AI prompt generator: let a model write your first prompt

Before you start

What comes back from a generator is usually a prompt template. It is a prompt with named gaps, called variables, that you or your code fill in later with a customer's email or a company name. The general name for the instructions behind a generator is a meta-prompt.

This lesson covers two routes. Anthropic's route is a notebook in its cookbook that needs an Anthropic API key and a place to run notebooks, such as Colab. OpenAI's route is the Generate button in its Playground, and its guide also prints the meta-prompt behind it. Anthropic's notebook says the prompt it writes is not guaranteed to be optimal, so plan to change it. The prompt templates lesson covers variables in more depth.

Run Anthropic's metaprompt on one task

The notebook calls itself a tool for the "blank page problem". You enter a task and, if you like, the names of the variables you want, and it gives you a starting point to iterate on. It is built for single-turn question and answer prompts, not multi-turn ones. The metaprompt notebook holds a long prompt of example prompts for different tasks. The notebook's own words say half a dozen, and the text it prints has five. Your task goes in at the end, followed by three instructions: list the inputs, plan the structure, then write the instructions. It tells the model that long variable values should come before the directions that use them.

The cell you edit is short. These are its lines as the notebook prints them, with its default task and two variables:

TASK = "Draft an email responding to a customer complaint"  # Replace with your task!
# Optional: specify the input variables you want Claude to use. If you want
# Claude to choose them for you, leave VARIABLES as an empty list.
VARIABLES = ["CUSTOMER_EMAIL", "COMPANY_NAME"]

# TASK = "Choose an item from a menu for me given my preferences"
# VARIABLES = ["MENU", "PREFERENCES"]

1. Open the notebook in Colab or any place you run notebooks, and run the first code cell. It installs the anthropic package.
2. Put your API key where the notebook says to, or leave it blank and set the ANTHROPIC_API_KEY environment variable.
3. Replace the TASK line with your own task. Write what a model should do and for whom, not just a topic.
4. List variable names in capitals if you want specific ones. The notebook writes them as {$NAME}. Leave the list empty and Claude picks the names.
5. If you changed the task, put example values for your variables in VARIABLE_VALUES in the testing section near the bottom.
6. Run all the cells. The notebook warns that the long metaprompt takes 20 to 30 seconds. Your template appears in section 1.

Generate a prompt with OpenAI's meta-prompt

OpenAI's prompt generation guide says the Generate button in the Playground makes prompts, functions and schemas from a description of your task. For prompts it uses meta-prompts built on OpenAI's prompt engineering best practices, with a different one for each kind of output, such as audio. The guide prints the text meta-prompt below. Its code sends it as the system message, and the user message is "Task, Goal, or Current Prompt:" followed by your task.

Because it is plain text, you can also paste it at the top of a chat window and put your task underneath. That shortcut is this lesson's suggestion, and the guide does not describe it.

Given a task description or existing prompt, produce a detailed system prompt to guide a language model in completing the task effectively.

# Guidelines

- Understand the Task: Grasp the main objective, goals, requirements, constraints, and expected output.
- Minimal Changes: If an existing prompt is provided, improve it only if it's simple. For complex prompts, enhance clarity and add missing elements without altering the original structure.
- Reasoning Before Conclusions**: Encourage reasoning steps before any conclusions are reached. ATTENTION! If the user provides examples where the reasoning happens afterward, REVERSE the order! NEVER START EXAMPLES WITH CONCLUSIONS!
    - Reasoning Order: Call out reasoning portions of the prompt and conclusion parts (specific fields by name). For each, determine the ORDER in which this is done, and whether it needs to be reversed.
    - Conclusion, classifications, or results should ALWAYS appear last.
- Examples: Include high-quality examples if helpful, using placeholders [in brackets] for complex elements.
   - What kinds of examples may need to be included, how many, and whether they are complex enough to benefit from placeholders.
- Clarity and Conciseness: Use clear, specific language. Avoid unnecessary instructions or bland statements.
- Formatting: Use markdown features for readability. DO NOT USE ``` CODE BLOCKS UNLESS SPECIFICALLY REQUESTED.
- Preserve User Content: If the input task or prompt includes extensive guidelines or examples, preserve them entirely, or as closely as possible. If they are vague, consider breaking down into sub-steps. Keep any details, guidelines, examples, variables, or placeholders provided by the user.
- Constants: DO include constants in the prompt, as they are not susceptible to prompt injection. Such as guides, rubrics, and examples.
- Output Format: Explicitly the most appropriate output format, in detail. This should include length and syntax (e.g. short sentence, paragraph, JSON, etc.)
    - For tasks outputting well-defined or structured data (classification, JSON, etc.) bias toward outputting a JSON.
    - JSON should never be wrapped in code blocks (```) unless explicitly requested.

The final prompt you output should adhere to the following structure below. Do not include any additional commentary, only output the completed system prompt. SPECIFICALLY, do not include any additional messages at the start or end of the prompt. (e.g. no "---")

[Concise instruction describing the task - this should be the first line in the prompt, no section header]

[Additional details as needed.]

[Optional sections with headings or bullet points for detailed steps.]

# Steps [optional]

[optional: a detailed breakdown of the steps necessary to accomplish the task]

# Output Format

[Specifically call out how the output should be formatted, be it response length, structure e.g. JSON, markdown, etc]

# Examples [optional]

[Optional: 1-3 well-defined examples with placeholders if necessary. Clearly mark where examples start and end, and what the input and output are. User placeholders as necessary.]
[If the examples are shorter than what a realistic example is expected to be, make a reference with () explaining how real examples should be longer / shorter / different. AND USE PLACEHOLDERS! ]

# Notes [optional]

[optional: edge cases, details, and an area to call or repeat out specific important considerations]

• Reasoning Before Conclusions tells the generator to put reasoning ahead of any verdict, and to reverse examples that give the verdict first.
• Preserve User Content keeps your rules, examples, variables and placeholders. Paste what you already have into the task.
• Constants says to put guides, rubrics and examples inside the prompt, because they are not open to prompt injection.
• Output Format asks for the format in detail, with length and syntax, and leans toward JSON for classification.
• The guide adds that OpenAI may bring in more advanced techniques later, such as DSPy, which lesson 4 covers.

How the two generators differ

Both tell the model to put reasoning or a justification ahead of the verdict. They differ in how you run them and what they hand back. They also differ on output format. OpenAI's meta-prompt has an Output Format rule, while the notebook only says to name the tags when you want output inside tags.

What Anthropic says its Console generator does

Anthropic also builds a prompt generator into its Console. Its account on X announced it on 10 May 2024: describe what you want to achieve, and Claude writes the prompt using techniques like chain-of-thought reasoning. The post names the technique and not the steps, so the Console's own wording is not something you can read here. The cookbook's metaprompt is the one whose full text is public.

A screenshot of X's own embed of the post, taken on 6 November 2026, with the time in UTC. Select it to open the post on X.

Check what the generator wrote

A generated prompt looks finished even when it is not, because it arrives in tidy sections with headings and a firm tone. These five checks come from what the two generators themselves ask for, and the table is this lesson's own list.

Common mistakes

• Describing the job in three words. "Write emails" gives the generator nothing to shape. Say who the email goes to, what the input is and what a good result looks like.
• Handing a long prompt that already works to a generator and expecting a fresh rewrite. For a complex prompt, OpenAI's meta-prompt only adds clarity and missing parts and keeps the original structure.
• Using the metaprompt for a chatbot with many turns. Anthropic built it for single-turn prompts.
• Changing the task and forgetting VARIABLE_VALUES. The notebook then asks for each value with input(), or uses a placeholder when it cannot ask.
• Saving the first draft. Anthropic says outright not to be afraid to change it.

Try it yourself

• Pick one task you repeat at work, such as replying to a complaint or tagging support tickets.
• Write it as a task line plus the names of the inputs that change.
• Get one draft from the notebook and one from OpenAI's meta-prompt, then put both through the check table.
• Run the better draft on three real inputs and write down where it slips. Those slips are what the next lessons help you fix.

Next: improve a prompt you already have

A generated prompt is a first version. Lesson 2 starts from a prompt that already exists, shows how OpenAI's optimizer and edit meta-prompt rework it, and reports what Anthropic says its prompt improver does.

Frequently asked questions

Sources

1. Anthropic on GitHub: claude-cookbooks, metaprompt notebook (read November 6, 2026)https://github.com/anthropics/claude-cookbooks/blob/main/misc/metaprompt.ipynb
2. OpenAI Docs: Prompt generation (read November 6, 2026)https://developers.openai.com/api/docs/guides/prompt-generation
3. X: Anthropic (@AnthropicAI) on generating prompts in the Anthropic Console (read November 6, 2026)https://x.com/AnthropicAI/status/1788958483565732213