The short version: Most “AI prompts for marketing” lists fail because they treat the prompt as the whole job, when the prompt is maybe 20% of it. The other 80% is what you feed the AI before you ask it anything. Below are the prompts I use every week, the exact inputs that make them work, and one honest bit nobody selling a prompt pack wants to say out loud.
Why the prompt packs you’ve downloaded haven’t helped much
I’ve bought three of them. Nice PDFs, colour-coded, 50 or 100 “proven” prompts. I used maybe four before I gave up on all three. Not because the prompts were badly written. Because they were generic by design, meant to work for a plumber in Leeds and a SaaS founder in Austin and a jewellery brand in Tel Aviv, and something built to work for everyone works well for no one.
Here’s the bit that’s uncomfortable and mostly gets skipped: the prompt was never the bottleneck. Your business’s lack of a distinct point of view is. If you feed ChatGPT or Claude a bland brief about “a coffee brand for busy professionals,” you’ll get bland copy about busy professionals and coffee, because that’s all you gave it to work with. No clever prompt structure rescues a business that hasn’t decided what it thinks. I’ve written about this before because I keep hitting the same wall with clients: the AI sameness problem isn’t an AI problem at all. It’s a “we never wrote down what makes us different” problem, and the AI just makes that gap visible faster than a human copywriter would have.
The one input that changes everything (and takes 15 minutes)
Before I ask any tool to write anything, I paste in what I call a voice and fact block. It’s not a brand guideline document. It’s rough, specific, and about 300 words. Mine looks roughly like this:
- Who I am: 53, UK-based, ex-influencer, Forbes-listed, rebuilding my consultancy in public after five hard years, blunt, warm, British.
- What I believe that most people in my industry won’t say: that most AI marketing content fails because founders skip the thinking and go straight to the writing.
- A real number or story from this week: e.g. “I sent 200 cold emails last month using a personalised opener and got 11 replies, versus 3 replies from 200 generic ones the month before.”
- Words I never use: use, unlock, delve, handle, and anything that sounds like a corporate training video.
- The one thing this piece has to say that a competitor’s version wouldn’t.
That block goes in before every single prompt. Without it, I get competent, forgettable marketing copy. With it, I get something that sounds like me on a good day. This is the whole answer to why AI-written content so often sounds identical from brand to brand: nobody’s feeding it anything to disagree with. If you’ve ever read a piece and thought “this could have been written by anyone,” that’s exactly what’s happening, and I’ve gone into more detail on why AI-written marketing sounds exactly like your competitor’s in a separate piece, because it’s worth understanding before you touch a single prompt.
The prompts I use, word for word
1. The LinkedIn post that doesn’t sound like a LinkedIn post
“Here’s a story from my week: [paste 3 to 5 sentences, raw, unedited]. Write this as a LinkedIn post in my voice from the block above. Open with the specific detail, not a hook question. No emojis. No ‘here’s what I learned.’ End with an opinion, not a call to engage.”
The line “no hook question” matters more than people think. Ask an AI for a LinkedIn post and nine times out of ten it opens with “Ever wondered why…?” It’s the single most obvious tell that something was written by a machine, so I ban it explicitly, every time.
2. The email subject line batch
“Give me 15 subject lines for this email [paste email]. Five should be under 30 characters. Five should sound like a text from a colleague, not a marketer. Five should include a specific number from the email itself, not a made-up one. Do not use ‘you won’t believe’ or any variation of urgency language.”
I ran this exact prompt for a client’s monthly newsletter over three months in late 2025. Open rates went from 18% to 31% across the list of just under 9,000 subscribers, mostly because the “sounds like a colleague” batch kept winning the A/B tests. The AI didn’t invent that insight. I told it what to test because I already suspected our subject lines were too polished. The prompt executed the idea. It didn’t have the idea.
3. The blog outline that isn’t a listicle by default
“I want to write about [topic]. Before you outline anything, ask me three questions about my actual experience with this topic that a generic outline wouldn’t cover. Then build the outline around my answers, not around what usually ranks.”
This one prompt has stopped more bad blog posts than anything else on this list. Left to its own devices, AI defaults to “5 tips for X” structures because that’s what’s statistically common in its training data. Making it ask you questions first forces in the specific detail that makes a piece worth reading, and worth citing, rather than one more rehash of the same five tips everyone else already published.
4. The competitor gap prompt
“Here are the top 3 ranking pages for [topic]: [paste URLs or summaries]. What do all three avoid saying, and what’s one true, slightly uncomfortable thing about this topic that none of them mention?”
I use this before I write almost anything now. It’s not about being contrarian for the sake of it. It’s about noticing that most content on a given topic quietly agrees to avoid the same uncomfortable point, usually because saying it doesn’t help sell whatever’s being sold on that page.
5. The B2B outreach message that isn’t a template
“Write a first-line opener for a cold LinkedIn message to [role] at [company]. Use this specific detail about their business: [paste something real, a recent post, a product launch, a job listing]. No compliments about their ‘impressive career.’ One sentence only.”
For more depth on this specific use case, including the exact structure I use for sequences rather than single messages, I’ve laid out full B2B-specific prompt templates in a separate guide, because B2B outreach has its own rules that consumer marketing prompts don’t cover.
The mistake almost everyone makes with “write like me” prompts
Somewhere in 2024 or 2025, “write in my voice” became the default instruction people added to every prompt, as if typing that sentence would somehow transfer your personality into the output. It doesn’t. I tried it for months, pasting three or four of my old blog posts and typing “write like this” before every request, and got copy that used my sentence rhythm without saying anything I’d think. Style without opinion is just an impression. It’s a very good impression of someone saying nothing.
What works instead is feeding the AI a specific belief or fact for that particular piece, every single time, rather than relying on a stored voice to carry the whole job. I go into the mechanics of why this distinction matters, and what to do about it, in a piece on why “write like me” prompts fail for small business content. Short version: voice is tone, opinion is content, and no prompt fixes the second by adjusting the first.
A real example of a prompt saving me time (not just producing words)
Back in September, I had a webinar to promote in four days and no email sequence written. Normally that’s a half-day job of drafting, rewriting, second-guessing subject lines, the usual. I used this sequence:
- Pasted the webinar description, the actual registration numbers so far (41 signups against a target of 200), and my honest read on why signups were slow (“the title’s too vague, it doesn’t say who this is for”).
- Asked for three different email angles built around that specific problem, not generic “reminder” emails.
- Picked one angle, asked for a three-email sequence with different subject lines for each, spaced across four days.
- Edited by hand for maybe 20 minutes total, mostly cutting things that sounded too smooth.
Total time from blank page to three scheduled emails: 47 minutes. Registrations finished at 218. I’m not crediting the AI with the outcome, the fix was noticing the title problem, which is a marketing judgement call, not a prompting trick. But the tool turned a judgement call into finished copy in under an hour instead of half a day, and that’s the actual value. Speed on execution, not replacement of the thinking.
What a useful prompt structure looks like
If you strip away everything else, every prompt that’s worked for me follows roughly the same shape:
- A specific fact, number, or story that only I could have supplied.
- A clear instruction on what to avoid, by name, not just “make it engaging.”
- A format constraint (length, structure, number of options) so you’re not left editing a wall of text into shape yourself.
- Permission, explicitly stated, to have an opinion rather than stay neutral.
Compare that to a typical downloaded prompt pack entry: “Write an engaging Instagram caption for [product] that drives engagement and reflects our brand voice.” There’s nothing in that sentence the AI can use except guesswork, so it fills the gap with whatever’s statistically common for captions in that category. That’s why two businesses using the same prompt pack end up with copy that reads almost identically, just with different product names swapped in.
Where this breaks down (the bit worth admitting)
Prompts don’t fix a boring product, a confused offer, or a business that’s undifferentiated. I’ve sat with clients who wanted the perfect prompt to make their marketing “pop” when the real issue was that their service was identical to four competitors down the street and nobody had ever articulated why a customer should pick them specifically. No prompt writes a point of view that doesn’t exist yet. That conversation has to happen first, on paper, with a human, before any AI touches it. It’s the part of AI marketing consulting that’s hardest to package into a template, and it’s also the part that moves revenue, which is why brands that skip it end up with beautifully written content that converts at the same flat rate as before they started using AI at all.
It’s also worth saying that great brand voice work existed long before generative AI and still matters more than the tool. Look at how a brand like Twitch built an identity that felt unmistakably theirs, chaotic, community-first, unpolished on purpose. I broke down the mechanics of that in a separate piece on Twitch’s marketing strategy, and the lesson holds here too: distinct voice comes from consistent decisions made by people, not from a better prompt template.
A simple weekly routine that keeps this from becoming another abandoned system
I keep one running document, updated every Monday, with three things: one specific number from the past week (an open rate, a reply count, a sales figure), one honest observation about what’s not working, and one thing I believe about my industry that I haven’t said publicly yet. That document is the input for every prompt I run that week. It takes ten minutes on a Monday and it’s the single habit that’s kept my AI-assisted content from drifting into the same flat, forgettable tone that most business content has settled into. The prompt is the last step. The Monday document is the actual work.
Frequently asked questions
What’s the best AI prompt for marketing content in general?
There isn’t one, and any list claiming there is one is selling you convenience over results. The prompts that work are built around a specific fact, number, or opinion from your own business that you feed in first, then a clear instruction on format and what to avoid. Swap the specific fact and reuse the structure.
Why does AI-generated marketing copy sound the same across different businesses?
Because most people give it the same generic brief, so it fills the gap with whatever’s statistically common rather than something specific to that brand. The fix isn’t a cleverer prompt, it’s feeding in a real number, story, or opinion that a competitor couldn’t supply.
How much time can AI prompts save on marketing tasks?
In my own experience, a full email sequence that used to take half a day dropped to under 50 minutes including edits, once I had a clear input to feed the tool. The saving is almost entirely in execution speed, not in replacing the strategic decision about what to say.
Can AI prompts fix marketing that isn’t performing?
No, not on their own. If the offer, positioning, or product is undifferentiated, better prompts produce better-written versions of the same underlying problem. The thinking work, deciding what makes the business different, has to happen before any prompt gets typed.
Related reading: Monzo Marketing Strategy: How They Built a Brand That Wins and How to Reverse-Engineer Any Competitor’s Marketing Strategy Using AI (Without Paying for Fancy Tools).
