Stop 12 · Prompt EngineeringTopics · 6/10

Prompt Templates

In short

Copy-pasting the same prompt and editing a few words by hand works for one test. In an app, you need one well-tested prompt with blanks your code fills in. That's a prompt template.

Reusable prompt structures with variable slots — how production AI features avoid rewriting prompts from scratch and keep behavior consistent across inputs.

The blob works at a small printing press, sliding blank name blocks into a card template.

Think of it like this

The wedding card press

The printing press uses one beautiful card design for a whole order. Only the guest's name block changes for each card. The design is tested once and stays the same. A prompt template is that design, and the variables are the name blocks.

Watch it happen

From template to model call

Template

Variables

Check

Render

Send

1/5One versioned template: 'Summarise this {doc_type} for a {audience} in {n} bullets.'

The big ideas

What you need to know

Version your prompts

Keep templates in files under Git, not scattered strings. You can see what changed when quality drops.

Test like functions

Run each template against a fixed set of inputs before releasing a change.

Fence user input

Insert user text inside clear delimiters so it can't blend into your instructions.

Tools help

Plain Python f-strings work. Jinja2, LangChain and similar libraries add loops, defaults and conditionals.

See it in code

Walk through the code

python

1from string import Template2 3SUMMARY_V2 = Template("""Summarise the $doc_type below for $audience.4Use exactly $n bullet points. Use simple English.5 6<document>7$text8</document>""")9 10def build_prompt(doc_type, audience, n, text):11    if not 1 <= n <= 10:12        raise ValueError("n must be 1 to 10")13    return SUMMARY_V2.substitute(doc_type=doc_type, audience=audience, n=n, text=text)14 15prompt = build_prompt("college circular", "first-year students", 3, circular_text)

1/3The template, with user text fenced inside <document> tags.

A small, versioned template with a check before filling. Python's string.Template uses $name placeholders.

Save yourself the pain

Common mistakes

Oops

Building prompts with string concatenation spread across ten files.

Do this instead

Keep each template in one place, named and versioned, so changes are easy to review.

Oops

Changing a template in production without re-testing.

Do this instead

Run your saved test inputs first. Small wording changes can shift output a lot.

Your mission

Template your favourite prompt

~20 min

Quick check

Why put user-provided text inside delimiters in a template?

Go deeper

1Build a small template system with variable interpolation for a recurring task.
2Version your prompt templates so changes are trackable over time.
3Test a template across edge-case inputs, not just the happy path.