20 AI Prompts for Marketers Who Are the Whole Department

The words HELP ME WITH.. appear above a search bar graphic, with a large cursor clicking on it, set against a bright blue background—suggesting users can find ai prompts for marketers with ease.

ChatGPT, Claude, Perplexity, and Gemini prompts for marketing content, outreach, strategy, and design.

TL;DR

20 AI prompts I use every week as a solo B2B marketer, spanning ChatGPT, Claude, Perplexity, and Gemini. Each prompt is tagged by tool, includes variations and failure modes, and covers content production, stakeholder management, strategic decisions, data operations, and visual assets. These are not generic prompts. They are constraint-loaded, built from real workflows, and designed for marketers running a full department alone.

The Story Behind These Prompts

I am the marketing department at Workzone. That is the whole department. Me.

There is a fractional contractor handling production. A part-time designer for the brand work. Beyond that, I am the demand gen team, the SEO team, the webinar team, the paid search team, the social team, the email nurture team, the AEO team, the direct mail team, and the person writing copy at 11 PM because no one else is going to.

If you are a marketing team of one, you already know the shape of this. You are not behind because you are bad at your job. You are behind because there are five jobs inside your job description and a finite number of hours.

AI did not solve that. What it did was make it survivable.

I started building this prompt library a year ago. Not all at once. One prompt at a time, in moments of desperation. The Sunday night I realized I had a webinar four days out with no promotion copy. The Tuesday morning a board update was due in three hours and I had a wall of HubSpot data and no narrative. The Tuesday I realized the project request that fell through the cracks was one I could have caught in 10 seconds if I had looked at the dashboard instead of trusting my inbox. The Friday afternoon someone dropped a “quick favor” in Slack at 4:45 and I had dinner plans I was determined to keep.

Each prompt below has a story like that behind it. I am sharing them because nobody handed me a manual for doing a department’s worth of work alone. I figured it out by writing the same kinds of prompts over and over and saving the ones that worked.

I use Claude every working day. Sometimes 30 times in a day. But Claude is not the only tool in the stack. ChatGPT is faster for high-volume generation. Perplexity is better when I need current data or research. Gemini lives inside the Google ecosystem where half my analytics questions land. The prompts below are tagged with the tool I actually use for each one, and why.

A few honest things before you copy anything:

These are not magic prompts. They work because they are specific. The minute you strip out the constraints (the “no emdashes,” the “five options,” the “do not paraphrase”), they get generic. Keep the constraints. Add your own.

Every prompt below is a starting point. I tweak each one based on the situation. The vertical translator changes depending on what verticals I am bridging. The brief writer changes depending on who is executing the work. Treat what is below as the skeleton, not the finished thing.

These work better in a long thread than as a one-shot. I almost never paste a prompt, take the first output, and ship it. I push back. I redirect. I say “make version three sharper” or “you missed the point on number two, try again.” The output gets better the more I push. If you ship the first draft, you get B+ work. If you push three or four times, you get A work.

What follows are the 20 prompts I would not work without.

Before You Use Any of These: The Five Things You Need First

These prompts work because they are loaded with constraints. The constraints come from documents I built before I ever wrote a single prompt. If you skip this step, every prompt below will produce generic output and you will blame the tool when the problem was the inputs.

Here is what you need to have in place before any AI prompt library will save you time instead of creating more editing work.

A brand guide. Colors by hex code, typography rules, logo usage, button styles. AI image tools like Midjourney and Ideogram need exact hex values or they will guess, and they will guess wrong. Every visual prompt in this post references four specific hex codes. Those come from the brand guide. If you do not have one, build one. It does not need to be 40 pages. It needs to have your colors, your fonts, and your visual identity documented somewhere you can paste from.

A messaging and voice/tone guide. This is the single most important document for making AI sound like your brand instead of like every other B2B SaaS company. Mine defines five voice traits, a tone spectrum that shifts by channel, messaging pillars by vertical, and before/after examples that calibrate what “on-brand” actually sounds like. Without it, you will spend more time editing AI output than you saved generating it. Write the guide once. Reference it in every prompt. Update it when the brand evolves.

A banned words list. This is non-negotiable. Every AI model has default vocabulary it reaches for, and most of it is the same generic language your competitors use. My banned list includes words like “leverage,” “optimize,” “streamline,” “robust,” “innovative,” and a few that are specific to my brand voice (“real,” “weep,” “chaos”). You will notice the prompts below reference “banned words I gave you” repeatedly. That reference only works if the list exists. Build yours. Start with the 10 words that make you cringe when you see them in a draft. Add to it every time AI produces a word you would never use. The list grows fast.

Banned content patterns and punctuation rules. Separate from individual words. These are structural rules about how your brand writes. Mine include: no emdashes (restructure the sentence instead), no “[X] didn’t just do this, [X] also did this” constructions, one exclamation point per piece maximum, no passive hedging, no jargon soup, pain before product always. Your rules will be different. The point is that AI will default to every one of these patterns if you do not explicitly ban them. The prompts below include these constraints inline. Yours should too.

Design preferences documented for AI image generation. If you use Midjourney, Ideogram, or any image generation tool, you need your visual style codified: preferred illustration styles, composition rules, what to avoid (stock-photo cliches, specific visual tropes), aspect ratios by channel, and how brand colors should be applied. I default to editorial illustration with paper grain for Midjourney and bold typographic posters for Ideogram. Those defaults are baked into the prompts below. Without documented preferences, every image prompt becomes a coin flip.

None of this takes more than a day to build. Most of it you already know. You just have not written it down in a format you can paste into a prompt. Do that first. Then come back to these prompts. The difference in output quality will be immediate.

Content Production Prompts

1. The Sales Call Pain Miner (ChatGPT)

The first time I rewrote a vertical landing page using verbatim language from a sales call, the page outperformed the one I had spent two weeks writing from my own brain. That was the lesson. The prospect always says it better than I do. My job is to mine it.

This prompt is the one I run before I write anything for a vertical asset. Sometimes I run it on three or four transcripts at once and then look for repeated language across all of them. The phrases that come up multiple times become the headline candidates. The phrases that show up once but punch hard become section headers or testimonial frames.

Why ChatGPT: Fast extraction from long transcripts. ChatGPT handles large pastes well and is aggressive about pattern recognition across multiple inputs, which is exactly what this task needs.

You are reviewing a sales call transcript. Pull every verbatim pain point the prospect uses about their current process. Do not paraphrase. Use their exact words.

Group the pain points into themes: visibility, intake, accountability, tool fatigue, adoption, reporting, capacity, leadership pressure, team burnout.

Skip any sentence that does not contain a specific named pain. I do not want filler like "we're growing fast" or "things are busy." I want sentences like "her list is the notebook" or "we have 400 unread messages and no way to triage."

For each pain point, include:
- The verbatim quote
- The theme it belongs to
- One sentence on how I might use this in copy

End with three sentences describing the prospect's emotional state during the call.

[Paste transcript]

Variations I use: When I have multiple transcripts, I paste them all in and ask for “the top five repeated pains across all calls, in the language they were said.” When I only have one call but it is exceptionally good, I ask for “the 10 most quotable lines, ranked by how visceral they feel.”

When it fails: If the sales call was a discovery call with no real pain elicited, the prompt finds nothing. That is a signal that the call was light, not that the prompt is broken. Push the salesperson to dig deeper next time.

2. The Vertical Translator (Claude)

We sell into four verticals. Healthcare, higher ed, credit unions, and manufacturing. The pain is structurally similar (visibility, intake, accountability, tool fatigue) but the language is completely different. A hospital marketer does not say “her list is the notebook.” A credit union ops manager does not say “service line launches stalled in physician approval.”

Writing four versions of every piece of content from scratch would kill me. The vertical translator is the prompt that lets me write the strongest version for one vertical and then port the structure to the other three without losing the spine of the argument.

(If you work in higher ed specifically, I wrote a dedicated set of ChatGPT prompts for higher ed marketers that goes deeper on that vertical.)

Why Claude: Preserving voice and structure across a long rewrite while swapping out vertical-specific language requires nuance. Claude holds the constraints better than other tools on this kind of long-form, voice-sensitive transformation.

Below is a marketing asset written for [original vertical]. Rewrite it for [target vertical].

Constraints:
- Keep the structure and headline framework intact
- Replace examples, named pains, and language with terms native to [target vertical]. The pain is the same but the words are different.
- Replace customer proof points with the ones I'll paste below
- Replace any vertical-specific compliance, regulatory, or operational language with the target vertical's equivalent
- Do not soften or hedge. If the source is direct, the rewrite stays direct.
- Flag any section where the structural argument does not actually port across (sometimes a healthcare pain has no manufacturing equivalent and pretending it does makes the rewrite weaker than the original)

[Paste original asset]
[Paste vertical-specific proof points]
[Paste any vertical-specific language I want preserved]

Variations I use: When I want to test which vertical version is the strongest, I run the translator across all four and then ask Claude to “rank these four versions by emotional intensity and tell me which audience this piece naturally serves best.” Sometimes I discover that the original was secretly written for the wrong vertical the whole time.

When it fails: This prompt works when the underlying argument is universal. It fails when the original piece relied heavily on vertical-specific specifics that have no equivalent. If a piece is about HIPAA, you cannot translate it to manufacturing. You need a different piece.

3. The Repurpose Engine (ChatGPT)

I cannot write net-new copy for every channel. A blog has to become a LinkedIn post, an Instagram caption, a newsletter blurb, email subject lines, and a paid social headline. Doing each of those from scratch would take a full day. This prompt does it in 10 minutes and saves me a workday a week.

Why ChatGPT: This is a volume task. ChatGPT is faster at producing many derivative assets in one shot without needing to be coaxed into generating all of them.

Below is a long-form blog post. Generate the following derivative assets, each in my brand voice (direct, pain-forward, no emdashes, no banned words):

1. LinkedIn post (250 words, authoritative, ICP-focused, ends with a question to drive comments)
2. Instagram caption (90 words, shorter, hook-led, more personality)
3. Newsletter blurb (3 sentences plus a one-line CTA)
4. Email subject lines (5 options, pain-named)
5. Paid social headline (3 options under 60 characters, 3 options under 90)
6. One-sentence promo line for the registration or landing page
7. A 3-tweet thread version (if Twitter/X is in the mix)
8. A 30-second video script if I want to record a talking head

For each, give me two versions: the safe one (close to the original) and the punchier one (riskier, more personality).

[Paste blog]

Variations I use: When I am repurposing a webinar instead of a blog, I add “extract the three most quotable moments from the transcript and build social copy around each.” When I am repurposing for a customer case study, I add “make sure the customer name appears in every version because the proof is the point.”

When it fails: The repurpose engine is only as good as the source content. If the blog is generic, all 10 derivatives will be generic. Garbage in, garbage out. Spend the time on the source piece. Everything downstream gets easier.

Working with Contractors and Ghost-Writing

4. The Brief Writer (Claude)

I am the strategy and the direction. Someone else is the production. The handoff lives or dies on the brief. A vague brief gets three rounds of edits. A specific brief gets clean work the first time.

I learned this the hard way. The first six months of working with my contractor, I would write briefs that said things like “blog about higher ed marketing challenges, around 1,500 words, by Friday.” She would ship work I had to send back. Not because she was wrong, but because I had not given her enough to be right.

The brief writer below changed that. Now I spend 10 minutes on the brief and save an hour on edits.

Why Claude: A content brief is a constraint-heavy document. Claude holds 10+ requirements in its head without dropping any, and the output stays structured without needing reformatting.

Write a content brief for the asset described below. Use this structure:

1. Asset type and format
2. Audience (role, vertical, awareness level: cold, warm, in-funnel)
3. The specific pain we are naming, in the audience's own words
4. The angle (what makes this different from generic content on the topic)
5. The one thing the reader should walk away believing
6. Tone notes (direct, pain-forward, no jargon, no emdashes, banned words to avoid)
7. CTA (specific, one ask)
8. Length and format constraints
9. Two example headlines that match the angle
10. Internal links the piece must include
11. Anything to avoid
12. The reference assets the contractor should read first

Asset description: [describe]
Audience details: [paste]

Variations I use: For social briefs, I drop sections 8 and 10 and add “channel-specific tone notes” and “asset count.” For paid ad briefs, I add character counts and “what we are testing against.” For landing pages, I add “primary conversion event” and “secondary CTA if the primary fails.”

When it fails: If you do not actually know what the angle is, no brief writer prompt will save you. The prompt cannot invent strategy. It can only structure what you already know. If you find yourself writing “[need to figure this out]” in three sections, the issue is not the brief. The issue is you have not done the strategic thinking yet. Step away from the brief and figure out the angle first.

5. The Voice Match (Claude)

Sometimes I have to write in someone else’s voice. A guest post under a customer’s byline. A LinkedIn ghost post for the CEO. An email that goes out from sales but I am writing. The Voice Match prompt lets me feed Claude an example of how that person actually writes and then produce work that sounds like them, not like me.

Why Claude: For voice-sensitive work (brand copy, executive ghost-writing, anything that has to sound like a specific human), Claude produces longer, more nuanced first drafts that need less re-prompting on tone.

Below is a sample of how [person] writes. Study it carefully.

Then write [the new piece I need] in that exact voice.

Pay attention to:
- Sentence length patterns (do they write short? long? mixed?)
- Vocabulary choices (formal? casual? industry-specific?)
- Rhetorical moves (do they use questions? lists? stories?)
- Pacing (do they get to the point fast or build slowly?)
- What they avoid (jargon? hedging? emdashes?)
- How they open (with a hook? with context? with a stat?)
- How they close (with a CTA? with a question? with a reflection?)

After you draft it, point out three specific places where you matched their voice and one place where I might want to push the draft further to sound more like them.

Voice sample: [paste 200-500 words of their writing]
What I need: [describe the new piece]

Variations I use: When I am writing for an executive, I add “this person is operating at the strategic level. Do not use tactical language. Stay at altitude.” When I am writing for a customer, I add “this person is not a marketer. They will not use marketing language. Write the way an ops director would write.”

When it fails: If the voice sample is too short (under 200 words) or too uniform (one type of content), the match will be shallow. Feed it varied samples when you can. A LinkedIn post, an email, a blog comment. The richer the sample, the better the match.

Drafting and Polishing Prompts

6. The Multi-Option Generator (ChatGPT)

I almost never want a single recommendation. I want five versions so I can feel the differences and pick. This is the universal request shape I bolt onto 80% of my drafting prompts.

Why ChatGPT: Faster and more aggressive about offering options without being asked twice. When the task is “give me volume so I can compare,” ChatGPT is the right tool.

Give me [N] versions of [the thing I'm asking for].

For each version:
1. A clear label naming its angle (don't make me guess what's different)
2. The version itself
3. One sentence on when I'd choose this one
4. One sentence on the risk of this version (when it would land flat)

Order them from safest to boldest.

No preamble. No closing summary. Just the versions.

Variations I use: Sometimes I want 10 instead of 5. Sometimes I want them ordered by tone (warm to direct) instead of risk (safe to bold). Sometimes I want them grouped (3 short, 3 medium, 3 long). The structure is the same. The dimensions change based on what I am trying to feel through.

When it fails: When I do not give enough context, the options all collapse into variations of the same idea with different word choices. The fix is to specify the dimensions I want the versions to differ on. “Different angles,” “different tones,” “different audiences,” “different lengths.” Pick the axis. Then the options will actually be different.

7. The Punch-Up on What I Already Wrote (Claude)

Most of my drafts are 80% there. I do not need a rewrite. I need someone to make them sharper without flattening the voice. The Punch-Up prompt is the one I use when I am close but not done.

Why Claude: Editorial nuance. Claude is better at sharpening prose while preserving the original voice, where other tools tend to flatten everything into generic SaaS copy.

Below is a draft I wrote. Do not rewrite it. Punch it up.

Specifically:
- Cut every sentence that could be cut in half
- Replace soft verbs with specific ones (suggest → say, help → fix, support → run)
- Find the line that buries the strongest point and move it earlier
- Mark anything that sounds generic so I can decide whether to keep it
- Flag any banned words I gave you (real, weep, chaos, playbook, leverage, optimize, streamline)
- Flag any emdashes or "[X] didn't just do this, [X] also did this" constructions
- Identify any place where I am hedging when I should be direct
- Identify any place where I am making a claim I have not earned with proof

Show me the punched-up version with edits visible (track-changes style), then a clean final.

[Paste draft]

Variations I use: When I want a lighter touch, I tell Claude “do not change my voice. Just sharpen what is already there.” When I want a heavier hand, I say “rewrite anything that is weaker than it could be. Voice is secondary to impact on this one.”

When it fails: The punch-up prompt sometimes flattens voice when I do not specify “preserve my voice.” If the output reads like generic SaaS copy when your draft did not, that is the prompt failing the voice test. Re-run it with stronger voice constraints. Or just take the flagged issues and fix them yourself.

8. The Headline Punch-Up (ChatGPT)

My first headline draft is almost always too long, too soft, or too clever. This prompt gives me 10 alternatives so I can pick one and adjust.

Why ChatGPT: High-volume headline generation. When I need 10 options fast with real variation between them, ChatGPT delivers without making me ask for more.

Below is a headline I drafted. Give me 10 alternatives that:

- Lead with the pain, not the product
- Are specific (name the broken thing)
- Stay under 12 words
- Avoid rhetorical questions as the hook
- Avoid any banned words I gave you
- Do not use emdashes
- Do not use the "[X] didn't just do this, [X] also did this" construction
- Vary in tone: 3 safe, 4 medium, 3 bold

For each, name the tone bucket and one sentence on what kind of piece it would work best for.

My draft: [paste]
Audience and asset type: [paste]
Where this will appear (blog, ad, email subject, landing page hero): [paste]

Variations I use: For paid social headlines I add “must fit in 60 characters” and “must be testable against one another (the differences should be diagnostic, not just stylistic).” For blog headlines I add “must pull in search and in social (specific enough to be SEO-friendly, hooky enough to be share-worthy).”

When it fails: If you are writing a headline for a piece you have not finished, the headlines will feel hollow. Headlines come from understanding the strongest single line in the piece. Finish the piece first. Then headline it.

9. The 4:45 PM Friday Rescue (ChatGPT)

Someone drops a “quick favor” into Slack at 4:30 on a Friday. I have 15 minutes before I lose my evening. This is the prompt that buys me back my Friday night.

Why ChatGPT: Speed. When I have 15 minutes and need three usable versions, ChatGPT is faster to first output than any other tool in the stack.

I have 15 minutes. I need to draft [the thing]. Here is the context: [paste].

Give me three versions:
1. The safest version (uses what I gave you, nothing fancy)
2. The version with a stronger pain hook (riskier, more memorable)
3. The version I would send if I trusted my gut more than the room

For each, give me a one-sentence note on who I would send it to and what they would think when they opened it.

No preamble. No closing summary. Just the three versions, labeled.

Variations I use: When the ask is genuinely urgent, I skip the three versions and ask for one: “Just give me the best version. I do not have time to choose.”

When it fails: When I forget to paste the actual context. Half the time my 4:45 Friday rescue is delayed because I asked for help before I gave it any of the information it needed. The fix is to take 60 seconds to dump everything I know into the prompt before I hit enter. Even a messy context dump beats a clean prompt with no context.

Stakeholder Management Prompts

10. The Saying No Generator (Claude)

Someone asks for a one-pager Wednesday afternoon, due Thursday. Three people want priority on the same week. The CEO asks “can you also…” and the answer probably needs to be no. I cannot say yes to all of it. This prompt gives me three ways to push back without burning the relationship.

Why Claude: Navigating tone in a politically sensitive email requires nuance. Claude handles the warmth-and-firmness balance better than tools that default to either corporate softness or bluntness.

Write three versions of an email response declining or pushing back on the request below. The relationship matters; I am not trying to stiff-arm anyone. I am trying to protect my time without sounding precious about it.

Version 1: Soft no, offers a future date
Version 2: Hard no, suggests a different solution they can own themselves
Version 3: Conditional yes, names what would have to come off my plate to make it work

For each:
- Tone: warm, direct, no apologizing for having priorities
- Short sentences
- No corporate softening
- Use specifics about my current workload to ground the no in reality, not moods

Request: [paste]
Current top three priorities I'd have to deprioritize to say yes: [paste]

Variations I use: When the request is from leadership and saying no outright is not an option, I drop version 1 and 2 and ask for “three versions of a conditional yes, each one trading off a different priority. I will pick which trade is worth it.”

When it fails: When the answer should actually be yes. The saying no generator is not a tool for avoiding work I should do. It is a tool for protecting against work that genuinely is not the highest leverage thing on my plate. Use it carefully. Saying no too often is its own problem.

If the problem is less about saying no and more about writing the email at all, I wrote a separate piece on using Claude for the emails that make you hate your life. Different prompts, same desperation.

11. The Weekly Check-In Helper (Claude)

I have a weekly check-in with my manager. Three questions. Every week. By the time Friday rolls around, my brain is mush. I cannot remember what I did on Monday. This prompt is the one that turns my week of scattered notes into a clean update in 10 minutes.

Why Claude: Turning a messy pile of notes into a coherent narrative with the right framing requires judgment, not just formatting. Claude gets the tone right (confident, not breezy) without over-polishing.

I do a weekly check-in with my manager in three-question format:
1. What did I accomplish this week?
2. What is blocking me or where do I need help?
3. What is next?

Below are my notes from the week (Slack messages, calendar events, completed tasks, half-finished thoughts). Turn them into a check-in response that:
- Uses professional but human prose for questions 1 and 2
- Uses clean single-line bullets for question 3
- Frames challenges as active expansion or proactive steps, not waiting or patience
- Leads with the biggest wins, does not bury them
- Quantifies what I can quantify (SQLs, trials, demos, ad spend, content shipped)
- Notes the work in progress separately from the work that shipped
- Flags anything I might be forgetting based on what I told you last week (if I share prior weeks' check-ins)

Tone: professional, confident, honest. Not breezy. Not panicked.

This week's notes: [paste]
Last week's check-in for context: [paste]

Variations I use: When I am behind on something my manager flagged the week before, I add “specifically address the [thing] from last week’s check-in and how it has moved.” When I want the check-in to set up an ask, I add “the goal of this check-in is to surface [need] without making it the headline.”

When it fails: If my notes are too sparse, the check-in is too thin. The fix is to dump everything from the week even if it feels excessive. Calendar events, Slack threads, half-finished ideas, things that did not happen but should have. The more context, the cleaner the output.

12. The Leadership Update Builder (Claude)

Once a month, leadership sees a marketing update. They are smart, busy, and they do not want a marketing memo. They want outcomes in plain language. This prompt is the one I use to turn a HubSpot data dump into something they can read in 90 seconds.

Why Claude: Translating a wall of marketing data into an executive narrative requires judgment about what to foreground and what to cut. Claude handles that editorial decision-making well, and the output reads like a person wrote it, not like a report template filled it.

Draft a one-page marketing update for [audience: board, leadership team, exec staff]. The tone is outcomes-centered, plain language, no jargon.

Structure:
1. Headline number for the month (the one stat that frames everything)
2. Three wins (specific, with numbers)
3. Two risks (named honestly, with what I am doing about each)
4. The top three priorities for next month
5. One thing I need from leadership (decision, budget, intro, signoff)

Rules:
- Separate total volume (clicks, impressions, signups) from rate (CTR, conversion rate). Do not collapse them into one number.
- Name actual SQL counts, trial volume, demo bookings, pipeline. Vanity metrics go at the bottom or get cut.
- I pull most of this from HubSpot and from our own reporting inside the platform. The numbers are already there. The prompt's job is to give them a narrative, not to invent one.
- Do not editorialize. The numbers speak. My job is to give them the narrative around the numbers, not to spin.
- No marketing jargon. If a phrase requires explanation, rewrite it in plain language.
- Each section is short. One paragraph for wins. One paragraph for risks. Bullets for priorities. One line for the ask.

This month's data: [paste]
Last month's update for context and trend lines: [paste]

Variations I use: When the data is bad, I add “frame the misses as learnings without minimizing them. Be honest. The board respects honesty more than spin.” When the data is great, I add “do not make this read like a victory lap. Confidence without bragging.”

When it fails: When I do not have the data ready. The prompt does not invent numbers. If you are missing key metrics, get them first. A leadership update with question marks is worse than one delivered a day late.

13. The Outreach Sequence Builder (ChatGPT)

Most outreach work is not a single email. It is a 3-touch sequence with day intervals and a channel switch. Building those one at a time is the slowest part of any campaign. The sequence builder is the one I use for cold outreach, partnership asks, webinar speaker invites, and re-engagement to lapsed contacts.

Why ChatGPT: When I need five subject line options per touch across a three-email sequence, ChatGPT generates at volume and with variation. It does not second-guess the ask the way some tools do.

Build a 3-touch outreach sequence for [audience] aimed at [goal: webinar registration, demo booking, partnership ask, speaker invite, content collaboration].

For each touch:
- Subject line (5 options, varied by approach: question, stat, name-drop, pain, curiosity)
- Body (under 150 words, conversational, plain-text feel, specific to the audience's role and what they care about)
- CTA (one clear ask, no double-asks)
- Send day relative to touch 1 (touch 1 = day 0)
- Channel (email or LinkedIn DM)
- The hook that ties this touch back to the previous one (so it reads as a sequence, not three disconnected emails)

Touch 2 should reference the specific reason touch 1 might have been ignored without sounding needy. Touch 3 is a soft break-up that leaves the door open and gives them a reason to respond even if it is to decline.

Context on the audience: [paste]
Context on the offer: [paste]
What I know about this person specifically (recent post, job change, content they published): [paste]

Variations I use: For warm audiences (people who have already engaged once), I cut from 3 touches to 2 and remove the “specific reason touch 1 was ignored” angle since they are not cold. For partnership outreach with a specific person I want to land, I add “make touch 1 a personalized angle that references their work specifically, not just their company.”

When it fails: When the offer is weak. A great sequence on a weak offer is still a weak offer. Fix the offer first. Test it on five people before you scale to a sequence campaign.

Strategic Decision Prompts

14. The Strategy Pressure Test (Claude)

I do not have a peer I marketing-strategize with every day. The strategy pressure test is what I use when I am about to commit to a 90-day initiative and I need someone to push back hard before I sign anything.

Why Claude: This is the prompt where I need genuine pushback, not sycophancy. Claude’s reasoning depth makes the steelmanned objections feel like they come from someone who has actually thought about the problem, not someone parroting a framework.

I am about to commit to the initiative described below. Before I do, pressure-test it.

1. What is the strongest argument against doing this at all? (Steelman the objection. Do not give me a strawman.)
2. What is the second-best version of this idea that I am not seeing?
3. What assumption am I making that, if wrong, kills this entirely?
4. What is the cheapest experiment I could run in the next two weeks to test the core assumption before fully committing?
5. What would a smart skeptic say after reading my plan? (Be specific. Pretend you are someone who has shipped this kind of initiative three times before.)
6. What would I need to see in the first 30 days to feel confident? In the first 60? In the first 90?
7. What is the kill criteria? (What would tell me to walk away before I have sunk too much time?)

Be direct. Do not soften your critique. I would rather hear the hard version now than discover it in week six.

Initiative: [paste]
The reasoning that got me here: [paste]
What I am giving up to do this: [paste]

Variations I use: When I am evaluating a vendor or partner instead of an internal initiative, I drop question 4 and add “what would I need to know about this vendor’s track record before signing? What questions would a skeptical CFO ask me about this purchase?”

When it fails: When I am not actually open to hearing pushback. The strategy pressure test is only useful if I am willing to kill the idea or change the plan based on what comes back. If I have already decided to do the thing, do not use this prompt. Use it before the decision, not after.

15. The Channel Decision Helper (Perplexity)

Vendors pitch me constantly. New ad placements, new newsletters, new conferences, new tools. I cannot afford to chase all of them. This prompt forces a clean evaluation before I sign anything.

Why Perplexity: This decision needs current data: what the vendor’s actual reach is, what comparable placements cost, what the industry benchmarks are. Perplexity can pull live research and cite sources, which makes the evaluation grounded in current reality instead of whatever the vendor’s deck claims.

I am evaluating whether to invest in [channel/placement/tool]. The cost is [$X]. The expected outcome is [Y].

Help me think through:
1. What is the bear case (why this fails or underperforms)?
2. What does success look like in measurable terms, and what should I see at week 2, week 6, and week 12?
3. What is the smallest version of this I could test before going all in?
4. What evidence would tell me to kill it and reallocate?
5. What other initiative am I implicitly choosing not to fund if I do this?
6. What questions would I want to ask the vendor before signing? (Give me the questions a skeptical CMO would ask, not the polite ones.)
7. What is a fair benchmark to compare this against? (If I am spending $X on this, what is the equivalent spend on a channel I already trust, and what would I expect that to produce?)

Context I already know: [paste]
What the vendor is promising: [paste]
My historical performance on similar channels: [paste]

Variations I use: When the channel is something I have already tried and underperformed, I add “what would I do differently the second time? What conditions would have to be true for this to work this time when it did not work last time?”

When it fails: When I am evaluating something I have already emotionally decided to buy. The channel decision helper is for decisions I am genuinely on the fence about. If I have already committed in my head, the prompt becomes confirmation bias dressed up as evaluation.

16. The Campaign Post-Mortem (Claude)

Most campaigns leave the office without a debrief. That is how the same mistakes show up next quarter. The post-mortem prompt is the one I use to force a 30-minute reflection on every major campaign before I move on to the next.

Why Claude: Honest analytical assessment without spin. Claude handles the “what did not work” section without softening it into something unrecognizable, which is the whole point of a post-mortem.

I just wrapped the campaign described below. Draft a post-mortem in this format:

1. Goal (the original SQL or pipeline target, in the numbers I set at the start)
2. Actual outcome (the number, no spin, with comparison to the goal)
3. What worked (3 items, specific, with the evidence behind why I think it worked)
4. What did not work (3 items, specific, with the evidence behind why I think it did not work)
5. What I would do differently next time (3 items, concrete)
6. What to keep doing (2 items)
7. What I am still uncertain about (open questions I cannot answer yet)
8. The one big lesson (the thing I want to remember six months from now when I forget the details of this campaign)

Rules:
- No preamble
- Direct prose, no hedging
- One paragraph per section maximum
- No blame, no defensiveness. The campaign is over. The point is to learn, not to assign credit or fault.

Campaign details: [paste original goal, channels used, audience, timeline, budget]
Final results: [paste actual numbers]
Anything weird that happened during the campaign: [paste]

Variations I use: For campaigns that significantly outperformed, I add “be specific about which variables drove the lift. Was it the offer? The audience? The creative? The timing? I want to know what to repeat.” For campaigns that underperformed, I add “be honest about what was a strategic miss versus what was execution. Those have different fixes.”

When it fails: When the data is incomplete. Half a post-mortem is worse than no post-mortem because it tricks you into thinking you have learned something when you have not. Pull the data first. Then run the prompt.

Data and Operations Prompts

17. The Marketing Ops Translator (Gemini)

I am a marketer, not a data engineer. When something in HubSpot, GA4, our project management reporting, or my paid agency’s numbers does not match what I expected, I need a plain-language answer fast. This prompt gets me one without three hours of documentation reading.

Why Gemini: Most of my ops questions live inside the Google ecosystem (GA4, Search Console, Google Ads). Gemini has native context on these tools and can answer Google-specific questions without needing me to paste in documentation. For HubSpot-specific issues, I sometimes run this same prompt in Claude instead.

I am seeing [thing that does not make sense] in [HubSpot / GA4 / Google Ads / paid search reporting / SE Ranking / Search Console].

Help me understand:
1. What is the most likely explanation (the boring one, not the conspiracy)?
2. What is the second-most likely explanation?
3. What specific thing should I check first to confirm?
4. If this is broken, what is downstream of it (what other numbers am I now wrong about)?
5. What is the plain-language version of this so I can ask my agency or vendor a clean question without sounding like I just learned what UTMs are?
6. What is the fix, ranked by effort (smallest effort fix first)?

Specifics I have: [paste numbers, screenshots, settings, the URL or query in question]
What I think is happening (might be wrong): [paste my guess]

Variations I use: When the issue is across multiple tools (HubSpot says one thing, GA4 says another), I add “explain the discrepancy. Which one is closer to truth? What is the actual measurement difference between these two systems?”

When it fails: When the problem is genuinely a vendor issue and the answer is “you need to call your agency.” The prompt is great for understanding what is happening. It cannot fix things you do not have access to.

18. The Stat Replacer (Perplexity)

A stat I have been using for two years is no longer accurate. Or worse, it is technically true but no longer flattering. I need a replacement that says the same thing in a different way without making it up.

Why Perplexity: This is a research task disguised as a copywriting task. Perplexity can search for current data, verify whether a stat still holds up, and suggest alternatives with sources, which is exactly what this prompt needs.

The stat below is one I want to retire. Help me find replacements that:
- Stay in the same proof category (longevity, scale, reliability, customer outcomes, retention, satisfaction)
- Are more current or more specific
- Could be verified or sourced from data I already have
- Hold up under scrutiny (do not use stats that an analyst would immediately question)

For each option, give me:
- The proof category it falls under
- The candidate stat phrased as it would appear in copy
- What data I would need on hand to verify it
- Where it would land best (homepage, landing page, sales deck, email, social)
- The weaker version I should avoid (some stats sound good but are easy to puncture)

Original stat: [paste]
Where it currently lives (asset, surrounding copy): [paste]
What I know about my customer base, product, or company data: [paste]
What I cannot say (numbers I do not have, claims that would not hold up): [paste]

Variations I use: When I am refreshing a whole stats row instead of one stat, I add “I am looking for 4-6 stats that hang together as a proof story. Make sure they are not redundant. Each should answer a different question a prospect would ask.”

When it fails: When the data does not exist and I am trying to invent it. The stat replacer cannot make up facts. It can only help me think about what would be defensible to claim. If you do not have the underlying data, get the data first.

Visual Asset Prompts

A marketing team of one does not have a design team. I have a part-time designer for brand work, but the Tuesday afternoon LinkedIn graphic, the newsletter header, the webinar promo card, the Instagram post that needs to go out before EOD? Those are on me. Before Midjourney and Ideogram, each one was either a Canva template I hated or a request that sat in the designer’s queue until the moment had passed.

Now I generate most of my social and newsletter graphics myself in 10 minutes. Not all of them. The brand campaigns, the landing page heroes, the stuff that needs to feel crafted still goes to the designer. But the volume work, the graphics that need to exist by 3 PM and look like they belong to the brand? That is Midjourney and Ideogram. Between the two of them, they save me three to five hours a week I used to spend either waiting on assets or settling for templates that looked like everyone else’s.

Midjourney handles the editorial illustrations, the scene-setting imagery, the blog headers and social visuals that need to feel like a magazine spread. Ideogram handles the typographic assets, the stat cards, the bold poster-style graphics where the text is the design. Together, they cover about 80% of what I need for newsletters and social posts without opening a design tool.

19. The Midjourney Prompt Builder (Claude)

Midjourney v7 has a specific prompt structure that produces consistent results, and the structure is fussy enough that I was getting it wrong half the time before I let Claude hold it for me. The key is that Midjourney prompts read like a single continuous sentence: subject, action, setting, style, colors by hex, then the flags at the end. No bullet points. No instructions. Just a description of the image you want, written in the language Midjourney understands.

I use this for blog headers, newsletter hero images, LinkedIn visuals, and social graphics where I need an illustration rather than a typographic design. The editorial illustration style has become part of the brand look at this point. People recognize it in the feed.

Why Claude: Claude holds the full prompt structure (hex colors, aspect ratios, style anchors, the “no hands” rule, the cliché avoidance list) without dropping any of it. One missing flag produces a completely different image. Claude does not forget.

Here is what a finished Midjourney prompt looks like. This is the output I am training Claude to produce:

editorial illustration of a speaker at a podium on a warmly lit stage with a large screen behind showing project timelines and an audience of silhouettes leaning in attentively, flat textured illustration style, navy #003282 audience area, bright blue #2094e0 screen glow, green #29d296 on track indicators on the screen, orange #ff4f01 spotlight on the speaker and podium, botanical potted plants flanking the stage, retro editorial magazine aesthetic, paper grain --ar 16:9 --v 7

minimal continuous line art illustration of woman marketer leaning back in chair with satisfied smile, arms behind head, relaxed after completing work, single elegant line drawing with selective bold color fills, color pops in bright blue #2094e0, green #29d296, orange #ff4f01, navy #003282, white background, space for text, no text --ar 2:3 --v 7

The prompt I give Claude to produce these:

I need a Midjourney v7 prompt for [describe the image, the mood, the vertical, the use case].

Write it as a single continuous description, not a list of instructions. Follow this structure:
[style] of [subject], [action], [setting], [specific visual details], [color callouts using hex: bright blue #2094e0, green #29d296, orange #ff4f01, navy #003282], [style anchors], [aspect ratio flag] --v 7

Rules:
- Default to no-hands subjects unless I specify hands
- Use --ar 9:16 for vertical, --ar 16:9 for horizontal, --ar 2:3 for portrait
- Colors by hex, never by name
- Do not invent visual styles. Use the style anchors I provide or default to: flat textured illustration, retro editorial magazine aesthetic, paper grain
- If I name a vertical (healthcare, manufacturing, higher ed, credit union), build the setting in language native to that vertical
- Avoid stock-photo clichés: no smiling teams in glass conference rooms, no lone laptop on a cafe table, no diverse team high-fiving
- The prompt should read as one flowing description, the way the examples above do

Format: [landscape / portrait / square]
Where this will be used: [blog header, LinkedIn, Instagram, newsletter, social]
Reference images I am working from (if any): [describe or attach]

Variations I use: When I want a candid lifestyle look instead of editorial illustration, I tell Claude “use vsco aesthetic, iPhone photo, photo dump, off-center composition, slight motion blur, imperfect framing, subjects not looking at camera.” When I want the continuous line art style (great for newsletter headers and social posts where I need space for text overlay), I tell Claude “minimal continuous line art, single elegant line drawing with selective bold color fills, white background, space for text, no text.”

When it fails: When I am vague. “An image of marketing” generates garbage. “A marketing director at a hospital, mid-30s, looking at a project board on her laptop, late afternoon, slight motion blur, off-center, not looking at the camera” generates something usable. The specificity of the description is the specificity of the image. Every adjective earns its place or it costs you.

20. The Ideogram Prompt Builder (Claude)

Ideogram is where I produce the typographic and data-visual assets: stat cards for social, bold poster-style graphics for LinkedIn, newsletter section headers, and anything where the text is the design. The structure is completely different from Midjourney. Ideogram prompts lead with dimensions and composition, then describe the layout cell by cell, and the typography has to be called out explicitly or it renders illegibly.

I used to spend 20 minutes in Canva building each social graphic. Now I describe what I want to Claude, paste the prompt into Ideogram, and have a branded graphic in under five minutes. For a newsletter that needs four section graphics and three social posts to promote it, that is the difference between an hour and a half day.

Why Claude: Ideogram prompts are layout blueprints. They have to specify dimensions, grid structure, exact text strings, font choices, color assignments per cell, and composition rules, all in the right order. Claude does not forget the logo instruction at the end, which I do every single time.

Here is what finished Ideogram prompts look like. These are the outputs I am training Claude to produce:

1:1 square composition, bold modern editorial poster, navy (#003282) solid background, a 2x2 bento-style grid filling the center of the composition, top-left cell in bright blue (#2094e0) with massive white "94%" and small white text beneath reading "ON-TIME DELIVERY.", top-right cell in green (#29d296) with massive white "12" and small white text beneath reading "ACTIVE PROJECTS.", bottom-left cell in orange (#ff4f01) with massive white "3.2H" and small white text beneath reading "AVG TURNAROUND.", bottom-right cell in cream off-white with massive navy "100%" and small navy text beneath reading "VISIBILITY.", white clean sans-serif headline above the grid reading "The answer is already waiting.", smaller white text at the very bottom reading "Reporting. Workzone.", clean editorial poster.

1:1 square composition, bold magazine editorial poster, navy (#003282) solid background, large bold rounded white sans-serif text filling the upper half reading "Your best process", beneath it in bold orange (#ff4f01) type reading "shouldn't live in one person's head.", a thin bright blue (#2094e0) horizontal divider line, smaller white text below reading "Project templates turn proven workflows into repeatable blueprints.", smaller white text at the very bottom reading "Workzone.", clean editorial poster, type-dominant.

The prompt I give Claude to produce these:

I need an Ideogram prompt for [describe the asset, the format, the headline, the use case].

Write it as a single continuous layout description. Follow this structure:
[Dimensions first], [composition style], [background color by hex], [layout description cell by cell or section by section, with exact text strings in quotes, font callouts, and color assignments for each element], [style anchors]. End with the brand attribution line.

Rules:
- Dimensions: 1:1 square (1080x1080) for social, 16:9 for newsletter headers, 9:16 for stories
- Default font: clean sans-serif (Montserrat unless I specify otherwise)
- Brand colors by hex: bright blue #2094e0, green #29d296, orange #ff4f01, navy #003282
- Typography must be legible at small sizes (Instagram thumbnail, LinkedIn preview)
- Bold, type-dominant, one clear visual hook beats a cluttered design every time
- No competitor brand names in any generated text
- Brand colors should appear deliberately as primary fills, not as accents on white
- If the design is a stat card or data visual, call out each number and its label explicitly
- If the design is type-dominant, describe the text hierarchy: headline size, subhead, body, attribution
- The prompt should read as one flowing layout blueprint, the way the examples above do

Format: [square / landscape / portrait / story]
Where this will be used: [LinkedIn, Instagram, newsletter, social]
The headline or text I want on the graphic: [paste exact text]
Reference assets if I am matching style: [describe or attach]

Variations I use: For stat cards (the bento-grid style), I tell Claude “2×2 or 3-column grid, one stat per cell, massive number with small label, each cell a different brand color.” For type-dominant posters, I tell Claude “purely typographic, no illustration, headline fills the top half, supporting text below a divider line, solid navy background.” For newsletter headers, I switch to 16:9 and add “space for a text overlay on the left third, visual weight on the right.”

When it fails: When the text strings are too long. Ideogram renders text well, but if you give it a 15-word headline it will shrink the type to fit and the graphic becomes unreadable at thumbnail size. Keep headlines under eight words for square formats. If you need more text, move it to the subhead line where smaller type is expected.

How to actually use these so they save you time

Three patterns I learned the hard way.

Save them where you can find them in three seconds. Mine live in one doc I keep pinned to my dock. If a prompt takes longer to find than to rewrite, you will rewrite it. Make your library frictionless. I have tried Notion. I have tried Apple Notes. I have tried a Slack channel I message myself. What works for me is a single Google Doc with anchor links. What works for you might be different. The point is friction is the enemy.

Iterate on the prompt, not the output. When a prompt gives me a weak draft, I do not rewrite the draft. I rewrite the prompt. Five minutes spent making the prompt sharper saves an hour next week. The prompts above are version 17 of each. They started as one sentence. I added constraints over time. Every weak output taught me what to add. Treat your prompt library like a living document.

Treat AI like a junior teammate, not an oracle. The output is a first draft. You are still the editor. The minute you start shipping unedited AI work, you stop being the marketing lead and start being a prompt operator. The way I check myself: would I let my contractor ship this without my review? No? Then do not let Claude ship it without your review either.

Build prompts in conversation, not in isolation. The best prompts come from frustration. You ask Claude for something. The output is wrong. You explain what was wrong. The next output is better. You explain what was still wrong. By the third or fourth round, Claude understands what you actually wanted. That conversation is the prompt. Save the cleaned-up version of that conversation. That is what goes in your library. Prompts I tried to write from scratch were always weaker than prompts I distilled from a working session.

Push back when Claude is sycophantic. If you ask “is this idea any good?” Claude will tell you yes. Always. You have to ask it to disagree. The strategy pressure test prompt above is the one I built specifically because I noticed Claude was telling me my ideas were great when they were not. Build pushback into the prompts you use for decisions. Otherwise you are just hearing yourself agree with you.

Match the tool to the task. I used to run everything through one tool. That was a mistake. ChatGPT is faster at volume. Claude is better at nuance. Perplexity is stronger at research. Gemini lives inside the Google ecosystem. The prompts above are tagged with the tool I use because it matters. A prompt that works perfectly in Claude might produce flat results in ChatGPT, and vice versa. Pay attention to which tool gives you the best output for each type of work.

Frequently asked questions

What AI prompts should a marketing team of one start with?

Start with three: the Sales Call Pain Miner, the Repurpose Engine, and the Multi-Option Generator. The pain miner gives you the language to write everything else. The repurpose engine turns one piece of content into six channels. The multi-option generator stops Claude from giving you one weak answer when you needed five varied ones to choose from. Together, these three save the most hours per week for the least setup effort. Add the others as the workflows come up.

What makes a good AI prompt for marketing work?

Three things. Specificity: name the asset, the audience, the angle, and the constraints. “Write a LinkedIn post” produces generic output. “Write a 250-word LinkedIn post for higher education marketing directors who have tried Asana and bounced back to spreadsheets” produces a usable draft. Constraints: tell the model what to avoid. Banned words, banned constructions, voice rules. The more rules, the less generic the output. Multi-option requests: ask for five versions, not one. The act of comparing variations is where your judgment lives.

Can AI replace a marketing hire?

No. AI does not replace strategy, brand judgment, or the ability to sit in a sales call and read the room. What it replaces is the first draft of everything. The 30 minutes you used to spend on a weekly check-in becomes 10. The half day you used to spend repurposing a blog into social becomes an hour. The first draft of an outreach sequence comes back in five minutes instead of a morning. For a marketing team of one, the value is not “AI does the work.” The value is “AI gets me to a usable starting point so I have time left for the work only I can do.”

How do I write AI prompts that produce usable output instead of generic copy?

Three patterns. First, give the model your voice rules explicitly. Banned words, sentence-length preferences, formatting rules. Without them, the model defaults to generic SaaS voice. Second, give it specific context, not abstract context. “Write for marketing directors” is abstract. “Write for a marketing director who manages a team of two, just lost her senior writer, and has been burned by two previous PM tool rollouts” is specific. Third, treat the first output as a diagnostic, not a deliverable. The first output tells you what your prompt is missing. Fix the prompt and run it again.

What is the best way to organize an AI prompt library for marketing?

One document. Pinned where you can reach it in three seconds. Organized by use case, not by tool. Group the prompts by the moment you need them (“Monday morning planning,” “campaign post-mortem,” “outreach building,” “ad copy production”) rather than by which AI you use. The grouping matters more than the file format. I use a Google Doc with anchor links. Others use Notion databases, Apple Notes, or a private Slack channel they message themselves. Use whatever your hand reaches for fastest.

How is using Claude different from using ChatGPT for marketing work?

Both can do the work below. The differences are stylistic. Claude tends to produce longer, more nuanced first drafts that need less re-prompting on tone. ChatGPT tends to be faster and more aggressive about offering options without being asked. For voice-sensitive work (brand copy, executive ghost-writing, anything that has to sound like a specific human), I lean Claude. For high-volume production work where I need 30 variations of a headline in one shot, ChatGPT can be faster. Perplexity is the strongest option when the task involves research or current data (evaluating a channel, replacing a stat, checking a benchmark). Gemini earns its spot when the question lives inside Google’s ecosystem (GA4, Search Console, Google Ads). The prompts above work across tools, but the “why this tool” notes reflect where I get the best results.

The bigger picture

The work of a marketing team of one will not get smaller. The volume of asks does not slow down because you put in your notice on the second headcount you were never going to get. What changes is how much of your week goes to the first draft of everything. AI shrinks that. The rest you still own.

Some weeks I am ahead. Most weeks I am behind. Every week, these prompts are the difference between “behind but in control” and “behind and underwater.” They are not a substitute for strategy. They are not a substitute for craft. They are a substitute for the unpaid overtime I used to put in just to keep the basics moving.

If you are running a marketing function alone, I hope a few of these save you a Tuesday night. That is the only reason I wrote them down.

If any of these prompts are useful to you, take them. If you want to see how we run the projects behind them, the free trial is open and 15-minute demos are on the calendar.

Last updated on June 3, 2026