AI Prompt Templates: 10 Copy-Paste Formats (2026)

November 11, 2026 · 12 min read

AI Prompt Templates: 10 Formats for ChatGPT, Claude and Gemini

An AI prompt template is a reusable prompt with fixed instructions and blanks you fill in each time, such as [TOPIC] or [AUDIENCE]. This page gives you one five-part structure and ten copy-paste templates built on it, written in Markdown so they work in ChatGPT, Claude and Gemini.

What Is an AI Prompt Template?

A one-off prompt solves today's problem. An AI prompt template solves the same problem every week, because only the bracketed parts change. You write the role, the rules and the output shape once, then swap in new text, names or numbers.

Compare a one-off request like "summarize this" with template 1 below. The one-off version leaves the model to guess the length, the audience and what to do with numbers. The template answers all three before you paste a word, so the summary comes back in the same shape every time.

The payoff is consistency. Two people on a team using the same template get answers in the same format, which makes them easier to compare and reuse. Developer libraries such as LangChain have their own template objects for code; this page covers templates you paste into a chat window.

A template also isn't a system prompt. You paste a template into the message for one job, while a system prompt or custom instruction sits behind every reply in the chat.

What Should a Prompt Template Include? The Five-Part Skeleton

Every template on this page uses the same five parts. Here's the bare skeleton, using Markdown headings to separate them:

## Role
You are a [ROLE] who [WHAT THEY ARE GOOD AT].

## Context
[WHO THIS IS FOR, WHAT IT IS, WHY IT MATTERS]

## Task
[ONE CLEAR INSTRUCTION, STARTING WITH A VERB]

## Constraints
- [LENGTH LIMIT]
- [WHAT TO AVOID]
- If anything in square brackets is still unfilled, ask me for it first

## Output format
[THE EXACT SHAPE OF THE ANSWER: LIST, TABLE, SECTIONS]

The model sees five labeled blocks instead of one long paragraph:

  • Role sets the expertise and voice, so the answer starts at the right level.
  • Context says who the result is for and what the input is. Most weak answers come from missing context.
  • Task is one instruction. Two tasks in one template usually means you want two templates.

The last two parts do the most work:

  • Constraints are the guardrails: length, tone, and things the model must not change or invent.
  • Output format describes the finished answer. Many templates online skip it, yet it's the part that makes results consistent.

For short prompts, bold labels such as **Task:** do the same job in less vertical space, which is why the ten templates below use them. Our guide to Markdown headings covers the heading syntax.

Placeholders: Square Brackets, Curly Braces and When Each Fits

Template sites mix three placeholder styles: square brackets, single braces and double braces. Here's the rule we follow: for chat use, write placeholders in square brackets and capitals, like [AUDIENCE] or [PASTE TEXT HERE]. They stand out to you when filling them in, and nobody mistakes them for code.

Curly braces belong to software that substitutes values for you. Google Cloud's template docs mark variables with single braces, as in {animal_name}. Anthropic's own prompt examples use double braces such as {{ANNUAL_REPORT}}, the same style Mustache-type template engines use. If a tool fills the blanks, use its syntax. If you fill them by hand, square brackets are clearer.

Two habits make placeholders safer. Keep one convention per template library, and add a line telling the model to ask about any bracket you forgot to fill. Without it, the model often guesses.

Here's the skeleton filled in for a summary task. Edit the brackets and watch the preview update:

Role

You are an editor who writes for busy readers.

Context

The text below is a [DOCUMENT TYPE] for [AUDIENCE].

Task

Summarize the text.

Constraints

  • Use at most [NUMBER] bullets
  • Keep names, numbers and dates exactly as written
  • If a placeholder is still in brackets, ask me for it first

Output format

  1. One-sentence summary
  2. Bullet list of key points
  3. Anything important that is missing

Text

[PASTE TEXT HERE]

80 words437 characters16 lines
Markdown

Prompt Templates for Writing: Summarize, Rewrite, Email

These three cover most everyday writing requests. Each uses bold labels instead of headings, so it stays compact when pasted.

1. Summarize a document

**Role:** You are an editor who writes for busy readers.
**Context:** The text below is a [DOCUMENT TYPE] for [AUDIENCE].
**Task:** Summarize it.
**Constraints:** At most [NUMBER] bullets. Keep names, numbers and dates exactly as written.
**Output format:** A one-sentence summary, then the bullets, then a line starting "Missing:" that lists anything a reader would expect but the text leaves out.

[PASTE TEXT HERE]

The "Missing:" line is our favorite small addition. A summary only shows what the text says, so asking for gaps catches the missing deadline or budget that a skim would miss.

2. Rewrite for a new tone

**Role:** You are a copy editor.
**Context:** The draft below is for [AUDIENCE], who read at a [READING LEVEL] level.
**Task:** Rewrite it in a [TONE] tone.
**Constraints:** Keep every fact, figure and name unchanged. Don't add claims. Stay within 10% of the original length.
**Output format:** The rewritten text, then a short list of the three biggest changes you made.

[PASTE DRAFT HERE]

3. Draft an email

**Role:** You write clear, polite work emails.
**Context:** I'm writing to [RECIPIENT], my [RELATIONSHIP, e.g. client or manager]. Background: [ONE OR TWO SENTENCES].
**Task:** Draft an email that gets them to [GOAL].
**Constraints:** Under [NUMBER] words. One request only. No filler opening line.
**Output format:** A subject line, then the body, then a one-line follow-up I can send if there's no reply in a week.

Prompt Templates for Work and Code: Code Review, Research Brief, Action Items

The code review template contains a code block inside the prompt. Because the outer fence and the inner fence both use backticks, the outer one needs four. Our code block guide explains fences in detail.

4. Review code

**Role:** You are a senior [LANGUAGE] developer doing a code review.
**Context:** This function [WHAT IT DOES]. It runs [WHERE, e.g. in a web API].
**Task:** Review it for bugs, security issues and readability.
**Constraints:** Point to line numbers. Don't rewrite the whole function. Say "none found" for an empty category.
**Output format:** Three headed lists (Bugs, Security, Readability), each item with a line number and a suggested fix.

```[LANGUAGE]
[PASTE CODE HERE]
```

5. Build a research brief

**Role:** You are a research assistant.
**Context:** I need background on [TOPIC] for [PURPOSE, e.g. a client pitch].
**Task:** Write a one-page brief.
**Constraints:** Name the source for each key claim. Mark anything you're unsure of with (unverified). Don't invent statistics.
**Output format:** Four sections: Overview, Key facts, Open questions, Sources to check.

6. Turn meeting notes into action items

**Role:** You are a project coordinator.
**Context:** Below are raw notes from a [MEETING TYPE] on [DATE].
**Task:** Extract every action item.
**Constraints:** Only include actions someone agreed to. Write "unassigned" when no owner was named.
**Output format:** A Markdown table with the columns Owner, Action and Due date, followed by a list of decisions made.

[PASTE NOTES HERE]

The table output is plain Markdown, so you can paste the reply into the editor and the preview renders it as a real table.

Career and Learning Templates: Resume Bullets, Concepts, Quizzes, Decisions

These four round out the set of ten. If you need whole document files rather than prompts, such as a README or a resume file, our Markdown templates post has those.

7. Sharpen resume bullets

**Role:** You are a recruiter who has read thousands of resumes.
**Context:** I'm applying for [JOB TITLE]. Here are my current bullets for my role as [PAST ROLE].
**Task:** Rewrite each bullet using the pattern action verb, what I did, result.
**Constraints:** Never invent numbers. Where a number would help, write [ADD METRIC] for me to fill in.
**Output format:** A numbered list, one rewritten bullet per original.

[PASTE BULLETS HERE]

8. Explain a concept

**Role:** You are a patient teacher.
**Context:** I'm a [LEVEL, e.g. beginner] and I know [WHAT I ALREADY KNOW].
**Task:** Explain [CONCEPT].
**Constraints:** Use one everyday analogy. Define every technical term the first time you use it.
**Output format:** A short explanation, the analogy, then one question I can answer to check I understood.

9. Make a study quiz

**Role:** You are a tutor writing practice questions.
**Context:** I'm studying [SUBJECT] from the material below.
**Task:** Write a quiz.
**Constraints:** [NUMBER] questions, mixed multiple choice and short answer. Only use facts from the material.
**Output format:** Numbered questions first, then an Answers section at the end so I can test myself.

[PASTE MATERIAL HERE]

10. Compare options

**Role:** You are a neutral advisor.
**Context:** I'm choosing between [OPTION A], [OPTION B] and [OPTION C] for [GOAL].
**Task:** Compare them.
**Constraints:** Use only the criteria I list: [CRITERIA]. Say when you lack information instead of guessing.
**Output format:** A Markdown table with one row per option and one column per criterion, then a two-sentence recommendation.

Treat all ten as starting points, not finished prompts. After three or four runs, look at what you keep fixing by hand and turn it into a constraint. If you always cut the last paragraph, add a word limit. If the tone keeps drifting, name the tone in the Role line. That habit is most of what a prompt engineering template is: a prompt that improves each time you edit it, instead of one you retype from memory.

Two more habits help once a template is stable. If the output shape is hard to describe in words, add an Examples section with one short sample of a good answer. And keep confidential material out of the brackets unless your workplace allows it, because whatever you paste goes to the assistant's provider like any other message.

Do Prompt Templates Work the Same in ChatGPT, Claude and Gemini?

Mostly, yes. The five-part structure transfers because all three vendors recommend the same basics: clear instructions, context, examples and explicit rules for the output. The differences are in how each one suggests separating the parts.

A ChatGPT prompt template can stay exactly as written. OpenAI's prompt engineering guide says Markdown headers and lists help mark distinct sections, and that XML tags help show where content starts and ends.

A Claude prompt template is where XML tags pay off. Anthropic's prompting best practices recommend wrapping instructions, context and inputs in their own tags. Here's template 3 converted:

<role>You write clear, polite work emails.</role>
<context>I'm writing to [RECIPIENT], my [RELATIONSHIP]. Background: [ONE OR TWO SENTENCES].</context>
<task>Draft an email that gets them to [GOAL].</task>
<constraints>Under [NUMBER] words. One request only.</constraints>
<output_format>Subject line, body, then a one-line follow-up.</output_format>

Gemini is the most relaxed of the three: a Gemini prompt template works in either style. Google's prompting strategies page says XML-style tags or Markdown headings both work for separating parts such as context and task, and advises picking one format per prompt. It also recommends including a few examples in your prompts.

Assistants also tend to answer in Markdown. Our post on Markdown in AI explains why and how to control that output. If a pasted reply carries stray symbols or hidden characters, our AI text cleaner guide shows how to strip them.

Where to Keep Your Prompt Template Library

Each assistant has somewhere to store standing instructions. ChatGPT has custom instructions and Projects. Claude's Projects hold project instructions plus a knowledge base of uploaded files. Gemini has Gems, which you create and save to customize Gemini's responses.

Those features suit settings that apply to every chat. For task templates, we prefer one .md file per template in a single folder. It's portable across all three assistants, it doesn't depend on one vendor's account, and plain text is easy to diff when you change a template. If you later move from ChatGPT to Claude, that switch starts from a Markdown context file too.

You can write and preview those files in the editor: Markdown on the left, live preview on the right. Use Copy Markdown to grab the filled-in prompt, download the .md file, or sign in to save it to My Files. If your team keeps a prompt library on a wiki page, the Markdown to HTML converter turns a template into HTML.

One limitation to be clear about: our editor edits, previews and exports Markdown, and it has no AI features. It doesn't fill placeholders or run prompts, so you paste the finished text into the assistant yourself.

Frequently Asked Questions

A good AI prompt template is five labeled parts, clear [PLACEHOLDERS] and an output format line that says exactly what you want back. Start with the template you'd use most this week, adjust the constraints after a few runs, and keep each version as a .md file. Write and preview your library in the editor, then paste the filled-in prompt into whichever assistant you use.