Prompt Engineering for Marketers: 20 Prompts That Actually Work

If you have spent 30 minutes arguing with ChatGPT to get a usable blog intro, you already know the core lesson of prompt engineering. The model is not reading your mind. It is reading your text. Ambiguity in the prompt becomes mush in the output. Precision in the prompt becomes craft in the output.

This post is the working library I use with clients and inside the FastStrat team. Every prompt here has been tested against real SMB marketing work. Some came from reviewing more than 400 client sessions. Others from copying patterns from top-performing agency briefs and translating them into model instructions. None of them are magic. All of them outperform “write me a blog post about X”.

If you are new to AI for marketing, start with the AI marketing playbook for SMBs to get the strategic frame. If you want a side-by-side on the models themselves, see ChatGPT vs Claude vs FastStrat for marketing.

1. The framework: every good prompt has 5 parts

Before the 20 prompts, the framework. Every prompt in this library is built on the same skeleton. Once you internalize it, you can write your own.

1. Role. Tell the model who it is. “You are a senior B2B SaaS copywriter with 10 years of experience writing for developer audiences.” Role sets the vocabulary, the references, and the implicit rules the model will follow.
2. Context. Tell the model what it needs to know. Brand, audience, product, prior work, success criteria. This is where 80% of output quality lives. A great prompt with no context produces generic work. A decent prompt with rich context produces specific, useful work.
3. Task. Tell the model exactly what to do. One verb, one deliverable. “Write three subject-line variants.” Not “help me with email marketing.”
4. Constraints. Tell the model what not to do. Word counts, tone bans, structural rules, things it should never include. Constraints do more work than adjectives.
5. Format. Tell the model what the output should look like. Table, bullet list, JSON, three numbered options, a headline plus three-line body. Explicit format produces clean artifacts you can actually use.

A one-sentence prompt has zero of these. A professional prompt has all five. Every template below follows this skeleton. Read one out loud and you will hear it.

One more rule. Put the most important instruction first and last. Models weight the beginning and the end of prompts more heavily than the middle. If there is a constraint that must not be violated, say it early and repeat it at the end.

Competitor research prompts (1 to 3)

Prompt 1. Competitor positioning teardown

What it does. Extracts the positioning, value propositions, and implicit target audience from a competitor’s homepage text.

When to use it. At the start of any competitor analysis, before you write your own positioning. Part of the 7-step method in how to do competitor analysis for small business.

You are a senior positioning strategist with 15 years of B2B and B2C experience.

Here is the full text of a competitor's homepage, between triple quotes:
"""
[PASTE FULL HOMEPAGE COPY HERE]
"""

Analyze it and return:
1. The primary positioning statement in one sentence (what they say they are)
2. The underlying positioning in one sentence (what they actually are, given the copy)
3. The top three value propositions in order of prominence
4. The implied target audience (role, company size, industry, pain)
5. The three biggest gaps or weaknesses in the positioning
6. One specific angle a smaller competitor could attack

Constraints: no generic marketing jargon, no "innovative" or "cutting-edge", quote
specific phrases from the text to back each claim.

Output as a markdown table with columns: Observation, Evidence (quoted), Implication.

Example output. A table with rows like “Primary positioning: ‘the all-in-one HR platform’ | Evidence: ‘manage hiring, payroll and benefits in one place’ | Implication: they are competing on breadth, not depth. Attackable with a focused single-function product.”

Tip. Run the same prompt on five competitors and paste the five tables into a single doc. Patterns across the five is where your own positioning gets sharp.

Prompt 2. Competitor content gap analysis

What it does. Given a competitor’s list of blog post titles, identifies the topic clusters they dominate and the gaps you can attack.

When to use it. SEO planning, topic cluster design, content calendar input.

You are a senior SEO content strategist.

Below is the list of the last 50 blog post titles from a competitor in
the [INDUSTRY] space, serving [ICP].

[PASTE 50 TITLES, ONE PER LINE]

Do this:
1. Group the titles into 4 to 7 topic clusters, name each cluster.
2. For each cluster, estimate the apparent search intent (informational,
   commercial, transactional).
3. List three topic clusters they are clearly NOT covering that would
   be natural for this ICP.
4. For each gap cluster, propose five specific blog post titles a
   competitor could publish to attack.

Format as four sections: Clusters | Intent map | Gaps | Attack titles.
Do not recommend generic "ultimate guide to X" titles. Each title must
be specific enough that only one person could write it well.

Tip. Feed the output back into the model with “Now rank the 15 attack titles by likely difficulty-to-rank assuming we are a new domain” to get a prioritized list.

Prompt 3. Feature war-gaming

What it does. Simulates how a competitor would respond to a feature you are considering launching.

You are a product marketing lead at [COMPETITOR NAME], a company that
positions itself as [POSITIONING] and serves [ICP].

We are considering launching the following feature:
[DESCRIBE FEATURE IN 3 SENTENCES]

From the competitor's point of view, answer:
1. What is the most likely counter-launch within 90 days?
2. Which existing feature do they double down on to neutralize ours?
3. What narrative do they push in content and sales to reframe it?
4. What is the one message our buyers might believe from them that
   undercuts us?

Be honest and specific. Do not be polite. Use the competitor's real
brand voice if you have context on it.

Tip. After the first run, add: “Now write the blog post title they would publish within 30 days.” That gives you the headline to pre-empt.

Copywriting prompts (4 to 7)

Prompt 4. Value proposition generator

What it does. Turns a product description plus target audience into 10 value proposition candidates using proven copywriting frameworks.

You are a direct-response copywriter trained on the work of Eugene Schwartz,
Gary Halbert, and Joanna Wiebe.

Product: [DESCRIBE PRODUCT IN 3 SENTENCES]
Target audience: [ROLE, COMPANY SIZE, INDUSTRY, CORE PAIN]
Primary competitor alternative: [WHAT THEY USE TODAY]

Write 10 value propositions using the following mix:
- 3 using the "for [audience] who [pain], [product] is the [category]
  that [benefit], unlike [alternative]" framework
- 3 "only" statements (we are the only X that Y)
- 2 outcome-led (two lines, before/after)
- 2 contrarian (name the industry's default wisdom and reject it)

Constraints: no "leverage", "unlock", "seamless", "revolutionize",
"game-changer". Specific verbs over abstractions. One value proposition
per line, numbered.

Tip. Test the top three in ad copy for two weeks. The one with the highest click-through rate becomes your homepage hero. For the full methodology on writing the winning version, read how to write a value proposition that sells.

Prompt 5. Headline matrix

What it does. Generates a matrix of headlines across emotional registers so you can A/B test systematically rather than randomly.

You are a senior conversion copywriter.

Topic: [ARTICLE OR LANDING PAGE TOPIC]
Primary keyword: [KEYWORD]
Target reader: [ROLE, PAIN, STAGE OF AWARENESS]

Write 15 headlines, 3 each in these five registers:
1. Problem-focused (name the pain)
2. Outcome-focused (name the result)
3. Curiosity (implicit gap, ethical not clickbait)
4. Contrarian (reject a common belief)
5. Specific number (measurable promise)

Constraints: max 70 characters each, include the primary keyword in at
least 8 of the 15, no questions as headlines, no colons used as
pause-tricks.

Output as a 5x3 markdown table.

Tip. For the full method on deciding which register to ship, read how to write headlines that convert.

Prompt 6. Brand voice cloning

What it does. Forces the model to produce output in your actual brand voice instead of the generic “helpful assistant” voice.

You are a copywriter. Below are three samples of our actual brand voice,
between triple quotes. Study them before writing.

Sample 1:
"""[PASTE 200-400 WORDS OF YOUR REAL COPY]"""

Sample 2:
"""[PASTE 200-400 WORDS]"""

Sample 3:
"""[PASTE 200-400 WORDS]"""

In one paragraph, describe the voice in concrete terms (sentence length,
word choice, rhetorical moves, what it avoids). Then write [DELIVERABLE]
using that voice exactly.

Constraints: if the samples avoid a certain word or phrasing, you must
also avoid it. If the samples use self-deprecation, use it. If the
samples use specific industry references, use them.

Tip. Save the “voice description paragraph” the model writes after the first run. Reuse it at the top of every future prompt. That one paragraph is worth more than any “professional tone” instruction.

Prompt 7. Landing page full draft

You are a landing page copywriter working in the long-form direct response tradition.

Inputs:
- Product: [DESCRIPTION]
- Target audience: [ICP + AWARENESS STAGE: unaware, problem-aware,
  solution-aware, product-aware, most-aware]
- Primary objection: [THE ONE THING THAT STOPS THEM FROM BUYING]
- Proof assets available: [CUSTOMER QUOTES, DATA POINTS, LOGOS]

Write a full landing page with these sections, in order:
1. Hero headline + sub (max 15 + 30 words)
2. Problem section (3 paragraphs naming the pain)
3. Solution section (introduce the product as the answer)
4. How it works (3 steps)
5. Three objections + three rebuttals (use the proof assets)
6. Social proof block
7. Closing CTA with one-line urgency reason

Constraints: no "unlock", "leverage", "seamless". Each section must be
useful if read standalone. The objection section must quote real proof.

Tip. Pick the awareness stage honestly. Writing solution-aware copy for an unaware audience is the number one reason landing pages under-convert.

Strategy prompts (8 to 10)

Prompt 8. Annual marketing plan outline

You are a senior marketing strategist who has built 50+ annual plans
for SMBs in the $2M to $20M revenue range.

Company context:
- Industry: [INDUSTRY]
- Revenue last year: [NUMBER]
- Revenue goal this year: [NUMBER]
- Team: [SIZE AND COMPOSITION]
- Current channels working: [LIST]
- Current channels not working: [LIST]
- Marketing budget: [NUMBER or % OF REVENUE]

Produce the skeleton of an annual marketing plan with these sections:
1. Situation analysis (3 bullet points)
2. 3 strategic priorities for the year (ranked by impact)
3. Channel allocation (% of budget per channel, justified)
4. 4 quarterly milestones tied to revenue
5. Top 5 KPIs with target values

Constraints: no generic "increase brand awareness". Every priority and
KPI must be tied to a number. If a channel is in "not working",
justify re-trying or dropping it.

Tip. This prompt writes the 80% skeleton. The 20% that matters is the judgment you add. For the full 6-phase methodology, see how to build an annual marketing plan for small business, and the SAGA Audiovisual case study for a walked-through example.

Prompt 9. Budget allocation stress test

You are a marketing CFO.

Here is the proposed marketing budget for [COMPANY]:
[PASTE BUDGET TABLE WITH CHANNEL + DOLLAR AMOUNTS]

Revenue goal: [NUMBER]
Current CAC: [NUMBER]
Current LTV: [NUMBER]

Stress-test this budget by answering:
1. Which line items are most overfunded given CAC/LTV math?
2. Which are most underfunded given the revenue goal?
3. Where does the budget assume a miracle (e.g., 10x conversion rate,
   unrealistic click-through rate)?
4. What is the minimum viable budget that still hits the goal?
5. What is the single biggest risk in this allocation?

Show the math for each answer. Do not flatter the plan.

Tip. Run it twice. Second run, prompt “Now argue the opposite of your first answer.” You get a better board-ready view. See how much should an SMB spend on marketing and how to calculate CAC and LTV for the underlying math.

Prompt 10. Quarterly priority ranking

You are a COO helping a marketing team ruthlessly prioritize.

We have the following initiatives on the list for Q[X]:
1. [INITIATIVE + ESTIMATED IMPACT + ESTIMATED EFFORT]
2. ...
[UP TO 15 INITIATIVES]

Do this:
1. Rank all initiatives on a 2x2: impact (low/high) and effort (low/high).
2. Mark which 3 we must ship this quarter.
3. Mark which 3 to drop entirely, with one-sentence justification.
4. Flag any two initiatives that are really the same initiative rebranded.
5. Identify the one initiative that is politically protected but should die.

Be blunt. No diplomatic hedging.

Tip. If the output hedges, follow up with “Which of these rankings would the CEO disagree with most, and why would they be wrong?” This is where the useful insight usually hides.

Persona and ICP prompts (11 to 12)

Prompt 11. Persona from customer interview transcripts

You are a customer research analyst.

Below are transcripts of 5 customer interviews, between triple quotes.
Read them carefully before writing anything.

"""
[PASTE TRANSCRIPTS, LABELED CUSTOMER 1 THROUGH 5]
"""

Extract a single persona based only on patterns that appear in at least
3 of the 5 interviews. Include:
1. Role and seniority (quote-backed)
2. Top 3 pains in their own language (direct quotes required)
3. Top 3 desired outcomes (direct quotes required)
4. The specific moment they start looking for a solution (trigger event)
5. The top 2 objections they voice about buying
6. The words they use that our marketing currently does not use

Do not invent anything. If a pattern only shows up in 1 or 2 interviews,
mark it as "weak signal, not in persona".

Tip. This prompt is gold because it forces source-grounded output. For the full 7-step method, see how to define your ideal customer profile in 7 steps.

Prompt 12. Anti-ICP (who we are not for)

You are a sales enablement lead.

Our ideal customer profile is: [DESCRIBE ICP].

Write the anti-ICP: the customer segment we should refuse to sell to,
even if they want to buy, because they will churn, complain, or damage
our economics. Include:
1. 5 disqualifying characteristics (concrete, observable)
2. 3 questions a salesperson should ask on the first call to spot them
3. The polite script to decline them without burning the bridge
4. One segment that looks like anti-ICP but is actually adjacent ICP

Constraints: be specific enough that two salespeople reading this
would disqualify the same leads.

Tip. Marketers almost never write the anti-ICP. Doing so sharpens messaging because you know exactly who to exclude.

Email prompts (13 to 14)

Prompt 13. Sequence from a single pain point

You are an email copywriter trained on the work of Ramit Sethi and Val Geisler.

Context:
- Audience: [ICP + WHERE THEY ARE IN THE FUNNEL]
- Pain: [ONE SPECIFIC PAIN, QUOTED FROM INTERVIEWS IF POSSIBLE]
- Offer: [WHAT WE ARE SELLING]
- Proof available: [CUSTOMER RESULTS, DATA]

Write a 5-email sequence with:
- Email 1: Name the pain in the reader's language. No pitch.
- Email 2: Story of one customer who had this pain. Proof-led.
- Email 3: Teach the mental model that makes the solution obvious.
- Email 4: Introduce the offer. Reason-why pricing explanation.
- Email 5: Last-chance reminder with a specific deadline.

For each email: subject line (under 50 chars), preview text (under
90 chars), body (under 250 words), clear single CTA.

Constraints: no "just wanted to check in", no "hope you're doing well",
no fake urgency. Each email must be useful if it is the only one read.

Tip. The mistake most marketers make is pitching in email 1. Email 1 is about being read. Pitching starts email 4.

Prompt 14. Subject line ladder

You are a retention marketer.

Email body:
"""
[PASTE FULL EMAIL BODY]
"""

Write 12 subject line variants organized as a ladder from most
conservative to most aggressive:
- 3 plain/informational (describes the email content)
- 3 benefit-led (names the reader's outcome)
- 3 curiosity (ethical; gap-based, not clickbait)
- 3 pattern-break (formatting, emoji use IF brand-appropriate, or
  unusual word order)

Include character count next to each. Flag any that are risky for
spam filters.

Tip. Test one conservative and one pattern-break in every send. Over six months you learn where your list lives on the register spectrum.

Ad prompts (15 to 16)

Prompt 15. Meta ad variant generator

You are a senior paid social copywriter.

Offer: [DESCRIBE]
Audience: [ICP + AWARENESS STAGE]
Platform: Meta (Facebook + Instagram feed placement)
Primary conversion: [LEAD, SIGNUP, PURCHASE]

Write 10 ad variants using this matrix:
- 3 primary text variants (under 125 characters, 2 longer variants
  at 250 characters)
- 5 headline variants (under 27 characters)
- 2 description variants (under 27 characters)

Constraints: no "unlock", no "revolutionize", no "game-changer".
One variant must use a customer quote (if available). One must use
a specific number/stat. One must open with a question. One must open
with a contrarian claim.

Output as a table: Variant # | Primary text | Headline | Description | Angle.

Tip. After the model writes the 10, ask “which 3 would you predict to underperform and why?” The answer teaches you its implicit model of what works, which is worth auditing against your own data.

Prompt 16. Google Ads responsive search ad

You are a PPC specialist.

Product: [DESCRIBE]
Landing page URL: [URL]
Landing page headline: [H1 ON THAT PAGE]
Primary keyword: [KEYWORD]
Top 3 related keywords: [LIST]

Generate a Responsive Search Ad:
- 15 headlines (max 30 chars each)
- 4 descriptions (max 90 chars each)
- Recommended pinning: which 2 headlines to pin to position 1

Constraints: include the primary keyword in at least 5 headlines.
Vary by angle (price, speed, outcome, guarantee, social proof, urgency).
Never stuff keywords. Every asset must be a grammatical sentence.

Tip. Pin sparingly. Google’s algorithm learns faster when you give it variation, not rigidity.

SEO prompts (17 to 18)

Prompt 17. Content brief from a SERP

You are an SEO content strategist.

Target keyword: [KEYWORD]
Search intent: [INFORMATIONAL / COMMERCIAL / TRANSACTIONAL]
My domain authority vs top 3 competitors: [HIGH/SIMILAR/LOW]

Below are the H1s and H2s from the top 5 ranking pages for this keyword,
between triple quotes. [PASTE]

Produce a content brief for a page that can beat them:
1. Recommended title (under 60 chars, keyword-forward)
2. Recommended H1 (different from title)
3. Full H2/H3 outline covering all top-5 subtopics plus 3 subtopics
   none of them cover
4. Target word count based on top-5 average, plus 20%
5. 5 internal link targets (tell me what topic to link to)
6. 3 external citations a page like this needs
7. The specific search intent this page must satisfy in the first 150 words

Constraints: do not propose filler sections. Every H2 must earn its place.

Tip. The “3 subtopics none cover” line is where ranking gains live. If the answer is bland, push back: “Those three are already in the top 10. Give me subtopics that would feel fresh to someone who has read all five pages.”

Prompt 18. Featured snippet capture rewrite

You are an SEO writer.

Current page content for query "[KEYWORD]":
"""
[PASTE CURRENT CONTENT OF THE SECTION]
"""

Current snippet winner (what Google is showing today):
"""
[PASTE THE CURRENT FEATURED SNIPPET TEXT]
"""

Rewrite my section to be more likely to win the snippet. Deliver:
1. A 40 to 60 word paragraph answering the exact query in direct
   language
2. An ordered list version (5-7 steps) if the query is a how-to
3. A 3-column table version if the query is a comparison
4. Explicit placement advice (where in the page this must sit)

Constraints: answer the question in the first sentence. Use the exact
keyword phrasing Google is using in related searches.

Tip. Do not use this unless the current snippet winner is under 70 words. If it is longer, Google already decided the query rewards depth.

Data analysis prompts (19 to 20)

Prompt 19. Campaign post-mortem

You are a senior marketing analyst.

Campaign data below, between triple quotes:
"""
[PASTE CSV OR TABLE: dates, channels, spend, impressions, clicks,
conversions, revenue]
"""

Target: [WHAT THE CAMPAIGN WAS TRYING TO ACHIEVE]
Actual result: [WHAT HAPPENED]

Produce a post-mortem:
1. 3 things that clearly worked, with the math to prove it
2. 3 things that clearly failed, with the math to prove it
3. 2 inconclusive findings (where data is thin or noisy)
4. The 1 structural lesson to apply to the next campaign
5. The 1 question we should have answered before launch that we didn't

Constraints: quote specific numbers from the data for every claim.
Do not invent data that is not in the table. If something is unclear
from the data, say "data insufficient" rather than speculating.

Tip. The last line is the anti-hallucination guard. Without it, models will happily invent a 23% lift that is not in your data. Pair this with AI hallucinations in marketing: 7 real mistakes for the broader guardrail set.

Prompt 20. Funnel leak diagnosis

You are a growth consultant.

Funnel data:
- Visitors last 30 days: [NUMBER]
- Email signups: [NUMBER] (conversion rate: [NUMBER]%)
- Trial activations: [NUMBER] (conversion rate: [NUMBER]%)
- Paid conversions: [NUMBER] (conversion rate: [NUMBER]%)

Industry benchmarks for [INDUSTRY]:
- Visitor to signup: [NUMBER]%
- Signup to trial: [NUMBER]%
- Trial to paid: [NUMBER]%

Diagnose the biggest leak. Answer:
1. Which stage is most below benchmark, by how much, in absolute terms?
2. Given typical SMB causes, list the 5 most likely root causes for
   that leak, ranked by probability.
3. For each root cause, what experiment would confirm or reject it
   in under 2 weeks?
4. What metric would prove the leak was fixed?

Do not propose 10 things. Five maximum, ranked.

Tip. The “absolute terms” clause matters. A 2 percentage point drop at the top of funnel is usually worth more than a 10 point drop at the bottom.

How to make any prompt 3x better

Four habits separate marketers who get usable AI output from marketers who give up.

1. Iterate, do not restart. First output is rarely the final output. After the model responds, say “Cut it by 30%”, or “Rewrite with a contrarian angle”, or “Now argue the opposite.” Most of the quality is in iteration, not in the first generation. Treat the first output as a draft by a junior copywriter.
2. Name what to avoid. Banning words is more effective than specifying words. “Do not use ‘leverage’, ‘unlock’, ‘seamless’, or ‘revolutionize'” does more work than “use a professional tone”.
3. Paste real artifacts. Customer quotes, competitor copy, brand samples. Context beats prompt cleverness. A five-line prompt with 400 words of pasted brand voice samples beats a 50-line prompt with no samples.
4. Use the model’s critique. After it writes, ask “what are the three weakest parts of this draft and how would you fix them?” Then paste its own critique back with “Now rewrite fixing those three.” You get near-expert editing for free.

When to stop prompting and start orchestrating

Here is the honest limit of prompt engineering. Prompts are great for one-off tasks. They break down when the work is a chain of dependent outputs that needs memory across sessions. You do not want to paste your brand voice paragraph into every prompt for a year. You do not want to re-explain your ICP to the model on Tuesday when you explained it Monday.

At that point, marketers either build their own wrapper (see build vs buy: should your SMB build an AI marketing stack) or move to an agentic platform that holds the context for them. FastStrat’s approach is to make the annual plan the persistent memory: every agent, from Martha writing a brief to Rikki pulling citations, reads from the same plan. That design choice exists because prompts alone, no matter how well-engineered, do not scale across months of marketing work. Details in behind the AI: what each FastStrat agent does. For the maturity lens on when to make that shift, see the 5 levels of AI maturity for marketing teams.

The 20 prompts above are the working tier of the MacGyver stage of SMB marketing maturity: the moment you go from “AI is magic” to “AI is a tool, and I know its hands”. For where this stage fits in the broader evolution, see the 4-stage SMB marketing maturity framework. For comparing the tools you run these prompts inside, Jasper vs Copy.ai vs FastStrat.

FAQ

What is prompt engineering for marketers?

The craft of writing model instructions (role, context, task, constraints, format) that produce usable marketing work instead of generic output. It is less about clever phrasing and more about precision and context.

Do I need to learn prompt engineering if I use an agentic platform?

Less, but not zero. Agentic platforms hold context for you, but you still write briefs, refine outputs, and guide agents. The same clarity skills transfer.

Which model should I run these prompts in?

For most long-form work, Claude. For quick drafting and brainstorming, ChatGPT. The framework is model-agnostic. See ChatGPT vs Claude vs FastStrat.

How do I stop the model from making up facts?

Paste the source material and explicitly say “If the answer is not in the pasted material, say ‘not in source’ instead of speculating.” This single constraint cuts hallucinations dramatically. More mitigations in AI hallucinations in marketing: 7 real mistakes.

How long should a good prompt be?

Long enough to include role, context, task, constraints, and format. That is usually 100 to 500 words for serious marketing work. Shorter prompts produce shorter thinking.

Can I just copy these 20 prompts and use them?

Yes. That is what they are for. Replace the bracketed placeholders with your real context. The specificity of your inputs is what separates a good output from a great one.

Next steps

Pick three prompts. Run them on your own work this week. If they produce something usable, save the version with your context filled in as a reusable template. Do that four times and you have a personal prompt library worth more than any AI tool subscription.

Explore the FastStrat AI agent team, check current pricing, or read the FAQ.

---

About the author. Walter Von Roestel is CEO of FastStrat. He has reviewed more than 10,000 marketing prompts written by clients and team members, and has the scar tissue to prove most of them could have been half as long and twice as clear.