Prompt Template QA Checklist: Role, Task, Context, Format, Constraints

Quick scan

• Problem: Prompts are often sent with one or two structural components missing, which makes output hard to reuse or predict.
• Approach: A five-component checklist you can scan before sending. Each component has a diagnostic question and a minimal fix.
• Use this now: Run a draft prompt through the five questions below before pasting it into any model.

This checklist does not measure output quality. It checks whether the structural components of a prompt are present and specific enough to reduce ambiguity.

Use it before sending, and use Prompt Mirror to see how a rules-based tool reads the same structure.

The five components

1. Role

Check: Does the prompt define who is producing the output?

Without it: The model chooses a default perspective. For specialized tasks, that default is often too generic.

Fix: Add "You are a [role]." at the start. Keep it specific to the task.

| Status | Example | | --- | --- | | Missing | "Summarize this article." | | Present | "You are a content editor. Summarize this article." |

2. Task

Check: Is the action verb specific and scoped?

Without it: The model infers the task from context, which can match your intent or miss it depending on the surrounding text.

Fix: Start the task instruction with a specific verb: summarize, extract, classify, rewrite, list, compare. Avoid "help me with" or "tell me about."

| Status | Example | | --- | --- | | Vague | "Help me with this email." | | Specific | "Rewrite this email so it opens with the decision, not the context." |

3. Context

Check: Does the prompt include the information the model needs but cannot infer?

Without it: The model works with whatever it can infer from the input text and its training data. That inference may be wrong for your specific situation.

Fix: Add one or two lines of situational context: audience, purpose, prior step, or constraint the model cannot see.

| Status | Example | | --- | --- | | Missing | "Write a product description." | | Present | "Write a product description for a browser-based text tool aimed at solo writers and developers. It runs locally with no account required." |

Context is not the same as constraints. Context describes the situation. Constraints limit what the output may include.

4. Format

Check: Does the prompt specify how the output should be structured?

Without it: The model chooses a format. For short tasks, that default is often acceptable. For outputs that will be pasted elsewhere, reviewed by others, or compared across prompts, the default format often creates reformatting work.

Fix: Specify the output format explicitly. Options: bullet list, numbered list, table with named columns, paragraph with a word limit, JSON, Markdown heading structure.

| Status | Example | | --- | --- | | Unspecified | "List the pros and cons." | | Specified | "List pros and cons in a two-column table. Each row is one item. Max 8 rows total." |

5. Constraints

Check: Does the prompt state what must not happen?

Without it: The model may expand scope, use hedging language, add qualifications, or include content that contradicts the intent.

Fix: State what must be preserved, avoided, or excluded. Use explicit negatives: "do not", "keep X unchanged", "exclude", "no more than".

| Status | Example | | --- | --- | | Missing | "Rewrite this for a general audience." | | Present | "Rewrite this for a general audience. Keep the word count under 100. Do not change factual claims. Avoid jargon." |

Quick scan checklist

Use this before sending any prompt you intend to reuse:

• [ ] Role is defined and specific to the task
• [ ] Task verb is explicit and scoped (not "help me with")
• [ ] Context includes what the model cannot infer from the input alone
• [ ] Output format is named (list, table, paragraph, structured fields)
• [ ] At least one constraint states what must not change or be added

A prompt that passes all five checks is structurally complete. It is not guaranteed to produce the output you want — but it removes the most common sources of structural ambiguity.

What this checklist does not assess

• Whether the context you added is accurate
• Whether the model has domain knowledge for the task
• Whether the output will be factually correct
• Whether the model version supports the requested format
• Whether the task itself is well-defined enough to be useful

If a structurally complete prompt still produces inconsistent output, the problem is more likely in the context or task definition than in the structure itself.

Related tools and notes

• Run a structural check in Prompt Mirror
• See five complete worked examples in Five Structure Gaps Prompt Mirror Can Expose
• Compare structured and manual approaches in Prompt Mirror vs Manual Editing

FAQ

Do I need all five components every time?

No. Simple, single-use prompts often only need task, format, and one constraint. All five components matter when you are building a reusable template or when output quality needs to be predictable across multiple uses.

What is the difference between context and constraints?

Context describes the situation the model needs to know. Constraints limit what the output may include or do. "This is for a legal audience" is context. "Do not give legal advice" is a constraint.

Should I run every prompt through Prompt Mirror?

Prompt Mirror is most useful for iterating on templates and for learning what structure gaps feel like. For quick, throwaway prompts it adds friction without much return.

How do I know if my constraints are too strict?

If the model outputs something short and generic, or repeatedly fails to include expected content, the constraints may be limiting too much. Remove one at a time and check what changes.