ChatGPT prompts do not need a secret formula. A useful prompt states the result you want, adds the context that could change it and tells ChatGPT what the finished answer should look like. The examples below cover everyday writing, summaries, comparisons, planning and research. Replace the bracketed fields, then review the answer before you use it.

OpenAI’s current Prompting guide groups the useful details into four parts: goal, context, output and boundaries. It also says you only need the parts that help. This guide turns that advice into copyable prompts without pretending one template fits every task.

On this page

A short ChatGPT prompt can still be specific

Start with the result, not a long role description. Add an audience or format only when it changes the answer.

A practical base template is:

Create [result] for [audience or purpose].

Use this context:
[paste the relevant facts or source material]

Return [format, length and level of detail].
Keep [facts or wording that must not change].
If information is missing, label it instead of guessing.

The last two lines are boundaries. They matter when an incorrect date, invented fact or unintended action would make the result unusable. For a casual brainstorm, you may not need them.

Prompt for a concise summary

Use this when you have the source material and want a shorter version, not new research.

Summarize the text below for [audience].

Put the main decision or takeaway first.
Then list:
- the three most important supporting points
- actions, owners and due dates that are explicitly stated
- information that is missing or unclear

Keep all dates and numbers unchanged.
Do not add facts that are not in the text.

Source:
[paste text]

Check the summary against the source. A polished sentence can still lose a qualification such as “may,” “up to” or “only in selected regions.”

Prompt to rewrite without changing the facts

This version gives the model a clear editing job and protects details that must survive the rewrite.

Rewrite the draft below for [audience].
Use a [friendly / formal / direct] tone and keep it under [length].

Preserve every name, date, number and product term.
Do not add examples or claims.
After the rewrite, list any sentence whose meaning was ambiguous.

Draft:
[paste draft]

OpenAI’s prompt engineering help article recommends clear instructions, enough context and descriptive words when you want a particular tone. “Friendly” or “formal” is more useful than “make this better.”

Prompt to compare two options

Tell ChatGPT what the comparison is for. Otherwise it may spend space on differences that do not affect your decision.

Compare [option A] and [option B] for [specific situation].

Use these criteria:
- [criterion 1]
- [criterion 2]
- [criterion 3]

Show verified differences in a table.
Then recommend one option and explain the tradeoff in three bullets.
If a required fact is missing or current information is needed, flag it.

Information I already have:
[paste facts, links or notes]

For prices, policies or availability, ask for current sources and check their dates. Do not treat an uncited comparison as a verified fact.

Prompt to turn notes into an action plan

This prompt separates decisions from work that has not been assigned.

Turn these notes into an action plan for [team or project].

Use a table with:
Action | Owner | Due date | Dependency | Status

Include only actions supported by the notes.
Write "not assigned" when there is no owner.
Write "not specified" when there is no due date.
After the table, list unresolved decisions and risks.

Notes:
[paste notes]

This is especially useful when meeting notes mix decisions, suggestions and background. The explicit labels stop an absent owner from becoming an invented one.

Prompt for research with sources

Ask for the scope, evidence and stopping point. “Research this” leaves too much open.

Research [question] for [audience or decision].

Cover:
- [aspect 1]
- [aspect 2]
- [aspect 3]

Use current primary sources where possible.
Limit the research to information published or updated since [date].
Cite the source beside each material fact.
Separate reported facts from your analysis.
End with the three questions that remain unanswered.

OpenAI’s Prompting guide recommends naming when web search is needed and asking for sources when you want to check the result. A citation helps you inspect a claim; it does not remove the need to read the source for an important decision.

Prompt to plan a week of work

Give the real constraints before asking for a schedule.

Plan my work from [start date] to [end date].

Fixed commitments:
[paste meetings, deadlines and unavailable times]

Tasks:
[paste tasks with estimated effort and priority]

Return a daily schedule with time blocks.
Do not move fixed commitments.
Leave [amount] of unscheduled time each day.
If the workload does not fit, show which task should move and why.

Review the plan against your calendar. ChatGPT cannot know about a commitment you did not include, and a connected calendar should only be used when you explicitly provide or authorize it.

Prompt to learn from your own material

This version keeps the explanation grounded in notes you choose.

Teach me [topic] using only the notes below.
Assume I am at [level].

First give a five-point overview.
Then ask one question at a time and wait for my answer.
After each answer, explain the correction using the notes.
If the notes do not cover something, say so.

Notes:
[paste notes]

The one-question instruction makes the task interactive. If ChatGPT sends a full quiz, ask it to return to the first question and wait.

Improve the answer with one focused follow-up

The first prompt does not need to predict every problem. OpenAI recommends reviewing the response and asking for the specific change you want.

Problem in the answer Follow-up prompt
Too long “Cut this to 180 words. Keep the two exceptions and every date.”
Wrong level “Explain the same idea for a reader new to the topic. Define each technical term.”
Unsupported detail “Remove claims that are not supported by my source text. List what you removed.”
Wrong format “Keep the content, but return it in a table with these columns: [columns].”
Missed requirement “Add [missing requirement] without changing [protected detail].”

Correct one problem at a time. A vague “try again” gives the model little information about what failed.

When to start a new chat

Stay in the same conversation when the follow-up depends on the earlier material. Start a new chat when you want a clean context, the topic has changed or earlier instructions are steering the answer in the wrong direction.

If you reuse a prompt, keep stable instructions separate from today’s input. Replace the source block, dates and audience without carrying over old facts by accident.

For deeper structure, see the system prompt guide and system prompt examples. The AI prompt examples guide covers vendor examples by task, while the prompt library holds individual reusable prompts.

Questions

What is a good prompt for ChatGPT?

A good prompt says what result you need and includes the context that could change it. Add the desired format, audience and boundaries when they matter.

Do I need to use a prompt formula?

No. OpenAI says there is no required rigid formula. Goal, context, output and boundaries are useful parts, not mandatory fields.

Should I tell ChatGPT to act as an expert?

A role can help when it changes the audience, vocabulary or criteria. It is less useful than naming the actual result, source material and constraints.

How do I stop ChatGPT from making up missing details?

Tell it to use only the supplied sources and to label missing information instead of guessing. Then verify important claims yourself.

Can I refine a prompt after ChatGPT answers?

Yes. Point to the specific problem, such as length, tone, missing evidence or the wrong format, and request one focused change.

Sources

OpenAI, Prompting; OpenAI Help Center, Prompt engineering best practices for ChatGPT. Documentation checked October 14, 2026. Features and interface controls can change.