The Four-Part Prompt Specification
Control AI output with instructions, context, examples, and constraints.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 99%
The Four-Part Prompt Specification converts a casual request into an actionable brief. Instructions identify the operation the model should perform, while context supplies the facts, audience, purpose, or source material needed to interpret that operation. Examples reveal the desired pattern, style, or structure concretely. Constraints define what the model must include, exclude, limit, or preserve. Together, these elements reduce the model's need to guess and make weak results easier to diagnose: an irrelevant answer may lack context, an incorrectly shaped answer may lack an example, and an overbroad answer may lack constraints. The framework is reusable because the same four questions can be applied across writing, coding, research, planning, and data-transformation tasks without depending on a particular AI product.
Origin
Extracted from Marketing Against The Grain during Kevin Hudson's explanation of the prompting level of AI fluency.
Core principles
- 01Clear task instructions define what the model must do.
- 02Relevant context helps the model interpret the task correctly.
- 03Examples demonstrate the expected pattern better than abstract adjectives.
- 04Constraints establish boundaries, exclusions, and output requirements.
- 05Prompt components should serve the task rather than add unnecessary length.
How to run it
- 1
Define the Instruction
State the operation and intended deliverable in direct language. Use an active verb such as analyze, compare, draft, classify, or transform.
Pro tip Separate the core task from supporting background.
Watch out Avoid bundling contradictory actions into one vague request.
- 2
Supply Context
Provide the audience, objective, source material, business situation, and relevant definitions. Include only information that can affect the answer.
Pro tip Explain why the output will be used when that changes the appropriate response.
Watch out Irrelevant context can distract the model from the actual task.
- 3
Demonstrate with Examples
Show one or more representative inputs or outputs when voice, formatting, or judgment criteria matter. Point out which features should be imitated.
Pro tip Use an approved real example whenever possible.
Watch out A flawed example can override otherwise good written instructions.
- 4
Set Constraints
Specify boundaries such as length, format, required fields, prohibited language, available evidence, and acceptable assumptions. Make critical constraints testable.
Pro tip Phrase exclusions explicitly when the model repeatedly makes the same mistake.
Watch out Too many rigid constraints may make the task impossible or internally inconsistent.
- 5
Evaluate and Isolate
Compare the result with each of the four components. Revise the component responsible for the failure and run the task again.
Pro tip Preserve successful prompt components while testing a correction.
Watch out Changing every component simultaneously makes it difficult to learn what improved the result.
In the wild
A manager instructs the model to draft a weekly update, provides project status and the executive audience, supplies one previously approved update, and constrains the result to five bullets with no unsupported claims. The model now has both the substance and the presentation standard.
→ The first draft requires fewer factual and stylistic corrections.
A creator asks an agent to turn a video into a lead magnet, provides the video and branding assets, describes the expected sections, and requires use cases, starter prompts, a cover page, and brand colors.
→ The agent can research, write, and design a much more complete first version.
Common mistakes
Giving Only the Topic
A topic tells the model what the request concerns but not what operation or deliverable is required.
Using Adjectives Instead of Examples
Terms such as professional or engaging are ambiguous; a representative example communicates the desired pattern more reliably.
Leaving Boundaries Implicit
If exclusions, evidence rules, or length limits are unstated, the model must invent its own boundaries.
Is it for you?
Best for
It is best for repeatable writing, research, analysis, and transformation tasks where output requirements can be stated clearly.
Not ideal for
It is not ideal for open-ended exploration where prematurely narrow constraints would suppress useful possibilities.
From the transcript
“So after that, you start getting into prompting, you just realize that when you ask things differently, you get a different result.”
“And so when you dive deeper into that, if you start putting in clear instructions, add context, examples, and constraints, it just changes everything.”
From the episode
The AI Skill Ladder (Beginner → Workflow Builder)