How marketers can write (much) better AI prompts

I’ll be honest: I’ve also typed things like “make this better” or “fix this draft” when ChatGPT was first released and hoped it would hand me something polished, sharp, and expert. And guess what: it didn’t.

We already know AI isn’t magic, but a pattern engine. What comes out depends entirely on what you put in.

Which means if you want better results, you need to get better at prompting.

This isn’t about becoming a so-called “prompt engineer”, though I love the concept. It’s rather about being a more effective marketer, someone who knows how to brief a tool clearly, guide the output, and turn a rough draft into solid copy fast.

And that’s what this article is about: how to write prompts that actually deliver what you’re looking for, from my own experience battling with the tools and models that have been out there since 2022.

Why prompt writing is now a core marketing skill

AI is now part of the day-to-day for most marketing teams, whether they’re generating content, testing messaging, researching competitors, or repurposing assets for different channels.

And while the tools are powerful, they only work if you know how to use them.

Prompt writing is the interface. Think of it like the bridge between your strategy and what AI actually produces. If your prompt is vague, generic, or missing key context, the output will reflect that. You’ll get something that technically fits the request but misses the mark in tone, depth, or direction.

In other words: garbage in, garbage out.

The difference between a forgettable headline and one that gets clicks, or between a generic LinkedIn post and something that drives real engagement, often comes down to how the prompt was written. The better the input, the less editing you have to do. The clearer your request, the more usable the result.

And it’s not just about saving time. Prompting well gives you leverage. You can produce first drafts faster, explore more creative angles, and scale content across platforms without sacrificing quality. You’ll also spend less time wrestling with the tool and more time focusing on what actually matters: message, impact, and results.

What makes a good AI prompt

If you’ve ever used an AI tool and thought, “This just isn’t very good,” the issue probably wasn’t the model. It was the prompt (aka you).

A good AI prompt doesn’t just tell the tool what to do. It gives it structure, context, and purpose. Without that, you’re relying on default behavior, which usually results in vague copy, generic ideas, or tone-deaf messaging.

The fix is simple: stop thinking of prompting as a command and start thinking of it as a brief.

Here’s what every good AI prompt includes:

1. A clear task

Start with a specific action. Do you want the AI to write, summarize, brainstorm, outline, rewrite, or compare? Be direct. Avoid open-ended requests like “create content” or “help with copy.” Say exactly what you need.

Example:

Bad: Make this shorter

Better: Summarize the following paragraph into 3 bullet points for a landing page

2. Context for the output

AI doesn’t understand the bigger picture unless you tell it. Where will this content live? What’s its purpose in the funnel? Who is it competing with?

Example: Write a blog post intro for a B2B SaaS company targeting operations managers. The article is about reducing churn through better onboarding. It will appear on the company blog and be promoted via LinkedIn.

That kind of context gives the tool a direction, not just a task.

3. Audience insight

Most prompts fail because they ignore the reader. If you don’t tell the AI who it’s writing for, it will guess. Usually wrong.

Be clear about who you’re targeting, how familiar they are with the topic, what they care about, and what tone they respond to.

Example: Audience: Heads of Marketing at fast-growing startups. They’re time-strapped, data-driven, and skeptical of fluff.

This changes the entire shape of the output.

4. Tone and style

Don’t assume the AI knows how your brand sounds. If you want output that matches your tone, you have to name it or describe it.

You can reference your voice guide, share examples, or give short descriptors like “professional but casual,” “playful and bold,” or “authoritative and concise.”

Example: Tone: Confident and direct, like someone giving expert advice in a meeting. Avoid filler, jokes, or vague hype.

Even better: show the model a few sentences that already sound like your brand.

5. Format and length

If you need a 50-character subject line or a 200-word summary, say so. The model won’t guess formatting unless you specify it.

Examples:

• Give me 3 subject line options, each under 50 characters

• Write a 4-sentence product description using sentence case and no technical jargon

6. Constraints

This is my favourite part. What rules, limitations, or brand guidelines should apply? If you have specific rules, state them upfront. This includes brand phrases, banned words (most importantly, banned em dashes!), formatting styles, voice consistency, and platform-specific limitations.

Examples:

• Don’t use the phrase “game-changer”

• Avoid technical jargon — write for someone non-technical

• Use sentence case for all headlines

• Limit response to 150 words max

Constraints give the AI clear boundaries, which usually improves the quality of the output.

A well-written prompt is like a creative brief. It focuses the output without boxing it in. And once you’ve practiced a few times, writing good prompts becomes second nature.

TLDR: The 6 elements of a strong AI prompt

Use this quick checklist for a refresher:

1. Task: What do you want the AI to do? (write, summarize, rewrite, etc.)

2. Context: Where will this be used? What's the goal or situation?

3. Audience: Who is it for? What do they care about?

4. Tone and style: How should it sound? Any voice or brand rules?

5. Format: What output structure do you need? (headlines, bullet list, email draft)

6. Constraints: Any limitations, banned words, or brand requirements?

You don’t have to include all six every time, but the more you cover, the less cleanup you’ll need later.

Prompt like a strategist: best practices for marketers

Knowing how to structure a prompt is one thing. But applying it consistently, across channels, under deadlines, with variable tools, is where most marketers hit friction.

This section is about reducing that friction. Here’s how to move from “OK prompt” to “I barely had to edit this.”

1. Start specific, not creative

You’re not briefing a copywriter. You’re briefing a model. Don’t lead with metaphors or vague instructions like “make it pop.” Lead with structure.

Start by describing what the content is, who it’s for, and what success looks like.

Bad prompt: Write a homepage headline

Better prompt: Write a homepage headline for a B2B SaaS company in the cybersecurity space. Audience: IT leaders at mid-market companies. Goal: Increase demo bookings. Tone: authoritative but clear.

The second one will save you 3 rewrites.

2. Assign a role

One of the most effective ways to improve AI output is to frame the prompt with a role. It anchors the response in a perspective and raises the quality immediately.

Examples:

• Act as a B2B copywriter with experience in conversion-focused landing pages.

• Write like a content strategist preparing thought leadership for a fintech CMO.

• Take the role of a social media manager creating a Twitter thread for SaaS founders.

This trick works in almost any context, and is especially helpful when you need tone consistency.

3. Use layering, not single-shot prompts

Don’t try to get everything in one go. Break down complex outputs into small steps:

1. Start with a rough draft

2. Ask the AI to rewrite it in your brand voice

3. Then prompt it to shorten, reformat, or create variants

Think of AI like an intern: great at following clear steps, but likely to mess up if you throw 10 tasks at once.

4. Save your best prompts and iterate

Every marketer should have a personal “prompt bank”, the prompts that reliably produce good results for common tasks.

These should be stored somewhere easy to access (Notion, Google Docs, whatever works), grouped by task:

• Subject lines

• Email intros

• LinkedIn posts

• Blog outlines

• SEO meta descriptions

Every time you tweak a prompt and it works better, save the updated version. This turns prompting into a repeatable skill, not a guessing game.

5. Build reusable templates

Templates aren’t just for the AI. They’re for you. If you often ask for similar outputs (like ads, CTAs, or product descriptions) create prompt templates that include all six building blocks.

Example template: Write a [format] for [context]. The audience is [persona], and the tone should be [tone]. Keep it under [length or word count]. Avoid [constraints].

Filling that in takes less than 60 seconds and usually gives you something you can work with on the first try.

Common AI prompting mistakes (and how to avoid them)

These are five of the most common prompt mistakes I've seen consistently repeated with other colleagues and team members, plus how to fix them fast.

1. Being too vague

I already mentioned this one: if your prompt reads like “make this better” or “write something engaging,” you’re handing AI a blank canvas and hoping it guesses right.

✅ Fix it: Add clarity. Define the format, goal, tone, and audience.

Instead of: Write something about our product launch

Try: Write a short LinkedIn post (under 200 words) announcing our new product feature for busy marketing ops managers. Tone: confident and direct. Highlight the time-saving benefit.

2. Skipping audience details

If you don’t mention who the content is for, the AI will default to general. That leads to content that feels flat, safe, and forgettable.

✅ Fix it: Be specific about the audience’s role, industry, goals, and mindset.

Instead of: Write a subject line for our new webinar

Try: Write a subject line for a webinar targeting revenue leaders at B2B SaaS companies. Focus on how they can shorten sales cycles without adding new tools. Keep it under 60 characters.

3. Asking for too much in one go

Trying to get an outline, intro, CTA, and headline from one prompt will usually result in a mess. The model doesn’t know what to prioritize and gives you a watered-down version of everything.

✅ Fix it: Break complex tasks into smaller steps. Ask for one format at a time.

Instead of: Write a blog outline and the first two paragraphs and give me headline options

Try: Start by outlining a blog post for mid-market ecommerce marketers on how to reduce abandoned carts using personalized product pages. Keep the outline to 5 main points. We’ll write the intro next.

4. Leaving out tone and brand voice

Tone is one of the easiest things to guide in a prompt, and one of the easiest to overlook. If your brand sounds polished and your output sounds robotic, this is probably why.

✅ Fix it: State the tone in plain language or reference an example.

Instead of: Write a product description

Try: Write a benefit-led product description for our new automation feature. Tone: conversational and confident, similar to how we write on our homepage. Use contractions and avoid technical language.

5. Taking the first draft as final

Even with a great prompt, the first version is rarely perfect. Many marketers take the first output, edit a bit, and move on, but the real power of prompting comes from iteration.

✅ Fix it: Ask the AI to rewrite, sharpen, expand, or adjust. It works like a feedback loop.

Example prompt: This is close. Can you rewrite it to be more concise, use stronger verbs, and remove any filler?

Or: Now give me three variations with different tones: one more playful, one more direct, and one more professional.

Prompt frameworks you can actually reuse

Writing good prompts on the fly is possible, but having a few go-to templates saves time and helps you stay consistent across different types of tasks. Whether you’re generating social content, product messaging, or thought leadership, these three frameworks will give you a reliable starting point.

Each one includes a breakdown and a sample prompt you can copy and adapt.

Prompting framework 1: RACE

Best for: General writing tasks where you need structured output

• Role. Who should the AI act as?

• Action. What do you want it to do?

• Context. What background should it know?

• Execution. What should the final output look like?

💡Example prompt: Act as a B2B content marketer. Your task is to write a blog intro. Context: the post is about how SaaS companies can reduce churn with better onboarding. The audience is growth-stage startup teams. Tone: professional but approachable. Format: 3 sentences max.

Prompting framework 2: DREAM

Best for: Creative tasks like messaging, ideation, and campaign angles

• Describe the goal. What are you trying to achieve?

• Refine the audience. Who are you targeting?

• Establish tone. How should it sound?

• Add references. Is there something to mirror or draw from?

• Make the request. What should the AI deliver?

💡Example prompt: The goal is to create a short, punchy value prop for a new B2B tool that helps sales teams write better emails. Audience: SDR managers in the US. Tone: casual but confident. Reference: the kind of language Gong or Lavender uses. Please generate five variations, each under 15 words.

Prompting framework 3: PASTA

Best for: Repurposing existing content into new formats.

• Primary asset. What are you starting with?

• Audience. Who’s the repurposed content for?

• Style. What tone or format should it match?

• Transformation. What do you want to turn it into?

• Action. What should the audience do afterward?

💡Example prompt: Here is a transcript of a podcast episode on marketing analytics [paste excerpt]. Repurpose this into a LinkedIn carousel targeting mid-level marketers at SaaS companies. Use a direct and engaging tone. Each slide should focus on one clear takeaway. End with a CTA to download our new guide.

You don’t have to memorize these. You just need to recognize the parts and plug in what matters. The idea is to avoid blank-page syndrome and reduce trial and error by giving AI a consistent structure to work from.

When to use AI, and when to step in yourself

Prompting well is powerful, but it doesn’t mean AI should handle everything. It’s a tool that speeds up execution and unlocks ideas, but it still needs your input, judgment, and direction. Knowing when to use it and when to write something yourself is what separates efficient marketers from lazy ones.

Here’s a clear breakdown of what to automate, what to guide, and what to own.

When to use AI 🦾

1. Drafting: AI is great at writing quick first versions: blog intros, subject lines, LinkedIn post ideas, meta descriptions. It won’t always get the final tone right, but it gives you something to work from.

2. Ideation: Need five different ways to position a benefit? Or a dozen options for a headline? Prompting the AI to generate variations gives you more angles to explore in less time.

3. Repurposing: This is where AI shines. Take a long blog post and ask for five Twitter hooks, or feed in a podcast transcript and get a list of key points for a carousel. It saves hours on manual edits.

4. Personalization at scale: You can prompt AI to create copy variations by segment, industry, or job title, especially helpful for outbound, ads, or localized landing pages.

5. Cleanup: Rewriting for tone, shortening long paragraphs, simplifying technical copy, or fixing structure. This works well when you already have the content but need it cleaner and tighter.

When to step in yourself 🦸

1. Strategy and positioning: AI can’t create your messaging framework. It can help refine ideas, but it won’t decide what your product actually solves or how it should be positioned in the market. That’s your job.

2. Emotional resonance: If your copy needs to land emotionally, build trust, or make someone feel something, do it yourself. AI can mimic tone, but it doesn’t understand context or nuance the way you do.

3. Ethical judgment and sensitivity: AI has no context for what’s legally risky, socially inappropriate, or politically tone-deaf. If the content touches anything remotely sensitive, like health, finance, or social issues, it needs a human pass.

4. Thought leadership: Real expertise comes from people. You can use AI to help shape or structure your thinking, but original insights, perspective, and real-world examples need to come from you.

5. Final edits: Never publish AI output without reading it critically. Check facts. Edit for tone. Refine the message. Remove the damn em dashes. AI gives you a solid draft, not a finished product.

If you’ve made it this far, you should know this article was partially written with AI, but every insight was carefully researched and fact-checked by a human 😉