MIA Event Recap: AI Prompt Frameworks & Templates for Marketers
A live MIA workshop on building prompt systems, human-sounding AI writing, and real GTM use cases for marketers and founders.
Thanks for reading MIA—everything you need to put marketing into action—templates, frameworks, community, data, insights, case studies, ideas, and martech/GTM tools. Share it with your network! ~James
Most marketers did the same thing when ChatGPT went mainstream. We opened a new chat, typed a generic input, and hoped magic would happen. Sometimes it did. Most times, it spat out something that sounded polished, confident, and painfully generic. The problem was never that AI “couldn’t write.” The problem was that we treated a system like a one-off trick.
In this live MIA workshop, I walked through how I moved from vibes-based prompting to building proper prompt systems. We talked about communication basics, the five ingredients of a strong prompt, and why “garbage in, garbage out” still applies when your interface talks back in full paragraphs. Then we layered on the nine-step prompt framework, agent flows that think in sequences, and real examples from positioning, messaging, and landing page projects. If you want AI to sound more like you and less like AI, this session is for you.
Full recap here:
Key Takeaways
1. Prompt engineering is communication, not magic
Most marketers do not have a “prompt problem.” They have a communication problem. The output you get still follows the old rule: garbage in, garbage out.
In the workshop we reframed prompting as a simple communication exercise. Be explicit about what you want, who you are, who the audience is, and what “good” looks like.
When you treat prompts like briefs instead of wishes, your AI tools start behaving more like collaborators than slot machines.
2. Five ingredients make prompts more reliable
Good prompts share the same backbone.
Clarity: clear task and outcome, not “help me with content.”
Context: who you are, what you are working on, why it matters.
Constraints: length limits, tone rules, “avoid this, do more of that.”
Continuity: building on previous steps and chat history.
Consistency: reusing structures and variables so you can move faster.
Once you start writing with these five in mind, you get less fluff, fewer hallucinations, and results that feel closer to your real voice.
3. Remove the “AI accent” from your writing
AI outputs often expose themselves with certain habits. Over-dramatic constructions. Awkward em dashes. Overused phrases that sound like a generic keynote.
In the session, we walked through common tells, then used an “AI writing humanizer” project to score and clean drafts.
The goal is not to hide that you used AI. The goal is to ship copy that feels like a person wrote it, with specific details, grounded examples, and simple sentence structures.
4. The 9-step prompt framework
Behind my 40+ custom GPTs is the same skeleton.
Role: who the model should act as.
Task: what it should do right now.
Goal: the outcome you care about.
Audience: who this is for.
Style: tone, pacing, point of view.
Constraints: word limits, no-go patterns, structure rules.
Inputs and variables: fields you will keep swapping in and out.
Example output: a sample that shows “this is what good looks like.”
Formatting: headings, bullets, tables, or sections you want.
Once you set this up once, you can clone it for blog posts, landing pages, onboarding emails, and more.
5. Think in agent flows instead of single prompts
We also explored “agent mode,” where you stop asking for one-step outputs and start describing sequences.
For example: “research five competing SaaS launches, extract positioning angles, summarise patterns, then draft a launch blog post, LinkedIn post, and email”.
A good agent prompt thinks like a checklist. Research, summarise, create, refine. That mental shift turns AI from a writing assistant into an operations partner that can handle more of the grunt work.
6. Projects, custom GPTs, and assistants keep context alive
Instead of starting from zero every time, we walked through how to use projects, gems, and assistants to store your prompts and knowledge files.
In practice, that means: your brand docs, examples of “good” copy, and prompt frameworks all live in one place.
When you open that project, the model already understands your world. From there you can spin out custom GPTs for public use or connect assistants to tools like Clay, Notion, or internal dashboards through APIs.
7. Real GTM use cases make it stick
Frameworks only matter when they survive contact with real work. We applied everything to a startup called Komfot Health.
Using deep research, we pulled together customer insights, market context, and competitor analysis, then fed that into prompt projects for positioning, messaging, landing pages, lifecycle campaigns, and case studies.
The result was a full product marketing stack driven by AI but shaped by human judgment. You can run the same play for your own product with your own research.
8. Build your own prompt library, not just a folder of chats
Finally, we ended on process. Instead of scattered chats, build a prompt library: a simple index of your best frameworks, links to associated custom GPTs or projects, and examples for each use case.
Over time this turns into an internal “AI playbook” for your team. New hires do not start from scratch. Freelancers can plug into your system. And you spend more time refining prompts that already work than chasing new tricks every week.
Soundbites
“Your prompt is a brief, not a wish.”
“If your instructions are fuzzy, your AI will be too.”
“Most AI copy fails, not from bad models, but from lazy prompts.”
“Clarity, context, and constraints will outwork any clever phrase.”
“The nine-step prompt framework frees you from starting from zero.”
“Agents are just prompts that think in sequences instead of sentences.”
“Projects and custom GPTs are how you give AI a memory.”
“You can not fix bad positioning with a better prompt.”
“A good prompt library is a growth asset, not a nerdy side project.”
“AI should sound like you on your best writing day, not like a corporate blog.”
🧩 Get the full AI prompt library.
Get access to 40+ Custom GPTs and AI-assisted marketing systems that’ll help you get better results, every single time. Get access here.
If you want to collaborate or feature your product in this newsletter, check out our media kit, and shoot me an email.
MIA is trusted by 2,000+ marketers across Africa, the US, and Europe, helping you put marketing in action better, faster, and with context.
We’re building a village of marketers and founders who want to increase their odds of succeeding at marketing. You’re one of us. Share this newsletter with them.
Thanks for writing this, it clarifies a lot. As someone teacing computer science, the "garbage in, garbage out" principle you highlight resonates so much; it's fascinating to see how basic communication skills are becoming central to leveraging AI, rather than just technical prowess. I'm curious if you see these prompt systems evolving to be less dependent on human 'engineers' in the future, or if that communication bottleneck will always remain? Your ability to explain this so clearly is realy impressive.