Use AI Prewarming
Before giving the main task, ask: "What do you know about [topic]?" Then follow with: "Using this knowledge, [goal]." This activates relevant context before the real prompt runs.
RK.OS / AI Prompting System
Five principles.
One prompting system.
Great AI outputs do not happen by accident—they are engineered. This interactive playbook turns vague requests into structured prompts by controlling direction, format, examples, quality, and scope.
05Core principles
03Real comparisons
01Interactive checker
∞Possible outputs
01 direction role + context + rules
02 format structure + length + constraints
03 examples show what “good” looks like
04 quality verify before final output
05 scope one task / one function
// vague input → controlled system → usable output
// vague input → controlled system → usable output
PIPELINE STATUS● READY
Why it matters
Your prompt is an interface.
Generative AI models do not read minds—they respond to the information and constraints you provide. Unstructured prompting leaves the model to fill in the gaps.
“Blind Prompting” is writing and deploying prompts without evaluating the output. A repeatable system reduces guesswork, costly iteration, and unpredictable results.
Blind promptingassumptions ↑
Structured promptingcontrol ↑
Same model. Same goal. Different instructions.
Better prompting reduces what the system has to guess.
Better prompting reduces what the system has to guess.
Framework / 05
Build the prompt like a system.
Select each stage. The five principles are not five independent tricks—they control different failure points in the same prompt.
RK.PROMPT / 01
Prompt transformation
Weak input → controlled output.
Choose a scenario and compare what the model has to guess against what the stronger prompt explicitly controls.
✕ WEAK / BLIND PROMPTINPUT
→
✓ STRUCTURED PROMPTCONTROLLED
Prompt dissection
Anatomy of a complete prompt.
The editor below shows all five principles operating inside one accounting prompt. Hover the colored phrases to see the structure rather than reading it as one giant block of text.
Act as an experienced accounting professor. This explanation is intended for first-year business students with no prior accounting knowledge.
Rules: use plain, jargon-free language. If a technical term is necessary, define it immediately after it appears. Do not assume the student knows any financial terminology.
Return your answer as 4 short paragraphs, under 250 words total. Include one real-world analogy. End with a single bolded key takeaway the student should remember.
For example, a strong analogy sounds like: "Think of accrual accounting like ordering food at a restaurant — you record the charge the moment you order, not when you actually pay the bill."
Before finalizing your response, verify: (1) every paragraph is free of undefined jargon, (2) the analogy is relatable to a college student, and (3) the key takeaway is one clear, memorable sentence.
Task (scoped): Explain the difference between cash-basis and accrual-basis accounting.
Interactive diagnostic
Run a prompt check.
Before sending a prompt, toggle each control your prompt satisfies. The diagnostic scores its structure using the five-principle framework.
✓
Give Direction
Did you assign a role, add audience/context, and define clear rules or boundaries?
✓
Specify Format
Did you define output structure, length, reading level, and important constraints?
✓
Provide Example
Did you show at least one concrete example of what a useful answer should resemble?
✓
Evaluate Quality
Did you include a quality check and keep human verification in the process?
✓
Divide Labor
Is this prompt focused on one task/function instead of several unrelated jobs?
Field notes
Beyond the five principles.
Additional habits that make prompting more reliable across models and real-world use cases.
Iterate and Refine
Your first prompt rarely yields the perfect output. Treat prompting as a conversation — ask the AI to improve, adjust tone, expand a point, or redo the format.
Test Across Models
Context dictates the model — LLM performance varies by task. Test your prompt on multiple models (GPT, Claude, Gemini) and select the best fit for your specific use case.
Always Verify Outputs
AI can be confidently wrong (hallucinations). Keep humans in the loop. Always verify facts, figures, and technical details from authoritative sources before using output professionally.
Be Specific, Not Verbose
Precision beats length. A focused 4-sentence prompt outperforms a rambling 10-sentence one. Every word in your prompt should earn its place.
Chain of Thought
For reasoning or analytical tasks, prompt the AI to "think step by step." This technique significantly improves logical accuracy and the quality of multi-step outputs.
Quick reference
The prompting cheat sheet.
A compact system reference for the five principles. Keep it open while you build or review a prompt.
01Give DirectionRole, audience context, clear rules (Do's & Don'ts), AI prewarming
"Act as... / Intended for... / Do not..."02Specify FormatOutput type, length, reading level, structural constraints
"Return 5 bullets ≤12 words. No jargon."03Provide Example1–3 sample inputs/outputs (few-shot). Give direction first.
"For example, a good answer looks like: '...'"04Evaluate QualityBuilt-in quality check, model selection, human oversight
"Before responding, verify each claim is specific and evidence-based."05Divide LaborOne task per prompt. Use separate prompts for separate functions.
"Only [task A]. [Task B] is handled in a separate prompt."