer- 2 questions none of them answer that the searcher likely has- 1 angle a competitor could take that none have takenOutput as a brief, under 300 words.
Paid and landing pages (5 prompts)
18. Ad concept brief from positioning
We are running [paid channel] ads for [product] targeting [audience]. Ourpositioning is "[one-sentence positioning]." Generate:- 3 ad concepts (problem-led, identity-led, social-proof-led)- For each: headline, body, CTA, image direction (one sentence)- One concept that breaks our usual pattern and why it could workConstraints:[paste platform character limits]No clichés. No "Imagine if..."
19. Landing page diagnostic
Below is the copy from our landing page for [product/offer]. Diagnose:- Whether the hero promise is specific enough to act on- Where the page asks for trust before earning it- Which section the bored visitor will actually scroll past- One thing on the page that should not be therePage:[paste full copy]Be specific. No vague advice like "improve clarity."
20. Objection-handling section
Below are the top 5 objections from [audience] for [product/offer]. Write alanding page section called "Common questions" that handles them:- One question per objection in customer language- One short answer per question (2-3 sentences)- No defensive tone; no apologies for the priceObjections:[paste]
21. Ad-to-landing-page consistency check
Below is an ad and the landing page it links to. Check:- Whether the page's hero matches the ad's promise- Whether the visitor will feel like they landed on the right page within 3 seconds- Where the page introduces something the ad didn't promise- One change to the page that would lift the conversion rateAd:[paste]Page:[paste]
22. Pricing page copy
Write the copy for a pricing page with [N] tiers. Each tier needs:- Tier name (3 words max)- One-line audience match (who this is for)- 4-6 included features- One sentence about who should NOT pick this tierInputs:- Tiers: [list with prices]- Audience: [paste ICP notes]- Product: [paste short description]No "Most popular." No fake urgency.
Email (4 prompts)
23. Welcome sequence outline
Write a 5-email welcome sequence for new [list/product] signups. The goal is[specific outcome]. For each email:- Subject line (under 45 characters)- Preview text (under 90 characters)- Send timing (relative to signup)- One-paragraph body summary (not the full draft)- The single CTAAudience and product context:[paste]The sequence should escalate from "you're welcome here" to one specific ask.
24. Subject line tester
Below is an email body. Generate 8 subject lines:- 4 specific (named feature, named outcome, named number)- 2 curious without being clickbait- 2 plain (almost boring, signal authenticity)Body:[paste]Rank them on likely open rate for [audience] and explain the top two.
25. Re-engagement email
Write a re-engagement email for subscribers who haven't opened anything in90 days. The email should:- Acknowledge the gap honestly- Offer something specific worth opening (not "we miss you")- Make staying or leaving equally easy- Be under 100 wordsAudience: [paste]Product: [paste]
26. Newsletter intro
Write a 3-paragraph newsletter intro for this week's edition. The hook is[topic/event]. The link of the week is [link]. The reader is [audience].- Paragraph 1: a specific observation about the week (no "in today's fast-paced world")- Paragraph 2: why this matters to the reader- Paragraph 3: a tight CTA into the rest of the newsletterVoice: [paste 1-2 examples of past intros]
Analytics and reporting (4 prompts)
27. Weekly marketing report draft
Below are this week's metrics for [company]. Draft a weekly report for[audience]:- Headline takeaway (one sentence)- 3 things that moved up or down meaningfully- One hypothesis for the biggest change- 2 actions for next weekMetrics:[paste]Match the format of last week's report (paste below if relevant).
28. A/B test result interpreter
We ran an A/B test on [page/email/ad] comparing [variant A] vs [variant B].Results below. Tell me:- Whether the result is real (call out sample size and lift size)- The single most likely reason the winner won- Whether the winner generalizes or is specific to this audience/context- What to test nextResults:[paste raw data]Don't say "more testing is needed" without naming the next test.
29. Funnel diagnostic
Below is our funnel data from the last 30 days. Identify:- The single biggest drop-off relative to industry benchmarks- The most likely reason for that drop-off- One change that would test the hypothesisFunnel:[paste stages with conversion rates]Be specific. Don't say "improve the landing page" without naming the change.
30. Channel comparison
Below is performance data across [channels] over [period]. Compare them on:- Cost per acquisition- Customer quality (downstream metrics if provided)- Trajectory (improving, flat, declining)- Capacity left to scaleData:[paste]Tell me which channel to double down on, which to maintain, and which tocut. Defend each call in one sentence.
How to actually use this list
A library of 30 prompts is only useful if you use it.
We do three things to keep these from becoming dead weight:
First, save them somewhere you can paste in two seconds. A snippet manager, a starred Notion page, a single doc. The prompts you have to dig for are prompts you don’t run.
Second, fill in the brackets every time. Generic prompts produce generic outputs. The “swap your real data in” step is where the leverage lives.
Third, treat the output as a draft, not an answer. Every prompt above is meant to start a thought, not finish one. The output goes in front of a human who knows the business. That is the whole point.
For the longer operating model that sits behind these prompts, our complete AI marketing stack post covers what the rest of the system looks like. For the team-shape question of who runs all this, our 2-person AI marketing team post is the companion read. For the deeper agent-based version of the same workflows, our Hermes Agent for marketing automation post covers what changes when the prompts run on a schedule.
If your team wants help building this kind of prompt-and-agent operating model, our services page explains how we work, and you can get in touch here.
FAQ
Do these prompts work in Claude as well as ChatGPT? Most of them, yes. Claude tends to write slightly more carefully and is better at refusing to make things up; ChatGPT tends to be punchier and more willing to fabricate confidently. For research and synthesis prompts (1-6, 13-17, 27-30) we lean Claude. For copy and creative prompts (7-12, 18-26) the two trade blows depending on voice. We compared them directly in our Claude vs ChatGPT for marketing post.
How long should a marketing prompt be? Long enough to remove ambiguity, short enough that you can scan it before pasting. Most of the prompts above are 80-150 words. Anything shorter usually produces generic output. Anything longer usually means the prompt is doing the thinking your brief should have done.
Should I use a prompt library tool or just save these in a doc? Either works. The risk with prompt library tools is the prompts become someone else’s product priorities, not yours. A flat doc you control is harder to outgrow. Pick the one that gets the prompts in front of you in under 5 seconds.
What’s the single biggest mistake marketers make with these prompts? Pasting them with the brackets still in. The brackets are placeholders. The prompt only works when you replace them with the actual brand, audience, data, and constraints. A prompt run with placeholders gives you a Wikipedia article. A prompt run with real data gives you something worth editing.