Teach AI to Write Like You: 3 Context Files + Prompts
Three markdown files that teach AI how you create content, who you create it for, and what your version of good looks like. Build these files once and use them in every future session to see better output.
Create, optimize, and measure content that actually drives leads with Content Hub.
Today, AI is a genuinely capable content creator, and that's a problem for marketers.
Everyone is creating content with the same tools reflecting the same writing patterns, structures, and not-so-hot takes. The result is a sea of “commodity content” that lacks perspective and a distinct point of view.
The good news? AI models aren’t lacking in the capability or knowledge to create better content. They’re just missing you — your judgment, unique perspective, and definition of quality.
Below, you'll find three context files to create a personalized creative context system. It’s a knowledge base that teaches your AI how to think, write, and create like you.
What is Personalized Creative Context System?
Kieran Flanagan built this system to solve a problem he kept running into: AI would produce technically competent content that sounded nothing like him. The fix was three simple files that give the model the context it was missing.
A folder called Creative Context containing three markdown files:
- creator.md — How you think and how you create
- audience.md — Who you create for and what they care about
- taste.md — What you think is actually good
Load these files into your AI tool (Claude, Codex, ChatGPT, or your preferred environment) and you'll have a stronger foundation for future sessions.
File One
Creator.md
How you create content
-
What it contains:
Available
- Core thinking patterns
- Writing style and tone
- Recurring formats
- What makes you, you
- Platform-specific notes
How to build it:
The best input is your actual work. Export your LinkedIn posts, Substack archive, or any other content you've published. Aim for at least five pieces of content that ideally performed well and feel distinctly yours.
I'm going to give you 5–10 pieces of my best content. Your job is
to extract the constants — the things that stay true across all of
it regardless of topic or platform.
Look for:
- How I build and structure arguments
- Recurring ways I explain ideas
- My tone and sentence style
- Recurring creative formats or modes I use
- What makes my work distinct from generic content on the same topics
- The North Star: what I'm always trying to preserve in my writing
For each platform I create for [LinkedIn / Substack / YouTube —
list yours], note how my style adapts while staying true to the core.
Do not just summarize what I wrote about. Extract how I write and
think. The goal is a file my AI assistant can read before any
content session so it creates things that sound like me, not like
an average of the internet.
Here is my content:
[paste your 5–10 pieces]
💡 If you're starting from scratch, Kieran recommends copywork — find a writer whose style resonates with you, use their content as your training set, and work with AI to adapt it toward your own voice. You still have to do the work of developing your perspective but this gives you something to react to.
File Two
Audience.md
Who you create for
-
What it contains
Available
- Detailed audience profiles
- What they care about
- What they already know
- Their main challenges
- Content angles that resonate
How to build it (three inputs):
1. Five Dream Readers
Pick five real people whose profiles represent your ideal audience. Grab their LinkedIn or X profiles and give them to your AI assistant.
Here are the LinkedIn/X profiles of 5 people who represent my ideal audience. Analyze these profiles and extract:
- Their professional context and priorities
- What problems or challenges they're likely navigating
- What they already know and don't need explained
- What would earn their trust
- What content angles would feel genuinely useful vs generic to them
Build a detailed audience profile from these five people that I can use to inform my content creation.
[paste profiles]
2. Real conversation data (optional but powerful)
If you have access to comments, DMs, survey responses, or community discussions from people who already engage with your content, this is your best signal.
Feed it to your AI assistant and ask it to extract audience characteristics.
💡 Important note: Kieran estimates roughly 60% of LinkedIn comments are now AI-generated. If you're using comment data to build your audience profile, be selective about what you include — generic engagement comments are likely to skew the picture.
3. External research
Ask your AI to research what your target audience currently cares about — their biggest priorities, recurring frustrations, what they're tired of hearing, what would feel novel to them.
I create content for [describe your audience in 1–2 sentences].
Research this audience using public sources — LinkedIn posts, Reddit, practitioner discussions, industry publications. Help me understand:
- Their current biggest priorities
- Recurring frustrations and problems
- What they already know (what I don't need to explain)
- What they're tired of hearing
- What earns their trust
- What would feel genuinely novel or useful to them right now
I already have this audience context: [paste any existing audience description you have, or leave blank if starting fresh]
Enhance or build my audience.md file from this research.
File Three
Taste.md
What content you think is good
-
What it contains:
Available
- Core taste principles
- What makes writing good to you
- When to use frameworks
- How to balance contrarianism
- Simplicity and clarity standards
- Publish test: could the reader explain the core idea in under 10 minutes?
How to build it:
You can’t ask AI to define your taste, so the manual work you put into creating this file is essential.
You can start with a list of content you love and can explain why it's good. Kieran's list included:
- Positioning by Al Ries and Jack Trout
- Kevin Kelly's 1,000 True Fans
- Paul Graham's Do Things That Don't Scale
- Ben Thompson's Curation and Algorithms
- Claude Hopkins' scientific copywriting
The why behind this list matters. Explain to AI what specifically makes them good for your audience and why they resonate with you as a creator.
I'm going to give you a list of content I love and explain why each piece is good. Your job is to interview me with the following 5 judgment questions, then use my answers and the reference list to build my taste.md file.
The 5 questions:
1. New idea vs elegant writing — which matters more to you?
2. How should contrarianism work — strong argument, or underlying mechanism even if not fully proven?
3. When is a framework useful?
4. How much polish matters vs raw thinking?
5. Simplify or keep nuance — where do you draw the line?
Ask me these one at a time. After I answer all five, combine my answers with the reference list below to produce my taste.md file.
It should include:
- Core taste principles
- What makes writing good to me
- What I never want in my content
- A publish test I can run on any piece before it goes out
My reference list: [paste your list with explanations of why each is good]
How to use this system
Once all three files are built, load them into your AI assistant at the start of any content session. A few ways Kieran uses his:
For ideation: Ask your AI to read your audience.md file and research what's happening across subjects your audience cares about right now. Ask it to return ideas with: the signal (what's happening), why your audience cares, your angle (based on your creator.md), and a thesis. Pick the one that resonates and develop it.
For drafting: Generate a first draft, then explicitly ask your AI to run it through your taste.md file and show you what it would change. The taste.md pass is where the output gets sharp. Kieran's example: a hook that read "nearly all marketers" became "Marketing teams aren't just getting smaller, they're changing shape" after a taste.md pass.
For improving underperforming content: Give your AI a post that didn't land the way you expected and ask it to run through your taste.md file and tell you specifically what to change. It won't just give you generic feedback — it'll diagnose against your own standards.