Intrenion

Intrenion Doctrine

Prompt Engineering with ChatGPT

Christian Ullrich
2026-07-06

Table of Contents

Audio Discussion

Section 1: General principles

Episode 1: Prompt purpose

Practice 1: Define the work situation

Problem
You start writing a prompt before you clearly understand what work ChatGPT should support.

Action
Describe the situation that led to the prompt and the work ChatGPT should help you complete.

Outcome
The prompt addresses the actual need rather than a vague request.

Practice 2: Name the intended result

Problem
The prompt describes an activity but not the result you expect.

Action
State the concrete result that ChatGPT should produce or help you achieve.

Outcome
The response is focused on the intended outcome.

Practice 3: Separate task from purpose

Problem
The prompt explains what to do but not why it matters.

Action
Distinguish the requested task from the purpose it should serve.

Outcome
ChatGPT can better prioritize its work toward the desired result.

Practice 4: Limit the prompt scope

Problem
The prompt tries to solve too many problems at once.

Action
Reduce the prompt to a single coherent objective.

Outcome
The response becomes more focused and reliable.

Practice 5: Check whether a prompt is needed

Problem
You invest time in writing a complex prompt for a simple request.

Action
Decide whether the task requires a structured prompt or a straightforward question.

Outcome
You spend effort only where prompt engineering adds value.

Episode 2: Prompt structure

Practice 6: Choose the needed components

Problem
The prompt contains unnecessary or missing components.

Action
Select only the prompt components that support the current task.

Outcome
The prompt stays complete without becoming unnecessarily complex.

Practice 7: Order the components logically

Problem
The prompt presents information in a confusing sequence.

Action
Arrange the components in the order ChatGPT should interpret them.

Outcome
The prompt is easier to understand and follow.

Practice 8: Separate stable rules from variable input

Problem
Reusable instructions are mixed with task-specific information.

Action
Keep permanent instructions separate from information that changes between uses.

Outcome
The prompt becomes easier to reuse and maintain.

Practice 9: Remove conflicting instructions

Problem
Different parts of the prompt compete with each other.

Action
Review the prompt for contradictory goals, rules, or expectations and remove conflicts.

Outcome
ChatGPT receives one consistent set of instructions.

Practice 10: Keep the prompt maintainable

Problem
The prompt becomes difficult to understand or update over time.

Action
Remove unnecessary complexity while preserving the intended behavior.

Outcome
The prompt remains easy to improve and reuse.

Episode 3: Prompt refinement

Practice 11: Identify the output problem

Problem
The result does not meet your expectations, but the reason is unclear.

Action
Describe the specific weakness in the output before changing the prompt.

Outcome
Improvements focus on the actual problem.

Practice 12: Find the weak instruction

Problem
A prompt produces inconsistent results because one instruction is ineffective.

Action
Identify the instruction that most likely caused the undesired behavior.

Outcome
Changes target the real source of the problem.

Practice 13: Change one component at a time

Problem
Several prompt changes make it impossible to identify what improved the result.

Action
Revise one component of the prompt before testing the prompt again.

Outcome
You understand which changes improve the prompt.

Practice 14: Test the revised prompt

Problem
A revised prompt is accepted without verifying its effect.

Action
Run the updated prompt and compare the new output with the previous result.

Outcome
Prompt improvements are confirmed through use.

Practice 15: Keep the useful version

Problem
Successful prompt improvements are lost during later revisions.

Action
Save prompt versions that consistently produce useful results.

Outcome
Effective prompts remain available for future work.

Episode 4: Prompt reuse

Practice 16: Identify repeated work

Problem
You repeatedly write similar prompts for similar tasks.

Action
Look for prompt patterns that occur across multiple work situations.

Outcome
Repeated work becomes easier to standardize.

Practice 17: Replace fixed details with variables

Problem
The prompt can only be used for one specific situation.

Action
Replace changing information with clearly identified placeholders.

Outcome
The prompt can be adapted with minimal effort.

Practice 18: Preserve the stable structure

Problem
Reusable prompts become inconsistent after repeated modifications.

Action
Keep the proven prompt structure unchanged while updating only the variable parts.

Outcome
The prompt remains reliable across different uses.

Practice 19: Document the intended use

Problem
Later users cannot tell when to apply a prompt.

Action
Describe the situations the prompt was designed to support.

Outcome
The prompt is used in the right context.

Practice 20: Update the prompt when ChatGPT behavior changes

Problem
A prompt becomes less effective after ChatGPT changes.

Action
Review and adjust the prompt when model behavior noticeably changes.

Outcome
The prompt continues to produce reliable results.

Section 2: Core prompt components

Episode 5: Goal

Practice 21: State the desired result

Problem
ChatGPT does not know what final result you expect.

Action
Describe the result that ChatGPT should produce or help you achieve.

Outcome
The response is focused on a clear objective.

Practice 22: Explain why the result matters

Problem
ChatGPT understands the task but not its purpose.

Action
Explain why the requested result is important.

Outcome
ChatGPT can better prioritize its work.

Practice 23: Define success for the result

Problem
The expected quality of the result remains unclear.

Action
Describe what a successful result should achieve.

Outcome
The response is aligned with your expectations.

Practice 24: Prioritize competing goals

Problem
The prompt contains several goals without indicating their importance.

Action
State which goals take priority if trade-offs become necessary.

Outcome
ChatGPT makes decisions that reflect your priorities.

Practice 25: Remove vague intentions

Problem
The goal contains broad or ambiguous wording.

Action
Replace vague intentions with concrete expectations.

Outcome
The goal becomes easier for ChatGPT to interpret.

Episode 6: Instructions

Practice 26: Set language rules

Problem
The response uses the wrong language or writing conventions.

Action
Specify the language and language-specific requirements.

Outcome
The response follows the expected language rules.

Practice 27: Set source rules

Problem
ChatGPT uses information that should not be considered.

Action
Define which sources ChatGPT may or may not use.

Outcome
The response is based on the intended information.

Practice 28: Set reasoning rules

Problem
ChatGPT approaches the task in an unsuitable way.

Action
Describe how ChatGPT should think through the task.

Outcome
The reasoning follows your intended approach.

Practice 29: Set style rules

Problem
The writing style does not fit the intended use.

Action
Describe the required writing style and tone.

Outcome
The response matches the expected style.

Practice 30: Set behavior rules

Problem
ChatGPT behaves differently from what you expect.

Action
State general rules that should guide ChatGPT throughout the conversation.

Outcome
The responses become more consistent.

Episode 7: Role

Practice 31: Assign the working perspective

Problem
ChatGPT approaches the task from the wrong perspective.

Action
Assign the role that best fits the work to be performed.

Outcome
The response reflects the intended perspective.

Practice 32: Define the role behavior

Problem
The role is too broad to guide ChatGPT effectively.

Action
Describe how the role should approach the task.

Outcome
The role leads to more consistent behavior.

Practice 33: Limit the role scope

Problem
The assigned role encourages unnecessary assumptions.

Action
Define the responsibilities and limits of the role.

Outcome
The role remains focused on the intended work.

Practice 34: Match the role to the task

Problem
The assigned role does not support the requested task.

Action
Choose a role that naturally fits the objective.

Outcome
The role improves the quality of the response.

Practice 35: Avoid decorative role labels

Problem
The role sounds impressive but provides little guidance.

Action
Describe the role through its working behavior rather than its title.

Outcome
The role provides practical direction.

Episode 8: Context

Practice 36: Provide the needed background

Problem
ChatGPT lacks the information necessary to complete the task.

Action
Provide the background needed to understand the work.

Outcome
The response is based on the correct situation.

Practice 37: Separate relevant from irrelevant context

Problem
The prompt contains background information that distracts from the task.

Action
Include only context that supports the requested work.

Outcome
ChatGPT focuses on the relevant information.

Practice 38: State the working environment

Problem
Important organizational or technical conditions remain unclear.

Action
Describe the environment in which the result will be used.

Outcome
The response better fits the intended environment.

Practice 39: Include relevant conditions

Problem
Important circumstances are missing from the prompt.

Action
Add conditions that influence how the task should be completed.

Outcome
The response reflects the actual situation.

Practice 40: Keep context usable

Problem
The context is too large or poorly organized.

Action
Present the background in a clear and structured way.

Outcome
ChatGPT can use the context more effectively.

Episode 9: Task

Practice 41: Name the concrete work

Problem
The requested work remains too general.

Action
State exactly what ChatGPT should do.

Outcome
The task is immediately understandable.

Practice 42: Define the object of work

Problem
ChatGPT does not know what it should work on.

Action
Identify the document, information, or subject of the task.

Outcome
The task has a clear focus.

Practice 43: Specify the action to perform

Problem
The required activity is open to interpretation.

Action
Use clear action verbs to describe the work.

Outcome
ChatGPT performs the intended activity.

Practice 44: Limit the task scope

Problem
The task covers more work than intended.

Action
Define the boundaries of the requested work.

Outcome
The response remains focused.

Practice 45: Avoid mixing several tasks

Problem
Multiple unrelated requests compete within the same prompt.

Action
Separate independent tasks into individual requests whenever possible.

Outcome
Each response addresses one clear task.

Episode 10: Workflow

Practice 46: Break the work into steps

Problem
The task is too complex to complete reliably in one step.

Action
Divide the work into a logical sequence of activities.

Outcome
ChatGPT can complete the work more systematically.

Practice 47: Put steps in sequence

Problem
The order of work is unclear.

Action
Specify the sequence in which the steps should be completed.

Outcome
The work follows a predictable process.

Practice 48: Separate thinking stages

Problem
Analysis, evaluation, and writing become mixed together.

Action
Assign different thinking activities to separate workflow steps.

Outcome
Each stage receives appropriate attention.

Practice 49: Define stopping points

Problem
ChatGPT continues beyond the intended stage of work.

Action
Specify where the workflow should pause or end.

Outcome
You remain in control of the process.

Practice 50: Prevent skipped steps

Problem
Important parts of the workflow are ignored.

Action
State that each workflow step should be completed before moving to the next.

Outcome
The process becomes more reliable.

Episode 11: Output

Practice 51: Specify the output type

Problem
The response uses the wrong format.

Action
State the type of output you expect.

Outcome
The response matches the intended format.

Practice 52: Define the output structure

Problem
The information is difficult to navigate.

Action
Describe how the output should be organized.

Outcome
The response becomes easier to use.

Practice 53: Set the level of detail

Problem
The response is too brief or unnecessarily detailed.

Action
Describe the expected level of detail.

Outcome
The response fits the intended purpose.

Practice 54: Set formatting requirements

Problem
The presentation does not meet your needs.

Action
Specify formatting requirements that support the intended use.

Outcome
The output is easier to reuse.

Practice 55: Make the output reusable

Problem
The response requires unnecessary rework before it can be used.

Action
Specify an output that can be used directly in your workflow.

Outcome
The result is immediately usable.

Section 3: Optional prompt components

Episode 12: Source material

Practice 56: Identify the source of truth

Problem
ChatGPT does not know which information it should rely on.

Action
Specify the documents, files, or other sources that should serve as the basis for the task.

Outcome
The response is grounded in the intended information.

Practice 57: Limit the usable material

Problem
ChatGPT considers information that is outside the intended scope.

Action
Define which sources may and may not be used.

Outcome
The response remains focused on the relevant material.

Practice 58: Separate input from instruction

Problem
ChatGPT cannot distinguish between instructions and source material.

Action
Clearly separate the information to analyze from the instructions that describe the task.

Outcome
ChatGPT interprets the prompt more reliably.

Practice 59: Preserve source meaning

Problem
Important details from the source material become distorted or omitted.

Action
Instruct ChatGPT to preserve the original meaning unless changes are explicitly requested.

Outcome
The response remains faithful to the source material.

Practice 60: Flag missing source information

Problem
The available source material is insufficient to complete the task.

Action
Ask ChatGPT to identify missing information instead of making unsupported assumptions.

Outcome
Information gaps become visible before they affect the result.

Episode 13: Assumptions

Practice 61: State known assumptions

Problem
Important assumptions remain implicit.

Action
List the assumptions that ChatGPT should use during the task.

Outcome
The reasoning is based on transparent assumptions.

Practice 62: Ask ChatGPT to expose assumptions

Problem
ChatGPT introduces assumptions without making them visible.

Action
Instruct ChatGPT to explicitly identify any assumptions it makes.

Outcome
Implicit assumptions become easier to review.

Practice 63: Separate facts from assumptions

Problem
Facts and assumptions become mixed together.

Action
Require ChatGPT to distinguish verified information from assumptions.

Outcome
The response becomes easier to evaluate.

Practice 64: Challenge weak assumptions

Problem
Uncertain assumptions remain untested.

Action
Ask ChatGPT to question assumptions that significantly influence the result.

Outcome
Weak assumptions are identified before they affect decisions.

Practice 65: Update assumptions after review

Problem
Assumptions remain unchanged after new information becomes available.

Action
Revise the assumptions whenever important information changes.

Outcome
The prompt remains aligned with the current situation.

Episode 14: Constraints

Practice 66: Name hard limits

Problem
ChatGPT does not know which limits must never be exceeded.

Action
Specify the non-negotiable boundaries for the task.

Outcome
The response respects the required limits.

Practice 67: Name forbidden actions

Problem
ChatGPT performs actions that should be avoided.

Action
State which actions, topics, or behaviors are not allowed.

Outcome
The response avoids unwanted behavior.

Practice 68: Set scope boundaries

Problem
The response expands beyond the intended scope.

Action
Define what should be included and what should be left out.

Outcome
The response remains focused.

Practice 69: Define trade-offs

Problem
Several objectives compete without guidance.

Action
Explain which priorities should take precedence when compromises are necessary.

Outcome
ChatGPT makes decisions that match your priorities.

Practice 70: Remove impossible requirements

Problem
The prompt contains expectations that cannot all be fulfilled.

Action
Review the prompt and eliminate conflicting or unrealistic requirements.

Outcome
The prompt becomes achievable.

Episode 15: Quality criteria

Practice 71: Define quality standards

Problem
ChatGPT does not know what distinguishes a good result.

Action
Describe the qualities the result should demonstrate.

Outcome
The response aligns with your quality expectations.

Practice 72: Prioritize the important criteria

Problem
Too many quality expectations compete with each other.

Action
Identify the criteria that matter most.

Outcome
ChatGPT focuses on the highest priorities.

Practice 73: Make vague quality words concrete

Problem
Terms such as “good” or “clear” are open to interpretation.

Action
Replace abstract quality descriptions with concrete expectations.

Outcome
The quality requirements become easier to apply.

Practice 74: Add checks for common failures

Problem
Recurring weaknesses appear in the response.

Action
Include checks that address the most likely quality problems.

Outcome
Common mistakes are reduced.

Practice 75: Use criteria to revise the result

Problem
The result is accepted without evaluating its quality.

Action
Ask ChatGPT to review the result against the defined quality criteria.

Outcome
The final response better satisfies the intended standards.

Episode 16: Interaction

Practice 76: Decide when ChatGPT should ask

Problem
ChatGPT asks unnecessary questions or fails to ask important ones.

Action
Specify when ChatGPT should request clarification before continuing.

Outcome
The conversation becomes more efficient.

Practice 77: Define when ChatGPT should stop

Problem
ChatGPT continues beyond the intended scope of the interaction.

Action
Specify when ChatGPT should stop and wait for further instructions.

Outcome
You remain in control of the conversation.

Practice 78: Set the iteration pattern

Problem
The interaction does not follow the intended working process.

Action
Define how ChatGPT should proceed through successive iterations.

Outcome
The conversation follows a predictable pattern.

Practice 79: Control approval points

Problem
ChatGPT proceeds without confirming important intermediate results.

Action
Define where ChatGPT should pause for your review or approval.

Outcome
Important decisions remain under your control.

Practice 80: Keep the conversation on track

Problem
The conversation gradually drifts away from the original objective.

Action
Instruct ChatGPT to stay focused on the defined task unless directed otherwise.

Outcome
The conversation remains aligned with the intended goal.