The short version: most AI marketing prompts fail because people ask ChatGPT to "be creative" instead of feeding it their own facts, and creativity without facts is just confident guessing. The prompts that work give the AI specific numbers, real customer words, and a clear job to do, then you edit hard. Below are the exact prompts I use every week, plus the structure behind them.
Why the "50 best AI prompts" lists don't help you
I've read a lot of these lists. I've probably written a version of one myself back in 2023 when I was still figuring this out. Here's what I've noticed after two years of using AI daily for client work: the prompt itself is maybe 20% of what makes the output good. The other 80% is what you feed it before you even ask the question.
Ask ChatGPT to "write a LinkedIn post about the benefits of email marketing" and you'll get 200 words that could have come from any marketing blog since 2015. It's not wrong, it's just generic. There's nothing in there that couldn't apply to any business, which means it applies to none of them and does nothing for your audience.
The uncomfortable bit nobody wants to say out loud: if your prompt doesn't include your own data (your open rates, your customer's actual complaint, a number from last month's campaign) then you are asking a machine trained on the average of the internet to sound like you. It can't. It doesn't know you. The fix isn't a cleverer prompt. It's a fuller one.
The prompt structure I use for almost everything
I stopped hunting for magic prompt words years ago and started using a fixed shape instead. It's boring, but it works every time:
- Role - who the AI is acting as (a direct-response copywriter, a HubSpot email strategist, a Brighton-based skincare founder)
- Context - real facts: audience, past results, tone, product, constraint
- Task - one clear output, not five vague ones
- Format - length, structure, where it's going
- Bar - what "good" looks like, ideally with an example of your own past work
Here's what that looks like in practice for a real client, a skincare brand I worked with last year whose Black Friday email open rate had been sitting at 19% for two years:
"You're an email copywriter for a UK skincare brand selling to women 35 to 55 who care about ingredients, not hype. Our last three Black Friday subject lines all mentioned a discount percentage and got 19% opens. Our best-performing subject line ever ('Your skin barrier called, it wants this back') got 34% opens because it was specific and slightly cheeky, not salesy. Write 12 subject lines for this year's Black Friday email using that same specific, slightly cheeky tone. No percentages. Under 45 characters. Flag which three you'd test first and why."
That prompt produced subject lines we used. The generic version, "write me 10 catchy Black Friday subject lines," produced the same tired formulas everyone else's inbox already has: "Don't miss out," "Last chance," "24 hours only."
We tested three of the AI-drafted lines against the old formula over three sends. Open rate moved from 19% to 27% on the first send, then settled around 24% average across the campaign, still a genuine lift on a list of 14,000. Not because the AI is magic. Because we fed it our own winning example instead of asking it to invent one from nothing.
Prompts for content and social that don't sound like AI wrote them
These are the ones I use most often, adjusted for whoever I'm writing for that week.
Turning one idea into a week of LinkedIn posts
"I'm going to give you a 300-word story about [specific event or client result]. Write 5 LinkedIn posts from this single story, each taking a different angle: one contrarian opinion, one practical how-to, one behind-the-scenes, one number-led post, one that asks a genuine question. Each under 150 words. Keep my voice: direct, a bit blunt, no corporate padding. Here's a sample of how I write: [paste 200 words of your own old post]."
Pasting your own writing as the example is the single biggest quality upgrade in this whole post. AI is very good at mimicking a style it can see. It's poor at guessing one from a job title.
If you want the full system I use to turn a single afternoon into a month of content across formats, I've written that up in detail in how I use AI to write a month of marketing content in one afternoon, including the exact folder structure and the order I do things in.
Rewriting a case study so it doesn't read like a brochure
"Here's a rough case study draft with the client's name, the problem, and the result: [paste]. Rewrite this as a story with a beginning, a specific moment where something almost didn't work, and the number-based result at the end. Cut any sentence that could apply to a different client. If a sentence is generic, either make it specific or delete it."
That last instruction, "cut any sentence that could apply to a different client," is doing most of the work. It forces specificity. I've used variations of this prompt to rebuild seven client stories, and you can see the actual before-and-after numbers from those in my AI marketing case studies breakdown.
Ad copy that isn't a wall of adjectives
"Write 8 Meta ad headlines (under 40 characters) and 8 primary text variations (under 125 characters) for [product]. Audience: [specific description, not "everyone"]. Their actual objection, in their own words from a support ticket or review, is: [paste real quote]. Address that objection directly in at least half the variations. No exclamation marks. No 'game-changing' or 'revolutionary'."
Pulling a real customer quote from a review or support ticket and pasting it straight into the prompt changes the output completely. The AI starts writing to a real objection instead of an imagined one.
Prompts for email and lifecycle marketing
If you run anything through HubSpot or a similar platform, prompts get more useful once you connect them to workflow logic rather than one-off emails. I've documented 14 of the actual builds I've shipped for clients in AI-powered HubSpot workflows, but here are two prompts that stand alone.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Abandoned cart sequence that isn't just "you left something behind"
"Write a 3-email abandoned cart sequence for [product type]. Email 1 sends after 1 hour, plain reminder, no discount. Email 2 sends after 24 hours, address the most common reason people abandon this specific product (for us it's shipping cost uncertainty), still no discount. Email 3 sends after 72 hours, small incentive, urgency without fake countdown timers. Each under 100 words. Subject lines under 40 characters."
Re-engagement for a cold list
"I have a segment of 3,200 subscribers who haven't opened an email in 6 months. Write a single re-engagement email that does not apologise for emailing them, does not beg, and gives them one clear reason to stay plus one clear way to leave. Tone: direct, slightly dry, respects their time. Under 120 words."
For a skincare client this exact prompt, tweaked for their product, pulled 340 people back into active engagement out of that 3,200 and let us clean 1,100 dead contacts off the list, which dropped their monthly email sending cost.
Prompts for product pages and ecommerce copy
Product description prompts are where I see the most wasted effort, mostly because people paste in a spec sheet and ask for "compelling copy," which produces the same three adjectives (premium, luxurious, versatile) across every product on the site.
"Here are the specs for [product]: [paste]. Here's one real review from a customer: [paste]. Write a product description that leads with the specific problem this solves, uses one detail from the real review, and ends with a single clear next step. 90 words max. No use of the words 'premium,' 'perfect,' or 'ultimate.'"
Banning specific words in the prompt itself is one of the most useful small habits I've picked up. AI defaults to marketing filler unless you explicitly rule it out. I go into a lot more of these ecommerce-specific workflows, including how to batch this across an entire product catalogue, in AI for ecommerce marketing.
The part everyone skips: organising your prompts so they keep working
A good prompt used once is a nice output. A good prompt saved, refined and reused is a system. I keep a running document of every prompt that's produced something I used, organised by task, and I update the "context" section each month with fresh numbers so it never goes stale. If you're using ChatGPT specifically, I set up folders and saved prompts exactly this way, and I've written the full workflow, including bulk export so client work doesn't live only inside a chat window, in the ChatGPT power-user workflow for writers.
One thing I'd say plainly: if you're new to any of this and terms like "prompt chaining" or "temperature" or "few-shot" make your eyes glaze over, don't push through it out of pride. I put together a plain-English AI marketing glossary for exactly this reason, because half the confusion around AI prompts is just people not knowing what the jargon means, not the prompts being hard.
What " work" really means here
I want to be blunt about something most posts on this topic won't say. A prompt is not a strategy. If your subject lines are weak because your offer is weak, no prompt fixes that. If your LinkedIn posts get no engagement because you've got nothing to say, a better prompt gives you a better-written version of nothing to say. AI prompts improve execution. They do not create a point of view, an offer, or a result you haven't earned yet.
What they do brilliantly is remove the blank page and the eight drafts it usually takes you to find your own tone. That's not nothing. For me it's turned a two-day content backlog into a two-hour one most weeks. But the prompts above only work because I feed them real numbers, real customer words, and real past examples every single time. Skip that step and you're back to generic sludge with better grammar.
Frequently asked questions
What's the single most important thing to add to any marketing prompt?
Your own real data: a past subject line and its open rate, a genuine customer complaint or review, or a piece of your own writing to copy the tone from. Prompts without specifics produce generic output regardless of how cleverly they're worded.
Do longer, more detailed prompts always produce better marketing copy?
Not automatically. Length only helps if the extra words are adding facts (audience, past results, real quotes). Padding a prompt with vague adjectives like "make it engaging and punchy" adds length without adding useful information.
Can AI prompts replace a marketing strategy?
No. Prompts improve how something is written, not whether the underlying offer or idea is worth writing about. A weak offer written brilliantly is still a weak offer.
How often should I update the prompts I use regularly?
Refresh the context section (your numbers, your best recent example) at least monthly. The prompt structure itself rarely needs to change, but stale data in the context makes the output stale too.
Related reading: The AI Subscription Stack: What I Pay For (and What I Cancelled) in 2026 and I Got AI to Chase My Late Invoices. Here's What Happened.
