Prompt Templates and Variable Substitution

You're Writing the Same Prompt 20 Different Ways. Every Week.

If you have been using AI for a few months, you have probably converged on prompts that work well for your most common tasks. The problem is that most people keep these prompts in their heads, reconstruct them from scratch each time, or dig through conversation history to find a version that worked. Every reconstrution introduces variation — and variation means inconsistency in output quality.

A prompt template is a compound prompt with variable slots where the task-specific details go. The structure, role, constraints, and output spec are fixed and reusable. Only the content that changes between uses — the topic, the document, the audience, the specific deliverable — gets filled in each time. Well-designed templates produce consistent output quality across uses because the prompting technique is already embedded in the structure.

Anatomy of a Good Template

A useful prompt template has three components:

• Fixed scaffolding: The role, purpose, constraints, and output spec that apply every time this type of prompt is used. Written once and reused without modification.
• Variable slots: Clearly marked placeholders for the content that changes between uses. Convention: use [BRACKETS] to mark variables. They should be named descriptively enough that filling them in is unambiguous.
• A one-line description: What this template is for and when to use it. This is for your own future reference and is not sent to the model.

Building a Template: Before and After

One-off prompt (not reusable): "You are a B2B copywriter. Write a LinkedIn post announcing our project management software launch. Goal: generate demo requests from startup founders. Under 250 words, no exclamation points, no bullet points, do not use game-changing or revolutionize. End with a question."

Template version:
"You are a [TONE_STYLE] copywriter specializing in [INDUSTRY] marketing.
Write a LinkedIn post announcing [PRODUCT_OR_LAUNCH_DESCRIPTION].
Goal: [DESIRED_ACTION] from [TARGET_AUDIENCE].
Under [WORD_LIMIT] words. No exclamation points. No bullet points — flowing prose only. Do not use [BANNED_PHRASES].
End with [CALL_TO_ACTION_TYPE]. Write one version."

The template version requires 30 seconds to fill in and produces the same caliber of output as a carefully written one-off — for any product, any industry, any launch, any audience.

Variable Naming That Works

The most common template failure is variables that are too vague to fill in consistently. [CONTENT] and [DETAILS] are bad variable names — they do not constrain what goes in. Good variable names are descriptive and unambiguous:

• [TARGET_AUDIENCE] is better than [AUDIENCE] — it signals who, not what
• [TONE_STYLE] is better than [TONE] — it signals the type of description expected
• [WORD_LIMIT] is better than [LENGTH] — it signals a number, not a vague scale
• [DESIRED_ACTION] is better than [GOAL] — it signals that you want a specific behavior, not an abstract outcome

Template Categories Worth Building

Most people benefit from templates in five recurring task categories:

• Email drafts: Follow-ups, requests, proposals, sensitive conversations
• Content creation: LinkedIn posts, blog introductions, summaries
• Analysis: Market analysis, competitive comparison, option evaluation
• Document work: Summarize document, extract requirements, identify risks
• Research: Breakdown of unfamiliar topic, steelman argument, synthesis of multiple sources

You do not need templates for everything — only for tasks you do more than once a month. One-off tasks are better served by a fresh, specific prompt than by forcing them into a template that does not quite fit.

Building Your Template Library

Start with your three most common prompting tasks. Write a one-off prompt for each, refine it until the output is consistently good, then extract the variable slots. Store the templates in a notes app, a document, or wherever you keep reference material you access regularly. The storage method does not matter — what matters is that you can find them quickly and do not have to remember them.

For each template, write the one-line description at the top. "Use this for: LinkedIn posts announcing product or feature launches. Fill in: TONE_STYLE, INDUSTRY, PRODUCT_OR_LAUNCH_DESCRIPTION, DESIRED_ACTION, TARGET_AUDIENCE, WORD_LIMIT, BANNED_PHRASES, CALL_TO_ACTION_TYPE." Six months from now, you will thank yourself for writing that line.

Template Maintenance

Templates need occasional updates as models improve and your preferences evolve. A template that produced great output six months ago might need a constraint adjustment because the model's defaults have shifted. Review your most-used templates quarterly, run a test prompt, and update the constraints if the output has drifted from what you want. Good templates are living documents, not static files.