Prompting ChatGPT

ChatGPT is OpenAI's general-purpose assistant. Most people use it as a chat window, but how well it performs depends less on clever wording than on three ordinary things: whether it knows what you want, whether it has the material it needs, and whether you told it what the finished answer should look like.

Instructions reach ChatGPT at two levels. Standing guidance, such as custom instructions in your settings or the instructions attached to a project or custom GPT, applies to every conversation it covers. The message you type applies to the task in front of it. Putting stable preferences (your role, your audience, the tone you want, formats you never want) in the standing layer keeps individual prompts short and stops you from retyping the same preamble.

Context works the same way. You can paste text, attach files, or keep reference documents in a project, and the assistant will work from them. It cannot work from material it has not been given, and when a prompt leaves a gap it tends to fill that gap with something plausible. Supplying the facts yourself, and saying what to do when information is missing, is the most reliable way to reduce invented detail.

The prompts in this library are written as structured text with labeled sections because that format suits this product well: OpenAI's own guidance recommends separating instructions from reference material with clear delimiters. Paste a full prompt for work that matters, and use the short version when you only need a first pass.

This page deliberately avoids version names, limits and prices, which change often. Feature availability also varies by plan and region, so check the product itself for what your account includes.

How to get better results

Put durable preferences in instructions, not in every message

OpenAI's guidance distinguishes high-level instructions, which set how the model should behave throughout, from the user message that carries the specific request. In the ChatGPT app the equivalent is custom instructions or project instructions. Use that layer for things that are true of all your work: who you are writing for, spelling conventions, how long answers should be. Keep the task-specific detail in the message.

Separate the parts of a long prompt with labels or delimiters

When a prompt contains instructions, background and source text, mark where each begins and ends. OpenAI's prompt engineering guide suggests Markdown headings and XML-style tags for this, so the model can tell an instruction from the material it is supposed to act on. This matters most when you paste a document: without a boundary, a sentence inside the document can be read as something you asked for.

Show an example of the output you want

Describing a format in the abstract is harder than showing one. The guide recommends few-shot prompting: include a small number of input and output pairs that demonstrate the pattern, and the model will usually follow it. Vary the examples enough that it picks up the pattern and not the surface wording, and say plainly if the examples are illustrative and should not be copied.

Give it the facts; do not expect it to know your situation

The guide advises including relevant context in the prompt, particularly information the model could not otherwise have, such as your own data, internal documents or recent decisions. Attach the file or paste the section. Then add a rule for gaps, for example asking it to list what is missing instead of assuming. That single instruction prevents a good share of confident but invented detail.

Match the amount of step-by-step direction to the mode you are using

OpenAI's documentation notes that reasoning-oriented models respond well to a clear goal and constraints, with less need for a prescribed procedure, while other models benefit from more explicit, ordered instructions. In practice: if you have selected a thinking mode, state the outcome and the criteria for success. Otherwise, spell out the steps. If results disappoint, try moving in the other direction.

State the output format last and be literal about it

Say whether you want a table, a numbered list, plain paragraphs or a specific file type, how long it should be, and what to leave out. Format requests placed at the end of a prompt are easy for you to check and easy to adjust on the next turn. If the first reply is close, ask for the specific change instead of rewriting the whole prompt.

Where it fits

Well suited to

  • Drafting and revising text where you can supply the audience, tone and source material
  • Structured planning tasks such as briefs, outlines and specifications
  • Working through an uploaded document, spreadsheet or image with follow-up questions
  • Repeated tasks where a project or custom instructions can hold the standing rules
  • Explaining code or concepts at a level you specify

Watch out for

  • It can state incorrect facts fluently; verify names, numbers, quotations and references before relying on them
  • Saved memory and custom instructions affect every answer, so review them if responses start drifting in an odd direction
  • Long conversations can lose track of early instructions; restate the key constraints or start a fresh chat
  • Features such as file handling, browsing and projects differ by plan and change over time
  • Check your organization's policy before pasting confidential or personal data

Prompts that work with ChatGPT

These prompts are expected to work with ChatGPT based on their structure. Check each prompt page for its testing status; see how compatibility is labelled.

Web Design

Web Development

Content Writing

SEO

Marketing

Business

Research

Coding

Graphic Design

Productivity

Education

E-commerce

Official documentation