How to use the prompts on this site

Last checked 3 October 2026

Every prompt on this site is a set of standing instructions for an AI assistant. It is written to be given to the model before your request, so that the model knows what role it is in, how to work through the task and what shape its answer should take. This page explains how the prompts are put together and the ways you can use one.

How a prompt is built

Each prompt is plain text, usually between about 1,500 and 4,000 words. In characters, the shortest is just over 9,000 and the longest is about 25,500; the middle of the range is about 14,000. Each prompt page shows the character count above the text.

The parts are the same across the library:

An opening line that starts "You are a ..." and names the role, such as an email assistant or a code reviewer.
Sections with "## " headings that set out what input to expect, how to approach the task, rules to keep to, situations that need care, and what to check before answering.
An output format section that says what the answer should contain and in what order.
A closing label followed by a bracketed slot, for example "Email task, along with any thread, draft, or context:" and then [REQUEST] on the next line.

The Email Assistant is a typical example: ten sections, about 2,400 words, ending in [REQUEST]. The Code Reviewer ends in [CODE_AND_CONTEXT] and the Proposal Writer in [PROPOSAL REQUEST AND SUPPORTING MATERIALS]. The slot name tells you what the prompt expects to receive.

The bracketed slot at the end

The slot marks where your own request goes. What you do with it depends on how you use the prompt:

If you paste the whole prompt as the first message of a chat, replace the slot with your request, your material, or both. The model reads the instructions and then your request in one message.
If you put the prompt in a setting that applies to every chat (custom instructions, project instructions, a custom assistant), delete the closing label and the slot. Your messages in the chat then take the place of the slot.

Leaving the slot in a standing setting does no harm in most cases, but the model may treat the literal text [REQUEST] as an empty request and ask you what you want.

Placeholders inside answers

Some prompts tell the model to leave bracketed placeholders in its drafts for facts you have not given it, such as [DATE], [NAME] or [LINK TO DECK]. These appear in the answer, not in the prompt you paste. They mark the places where the model would otherwise have had to invent a detail. Before you use a draft, search it for square brackets and fill each one in, or give the model the missing facts in a follow-up message and ask it to revise.

Three ways to use a prompt

As a system prompt or custom instructions. Chat apps and APIs have a place for instructions that sit above the conversation. Text there applies to every message without being repeated. See What is a system prompt? for how this differs from an ordinary message.

In a project or custom assistant. ChatGPT and Claude have projects, and Gemini has Gems and skills. You save the prompt once and every chat you start there uses it. This suits a prompt you use repeatedly.

Pasted as the first message. Copy the prompt with the Copy prompt button, paste it into a new chat, and replace the slot with your request. This works in any chat app and needs no settings. The instructions stay in effect for the rest of that chat, and a new chat starts without them.

Each prompt page also has a Raw .txt link, which serves the prompt as a plain text file. You can save it and upload it where an app accepts files as reference material.

Follow-up messages

After the first answer, carry on in the same chat. Many prompts include instructions for follow-ups: the email prompt, for example, tells the model to revise the current draft when you ask for it to be shorter or less formal rather than starting again. You do not need to repeat the prompt in later messages. If you start a new chat, the prompt has to be supplied again, unless it is saved in a project, custom assistant or instructions setting.

When a prompt is longer than a field allows

Some fields have character limits smaller than these prompts. ChatGPT's custom instructions, for instance, hold 1,500 or 5,000 characters depending on plan, so no prompt on this site fits there whole. You have three options:

Use a different place that holds the whole text: pasting it into a chat, project instructions, or a file uploaded to a project or custom assistant.
Put the full prompt in an uploaded file and write a short instruction in the field telling the model to follow that file.
Trim the prompt to fit.

To trim, keep the opening "You are a ..." paragraph, the output format section and any section of rules. Then remove sections that do not apply to how you will use it, such as guidance for input types you will never send. Removing whole sections keeps the remaining instructions consistent with each other. The answers will no longer reflect the instructions you removed.

Platform guides

Step-by-step instructions, with current limits, for each of the main chat apps and their APIs: