ChatGPT Prompting Guide

0:05 A good ChatGPT prompt says exactly what you want, for whom, in what format and how long, and gives the context ChatGPT cannot guess. That is the core of OpenAI's own advice: prompts should be "clear, specific, and provide enough context", and "Prompt engineering often requires an iterative approach", so you improve the answer over a few turns rather than expecting it in one. This page gathers what OpenAI publishes about prompting, for everyday ChatGPT use and for developers using GPT-6 through the OpenAI API, with examples and the mistakes to avoid, as of Nov 6, 2026.

Quick facts

The basics: OpenAI's three tips for ChatGPT

OpenAI's help article for ChatGPT users is short, and its three tips are the ones that matter most (OpenAI Help Center):

1. Be clear and specific. Give enough background for ChatGPT to follow you. OpenAI's example: instead of "Write about dogs", ask for "a 200-word guide to housetraining a puppy for first-time owners".
2. Refine over several turns. Start with a draft prompt, read the answer, then add context, narrow the request or simplify it. OpenAI's example moves from "Summarize this article" to a three-bullet summary "for a non-expert audience".
3. Ask for a tone. Name it with plain adjectives: formal, friendly, professional, humorous, serious.

Nine practices from OpenAI's API guide that also work in chat

OpenAI's older article for developers lists practices that carry straight over to ChatGPT (OpenAI Help Center):

Source: Best practices for prompt engineering with the OpenAI API, OpenAI Help Center, read Nov 6, 2026.

Example prompts, weak and strong

0:27 The stronger prompts follow the same pattern: the audience, the format, a length and the facts only you know. The two shapes come from OpenAI's guidance above; the specific examples are this page's own. OpenAI Academy's own video makes the same point with dinner: "Help me with dinner" leaves ChatGPT guessing, while "three easy dinners for two people this week" with a short shopping list says what you want, who it is for and what you want back.

Prompting GPT-6: OpenAI's newest advice

OpenAI's GPT-6 guide for developers adds a "Prompting best practices" section, written for the behavior of GPT-6 Astra and offered as a starting point for the whole family (OpenAI developer docs). Its points, in brief:

• Initiative. Astra tends to stop and ask when more input could change the result. If you want it to make reasonable assumptions and keep going, say so, and ask for approval only once a concrete result is ready, except before destructive or irreversible steps.
• Instructions in files. Astra reacts strongly to context such as skills and AGENTS.md files. Check them for conflicting instructions and say that the user's instructions win over a skill's.
• Writing style. Astra leans toward long answers full of lists and tables. Ask for the style you want: short paragraphs, lists only for steps or parallel items, plain words, the main point early.
• Delegation. Astra may hand work to subagents less than you want; tell it when to work in parallel.
• Testing. For code, Astra may test more than a small change needs; ask it to match testing to the change.

1:08 In ChatGPT itself the same habits apply: if the answer is too long or too formatted, say "answer in two short paragraphs, no headings" once, or put it in your custom instructions so it applies everywhere.

Prompting inside ChatGPT's tools

• Work and agent tasks. OpenAI's guide to ChatGPT Work says a good goal names the outcome, the sources or plugins to use, the constraints, what success looks like and when to stop for your review, such as "use only these sources" or "compare the top five options" (OpenAI docs). The agent help article warns against vague prompts like "handle everything". See How to Use ChatGPT Agent Mode.
• Deep research. Describe the outcome you want; ChatGPT proposes a research plan you can edit before it starts, so fix the plan rather than rewriting the prompt. See Deep Research.
• Images. Describe subject, style, lighting and aspect ratio, then edit in steps; ChatGPT Images keeps details steadier from one edit to the next.
• Reusing prompts. Put standing preferences in custom instructions (up to 5,000 characters on paid plans since Jul 15, 2026), subject-specific rules in project instructions, and repeatable procedures in Skills.

What not to do

• Don't be vague about the result. "Make it better" leaves ChatGPT to guess what better means.
• Don't pack ten tasks into one message. Break them into steps, as OpenAI's API guide advises.
• Don't paste secrets. OpenAI's agent article tells users not to type passwords or private information into messages.
• Don't treat a confident answer as checked. Ask ChatGPT to search and cite, then open the sources.
• Don't rely on "magic words". OpenAI's guidance is about clarity, context, examples and iteration; it names no secret phrases, and no source read for this page shows that one exists.

For developers: prompts belong in code

OpenAI's API prompting page now tells developers to "Treat prompts as application code": keep them in versioned modules, put role and tone in the system message and task details and examples in user messages, and run tests every time a prompt changes. Saved prompt objects were de-emphasized on Jun 3, 2026 and the v1/prompts endpoint shuts down on Nov 30, 2026 (OpenAI developer docs).

Frequently asked questions

How do I write a good ChatGPT prompt? Say what you want, who it is for, the format and the length, and include any facts ChatGPT cannot know. Then refine: OpenAI's own help article says prompting "often requires an iterative approach".

Does ChatGPT need special prompt templates? No. OpenAI's published guidance for ChatGPT asks for clear, specific requests with context; templates help only as reminders to include audience, format and length.

Should I give ChatGPT examples? Yes, when the format matters. OpenAI's API guide recommends a couple of sample inputs and outputs ("few-shot") before the real input, and the same works in a ChatGPT chat.

How do I make ChatGPT answers shorter? Give a concrete length, such as "3 to 5 sentences" or "under 100 words", and say "no headings or lists". To make it stick, add the preference to custom instructions in Settings > Personalization.

Is there a GPT-6 prompting guide? Yes. OpenAI's developer guide for the GPT-6 family has a "Prompting best practices" section, written for GPT-6 Astra, covering initiative, instruction files, writing style, subagents and testing.

Can I save prompts in ChatGPT? There is no prompt library as such. Use custom instructions for standing preferences, project instructions for one subject, and skills or plugins for repeatable procedures; Custom GPTs retire on Dec 11, 2026.

Sources

• Prompt engineering best practices for ChatGPT, OpenAI Help Center, read Nov 6, 2026
• Best practices for prompt engineering with the OpenAI API, OpenAI Help Center, read Nov 6, 2026
• Using GPT-6: Prompting best practices, OpenAI developer docs, read Nov 6, 2026
• Prompting, OpenAI developer docs, read Nov 6, 2026
• Pricing (fine-tuning note), OpenAI developer docs, read Nov 6, 2026
• Get started with ChatGPT Work, OpenAI docs (learn.chatgpt.com), read Nov 6, 2026
• ChatGPT agent, OpenAI Help Center, read Nov 6, 2026
• Deep research in ChatGPT, OpenAI Help Center, read Nov 6, 2026
• Projects in ChatGPT, OpenAI Help Center, read Nov 6, 2026
• ChatGPT release notes, OpenAI Help Center, read Nov 6, 2026
• How to Ask ChatGPT Better Questions / OpenAI Academy, OpenAI, Sep 16, 2026 (video)