Guide
Why every prompt is written the way it is, per tool, from the tool maker's own guidance.
Every prompt on this site starts as a General version: the parts that ChatGPT, Claude, Gemini and Copilot all ask for, in plain sentences without any tool's own syntax. It works in any tool, and the four tool versions are made from it.
The parts
Labels, each alone on its line, always in this order. A prompt has only the labels it needs; Task is always there.
- Task: one sentence with the verb and what must come back.
- Context: why, for whom, and how the result is used. Your own details sit inside the sentence as [a field to fill in].
- Rules: one rule per line, as a list. Limits as numbers, never "short".
- Result: the shape and length of the answer.
- Example: only when the prompt has one.
- Material: the text you paste, last.
A role, when the prompt has one, is the first line, "You are ...", not a block.
Where this comes from
The General is promptwiki's own template: the common core of the four tool makers' guidance, the parts each of them names. Open a tab above to see what each tool changes, and which page says so.
OpenAI's prompting guide for ChatGPT at work lays a prompt out in sections with Markdown headings: what the model is working on, what to do, the constraints, and the text to work on last under its own heading. A prompt for ChatGPT on this site follows that layout.
The parts
Each section is a heading with one # and its content, in this order.
- Context: who the model is, what you are working on, background. Your own details sit inside the sentence as [a field to fill in].
- Instructions: exactly what to do, one item per line.
- Additional info / Constraints: the goal, the audience and tone, the format and length, anything non-negotiable, and how to judge the result. Where a detail is missing, the prompt says "not specified", as the guide's template does.
- Material: a last heading named for it, Text, Memo, Code or Document, with the pasted text under it.
Where this comes from
The ChatGPT Enterprise prompting guide in OpenAI's cookbook, read in full on 2026-10-12. Only OpenAI's advice for the ChatGPT app binds these prompts; advice for the API is recorded, not followed.
Anthropic's prompting best practices for Claude recommend XML-style tags to separate the parts of a prompt, and long material at the top with the task after it. A prompt for Claude on this site follows that page: a role as the first line when the prompt has one, then the tags in a fixed order.
The parts
Each tag opens and closes alone on its line, so the model sees exactly where a part begins and ends, and so do you.
- document or input: the text to work on, first.
- context: why, for whom, how the result is used. Your own details sit inside the sentence as [a field to fill in].
- instructions: what to do and what the result must look like, as prose, or numbered when order matters.
- example: one example, only when the prompt has one.
Where this comes from
The prompting best practices page in Anthropic's documentation, the page Anthropic itself names as the living reference, read in full on 2026-10-12. Only Anthropic's advice for the Claude app binds these prompts; advice for the API is recorded, not followed.
Google's prompting guide for Gemini in Workspace names four ingredients, persona, task, context and format, and shows them as natural language, the way you would brief a colleague, not as a form. A prompt for Gemini on this site keeps to that: prose only.
The parts
One short paragraph each, in this order, without labels.
- Role: the first line, when the prompt has one.
- Task: one sentence with a verb.
- Context: background and your own details, written as [a field to fill in], with constraints as counts and limits.
- Format: shape, length, tone.
- Material: the pasted text, last.
Where this comes from
The Gemini for Workspace prompting guide, read in full on 2026-10-12. Only Google's advice for the Gemini app binds these prompts; advice for the API is recorded, not followed.
Microsoft's guidance for writing Copilot prompts names four elements, the goal, the context, the expectations and the source, in plain sentences. A prompt for Copilot on this site keeps to that order, one short paragraph each.
The parts
- Goal: what to do and what output, one or two sentences. Always there.
- Context: why, who is involved, background, with your own details as [a field to fill in].
- Expectations: length, format, tone, and an example when there is one.
- Source: last. A pasted text in quotation marks, or a file as /[file name].
Where this comes from
Microsoft Support's article on writing Copilot prompts, with its related articles and Microsoft Learn, read in full on 2026-10-12. Only Microsoft's advice for the Copilot app binds these prompts.