Write better prompts with Kayfabe
Kayfabe is a Professional Wrestling term. It is the scripted world the performers operate in, the motives, expectations and pre-determined goals of a character.
How Kayfabe prompting works
Kayfabe prompting is a clear, structured and modular way of writing prompts that forces an LLM to stay inside a defined frame.
You set create the world and the script and the model responds as if it's real.
Kayfabe reduces drift, prevents hallucination, and produces cleaner and more predictable outputs.
The Kayfabe model
The principles of Kayfave are defined roles, clear outcomes and efficient structures.
Structuring a prompt in a concise and organised way reduces the amount of processing needed which results in efficient and useful outcomes.
The Kayfabe model uses modular Prompt Blocks to help AI models stay focussed.
Use the Kayfabe Prompt Blocks to write your own prompts and start seeing precise results immediately.
Kayfabe Prompt Blocks
Blocks are defined by headers. Although your AI tool of choice doesn't recognise formatting, the capitalised headers with curly brackets help you structure your prompt.
Use these as the backbone of your task. Copy and paste them and replace the dummy text. Remember, clear and detailed prompts written in a concise way result in the best outcomes.
Follow this structure:
- Role
- Task
- Audience
- Context
- Constraints
- Structure
The blocks
Role
Who or what you want the model to act as.
LLMs mirror patterns. If you don’t set a role, the model falls back to a generic “helpful assistant” tone. A role locks in expertise, voice, and style.
Task
The specific outcome you want.
Models are literal. A vague task produces vague output. Stating the job sharply stops the model from drifting or over-explaining.
Audience
The group the work is aimed at.
The same answer is not right for everyone. Giving an audience helps the model judge tone, complexity, reading level, and examples.
Context
Everything relevant to the task—background, examples, source text, constraints, extra information.
Models don’t know your situation. If it isn’t in the prompt, it doesn’t exist. Poor or missing context is one of the biggest causes of hallucinations.
{{CONTEXT}}
Here’s what you need to know:
Constraints
Rules the answer must follow: tone, length, style guides, accessibility, format limits.
Constraints keep the output tight and predictable. Without them, the model will pad, ramble, or pick a style you didn’t want.
{{CONSTRAINTS}}
Follow these rules:
Structure
How the answer should be formatted: headings, bullets, steps, tables.
Models are sensitive to structure. If you say “use bullets”, you get clean bullets. If you don’t, you may get a long block of text. Structure also makes the output easier to copy into documents or design tools.
{{STRUCTURE}}
Format the output as
Advanced extras
You can add these or use them in place of other components. Be aware that too much context might overwhelm the models ability to reason.
Scope
What’s included and excluded.
Assumptions
Things the model can safely take as true.
Data
Tables, schemas, components, Figma frames.
Examples
Samples to imitate.
Process
Steps the model should follow.
Acceptance criteria
Pass/fail conditions.
Style
Tone, depth, pace.
Risks
Challenges or limits to assess.
Decision frame
How recommendations should be evaluated.
References / Evidence
Required sources or standards.
Success criteria
What “good” looks like.
Limits
What to avoid.
Output format
JSON, Markdown, YAML, tables, etc.
Inputs
Artefacts provided for analysis.
{{SCOPE}}
{{ASSUMPTIONS}}
{{DATA}}
{{PROCESS}}
{{ACCEPTANCE CRITERIA}}
{{STYLE}}
{{RISKS}}
{{DECISION FRAME}}
{{REFERENCES / EVIDENCE}}
{{SUCCESS CRITERIA}}
{{LIMITS}}
{{OUTPUT FORMAT}}
{{INPUTS}}
Full prompt
Copy, paste and complete.
{{ROLE}}
Act as a:
{{TASK}}
Your job is to:
{{AUDIENCE}}
This is for:
{{CONTEXT}}
Here’s what you need to know:
{{CONSTRAINTS}}
Follow these rules:
{{STRUCTURE}}
Format the output as: