Prompt Engineering: A Guide for Better AI Answers

Prompt Engineering: A Guide with Practiced Example

Dr. Rashid @ NAPD

Preface

I am writing this guide day by day to promote knowledge on efficient use AI tools Like Claude, Gemini, Copilot, MetaAI, ChatGPT etc.

I am updating this blog article following daily work experiences. I am trying to compare different AI tools using same prompt. Geting exciting experiences. Updating each experince. 

The objective is to enable your capacity getting accurate response avoiding erroneous and generic type response by AI tools.

Introduction:

Prompt engineering matters because the same AI model can produce a vague, generic answer or a sharp, usable one, depending entirely on how the request is framed. For an organization like NAPD, where AI tools are increasingly used to draft reports, summarize policy documents, or prepare training material, a shared vocabulary and a repeatable structure for writing prompts turns AI use from a hit-or-miss exercise into a dependable part of the workflow.

The frameworks ROCE, RCTSCF, PREP, TRACE, and CREATE, are exactly that: structured checklists an official can run through before sending a prompt, so the request carries enough role, context, and intent for the model to respond well the first time.

·       ROCE: Role, Objective, Context, Examples/Content

·       RCTSCF: Role, Context, Task, Specifics, Constraints, Format

·       PREP: Point, Reason, Example, Point

·       TRACE: Task, Role, Audience, Context, Expectation

·       CREATE: Character, Request, Examples, Adjustments, Type of output, Extras

Framework and its Domain Application:

Framework	Best Application
ROCE	Quick, general-purpose prompts like writing opening speech, Preface, Executive summery etc
RCTSCF	Detailed technical or specialist tasks that need tight scoping — risk assessments, policy analyses, anything with constraints on length, tone, or data.
PREP	Persuasive or argument-driven writing — recommendations, position papers, justifications that need to build a case and land on a conclusion.
TRACE	Audience-specific deliverables  like  training materials, presentations, or communications where tone and depth must match a defined audience.
CREATE	Iterative, output-format-sensitive tasks like drafting content that needs refinement, a defined structure, and specific formatting (lists, tables, reports).

 

Examples in Practice:

ROCE: "Act as a project appraisal specialist (Role) preparing a briefing for senior management (Objective/Context) to explain why a proposed rural road project should be prioritized (Examples/Content), and write it as a one-page executive summary (Format)."

RCTSCF: "You are a climate resilience advisor (Role) for a coastal development agency (Context). Draft a risk assessment for a proposed embankment project (Task), focusing specifically on salinity intrusion and flood recurrence (Specifics), keeping it under 500 words and avoiding technical jargon (Constraints), and present it as a two-column table of risks and mitigations (Format)."

PREP: "We should adopt participatory budgeting for the next fiscal year (Point), because it improves transparency and community buy-in (Reason), as seen in Sylhet's pilot ward where citizen-approved projects had higher completion rates (Example), so participatory budgeting should be piloted in two more districts next year (Point restated)."

TRACE: "Write a training module outline (Task) as a curriculum designer (Role) for mid-level NAPD officials new to AI tools (Audience), given that the training must fit into a single half-day session (Context), aiming for a practical, hands-on outline rather than a theoretical one (Expectation)."

CREATE: "Act as a monitoring and evaluation officer (Character). Draft three indicators for tracking a rural electrification project (Request), similar in style to the indicators used in the last annual report (Examples), keep each indicator to a single sentence (Adjustments), present them as a numbered list (Type of Output), and flag any indicator that would need new data collection (Extras)."

More Prompt Engineering Framework Example: 

Date- 30 September, 2026

I used following prompt and got nice result in Copilot and Claude AI tools. Giving comparative example-

RACE (Role, Action, Context, Execution): I am Direcrector (Research & Publication) of NAPD. Create a slide on ORCIDID for 30 miniutes session that will teach trainees to basics on ORCIDID,features, contribution to carreer and handon practice. Follow https://orcid.org and set example of https://orcid.org/0000-0001-8311-2363

Experience that learned-

CoPilot gave 13 slide in text form to make power point. Covered everything folowing above prompt. Expection meet by 90%. It is free version.

Claude gave 13 slide both ppt and pdf format. Covered everything folowing above prompt. Expection meet by 100%. Remeber it was licenced version.