Prompt Engineering Implementation Partner
Module 2: Practical Foundations of Prompting
Video 5: Structuring Prompts with RICE
Sushmitha Dixith
Data Scientist | GITAA Pvt Ltd
Recap: Video 4
Role shifts vocabulary and framing — not factual accuracy.
Audience determines depth: prior knowledge, vocabulary, example types.
Tone controls style independently of content — combine all three.
Layer role + audience + tone for precise, repeatable output.
The RICE Framework
RICE is a structured checklist, not a formula. Each component eliminates a specific failure mode.
Role — Eliminates: ambiguous perspective "You are a strategy consultant advising a
R Who is the model acting as? What expertise, viewpoint, and 150-person B2B SaaS company entering
India."
vocabulary should it bring?
Instructions — Eliminates: vague task definition "Identify the top 3 entry barriers,
I Exactly what should the model do? Use specific verbs, recommend one entry mode, justify with
data, under 400 words."
deliverables, and constraints.
Context — Eliminates: missing background "Output is for a board briefing. Company
C What does the model need to know that it cannot infer? has no Asia presence. Budget is limited,
risk appetite is low."
Audience, constraints, purpose, situation.
Examples — Eliminates: format uncertainty "Barrier 1: [name]. [2-sentence
E Show what good output looks like. Even one example explanation]. Mitigation: [1 sentence].
Entry mode: [choice] because [reason]."
communicates structure, tone, and depth.
Building a RICE Prompt - Step by Step
Task: Market-entry strategy for India. Watch the prompt improve as each component is added.
No RICE
Who is writing? For what company? What
"Write a market-entry strategy for India." should it cover? What format? All guessed.
+ Role (R)
Better framing — but task is still vague. What
"You are a strategy consultant advising a mid-size B2B SaaS
exactly should be covered?
company. Write a market-entry strategy for India."
+ Instructions (RI)
"...Identify top 3 entry barriers, recommend one entry Task is now precise. But model doesn't know
mode, justify with data, name one critical success factor. the company or output purpose.
Under 400 words."
+ Context (RIC)
"...HR software, US+UK only, no Asia presence, limited Model knows the situation. Add one example
budget, risk-averse. Output will brief the board next to lock in format.
month."
Full RICE
"...Example: Barrier 1: [name]. [2-sentence explanation]. Complete. Every ambiguity eliminated. Output
Mitigation: [1 line]. Entry mode: [choice] because is predictable and immediately usable.
[rationale]."
When to Apply RICE?
RICE adds value for complex, repeated, or high-stakes prompts. Not every prompt needs all four components.
USE RICE WHEN… KEEP IT SIMPLE WHEN…
Task has multiple deliverables or steps Quick one-off factual question
Output will be reused or shared with others Simple reformatting or summarisation
Consistency matters across many runs Zero-shot already works reliably
Something went wrong before (diagnose which
Speed matters more than precision
component)
High stakes: client-facing, board-level, published output You're testing/exploring, not producing final output
RICE as a debugging tool
When output is wrong, ask: Is it the Role (wrong perspective)? Instructions (vague task)? Context (missing info)? Examples
(wrong format)? Fix that one component. Don't rewrite the whole prompt.
Key Takeaways
RICE = Role, Instructions, Context, Examples — A checklist that covers the most common prompt failure
1
modes
Each component eliminates a different ambiguity — Role=perspective, Instructions=task,
2
Context=background, Examples=format
3 Structured prompts are easier to debug — When it goes wrong, you know which component to fix
Build incrementally — add and test one component at a time — This reveals what each component
4
contributes
5 RICE works for any task type — Strategy, writing, analysis, code, classification, summarisation
Up Next: Getting Usable Outputs