How to Write Better ChatGPT Prompts: 8 Steps With Examples

chatgptprompt-engineering

Most people learn how to write better ChatGPT prompts by trial and error: type something short, get a bland reply, add a detail, try again. That works, but it is slow. This checklist gives you the details to add up front, with a real prompt before and a rewritten prompt after for each step, so you can get a useful answer in fewer tries.

Short answer: To write a better ChatGPT prompt, tell it who the answer is for, what it should achieve, the facts it cannot know, the limits it must respect and the format you want back. Then ask it to check with you before it starts.

What makes a ChatGPT prompt better

A good prompt removes guesswork. Every detail you leave out, ChatGPT has to fill in with a guess, and a guess tends to be the most common choice. We explain that in more depth in why ChatGPT gives generic answers.

To show which details everyday prompts leave out, we ran every prompt in this article through the check behind our free prompt grader. The grader uses Jev, a classifier from TypeSafe. For four items (Who it's for, Format and length, Your details, and What a good result looks like) it asks two questions: does this task need it, and does the prompt state it? An item is flagged as missing only when the task needs it and the prompt leaves it out. Items close to that line show as "could add". We ran each prompt at least twice on 2026-11-03. Results near the line can shift slightly between runs, so where the runs disagreed, the table says how often each result came up. For the customer reply, Your details sat on the line: some runs showed it as "could add", others as not needed. These results come from Jev 1.14, checked on 2026-11-03, and can change when TypeSafe releases a newer version.

Before prompt Task type Missing Could add
write a marketing email for our new product launch, make it catchy Writing for people Who it's for, Your details none
reply to a customer who is angry their order is late Writing for people What a good result looks like; Format and length in 10 of 18 runs Format and length in the other 8 runs
make this resume bullet sound better: managed social media accounts and increased followers Writing for people Who it's for Your details, Format and length (11 of 12 runs), What a good result looks like (10 of 12 runs)
give me an outline for a blog post about remote work Writing for people none Who it's for
why is my python code so slow? it loops through a big list and checks stuff Factual question none none

Every "after" prompt below is a rewrite we wrote by hand for this article, and every one came back with nothing missing and nothing to add in our final run. Paste any of them into the free prompt grader to check them yourself.

How to improve ChatGPT prompts

Use these eight steps in order. For a quick question you may only need the first three. For anything you will publish, send or rely on, use all eight.

1. Say who the answer is for

The audience decides vocabulary and tone. Without one, ChatGPT has to pick a reader for you.

Before (could add: Who it's for):

give me an outline for a blog post about remote work

After:

Write an outline for a blog post for agency owners deciding whether to
go fully remote. Format:
- a working title under 60 characters
- 5 to 7 H2 sections, each with 2 to 3 bullet points
- a one-line summary of the main argument at the top

The first line names the readers. The format lines are not what made the original pass on format: the check judged that this outline task does not need a stated format, so it showed Format and length as "Not needed for this task".

2. State the goal

Tell ChatGPT what the answer should make happen. A goal gives the model a finish line and a way to decide what to leave out.

Before (missing: What a good result looks like; Format and length is missing or could add, depending on the run):

reply to a customer who is angry their order is late

After:

Write a reply to a customer of our online plant shop. She ordered a
gift for her mother's birthday, and the order is 4 days late. The new
delivery date is [date].

Goal: she keeps her order instead of cancelling, and she knows exactly
when it will arrive.

Write the email body only, under 120 words.

The "Goal" line covers What a good result looks like, and the last line covers Format and length.

3. Give the facts it cannot know

ChatGPT cannot see your product, your code or your data. Paste the real input.

Before (nothing flagged):

why is my python code so slow? it loops through a big list and checks stuff

The check passes this prompt. It reads "loops through a big list" as a stated detail, and a factual question needs no audience or format. That shows a limit of any automated check: it can tell whether you gave details, but not whether those details are enough. Nobody can say why a loop is slow without seeing the loop.

After:

This Python function takes about 40 seconds on a list of 200,000
email addresses. It checks each address against a list of 50,000
blocked addresses.

[paste the function here]

Explain why it is slow and show a faster version.

The 40 seconds and the list sizes here are part of our example, so replace them with your own measurements.

4. Swap vague words for testable ones

"Catchy", "better" and "professional" can mean almost anything. Replace each one with something you could check.

Before (missing: Who it's for):

make this resume bullet sound better: managed social media accounts and increased followers

After:

Rewrite this resume bullet for a mid-level marketing manager role at a
B2B software company. Hiring managers will skim it in a few seconds,
so it has to show a result.

Original: managed social media accounts and increased followers
Details: I ran our LinkedIn and Instagram accounts for 18 months and
grew followers from 4,000 to 11,500.

Make it one line under 25 words that starts with an action verb and
includes the follower growth as a number.

"Sound better" became three things you can check: an action verb, a number and a word limit. The follower figures are invented for the example, so use your own.

5. Set constraints

Constraints remove the safe defaults ChatGPT falls back on. Good constraints include length, reading level, words or claims to avoid, and what is out of scope.

Before (missing: Who it's for, Your details):

write a marketing email for our new product launch, make it catchy

After:

Write a launch email for TaskNest, a shared to-do app for families.
Readers are parents who joined our waitlist.

Constraints:
- under 150 words
- subject line under 50 characters
- no exclamation marks and no "revolutionary" or "game-changing"
- one call to action: start the free 14-day trial

TaskNest is a made-up product for this example. Use your own product details.

6. Ask for a specific format

When you define the shape of the answer, you decide how long it is and what goes where. Ask for a table, a numbered list, headings, a word count or a draft plus alternatives.

Before (missing: What a good result looks like; Format and length is missing or could add, depending on the run):

reply to a customer who is angry their order is late

After:

Write a reply to a customer of our online plant shop. Her birthday
gift order is 4 days late and the new delivery date is [date].

Goal: she keeps the order and knows when it will arrive.

Format: the email body only, 3 short paragraphs, under 120 words.
Paragraph 1 apologizes once, paragraph 2 gives the new date, paragraph
3 offers free shipping on her next order.

7. Give it a role only when the role adds knowledge

"You are an expert" on its own adds little. A role helps when it brings a specific point of view or set of standards with it.

Before (missing: Who it's for):

make this resume bullet sound better: managed social media accounts and increased followers

After:

Act as a recruiter who screens marketing manager applications for B2B
software companies. Read this bullet the way you would in a 30-second
resume scan, tell me what is missing, then rewrite it as one line
under 25 words.

Bullet: managed social media accounts and increased followers
Details: I ran our LinkedIn and Instagram accounts for 18 months and
grew followers from 4,000 to 11,500.

8. Invite questions before it answers

One sentence at the end of your prompt asks ChatGPT to tell you what it still needs, which can surface a detail you forgot. The check does not look for this sentence, so it adds no item. We include it because it costs one line.

Before (could add: Who it's for):

give me an outline for a blog post about remote work

After:

I run a 12-person design agency and I'm writing a blog post for other
agency owners who are deciding whether to go fully remote.

Goal: help them decide, using what we learned in our first year remote.
Our main lessons: client calls got easier, onboarding junior designers
got harder, and we cut our office costs.

Write an outline with:
- a working title under 60 characters
- 5 to 7 H2 sections, each with 2 to 3 bullet points
- one section on what we would do differently

Avoid generic pros-and-cons lists. Ask me up to 3 questions first if
anything is unclear.

How to write better ChatGPT prompts: put it together

Here is the angry customer prompt with steps 1, 2, 5, 6 and 8 combined:

Write a reply to a customer of our online plant shop. She ordered a
gift for her mother's birthday, and the order is 4 days late because
our courier lost a pallet.

Goal: she keeps her order instead of cancelling, and she knows exactly
when it will arrive.

Constraints:
- under 120 words
- apologize once, without blaming the courier
- offer free shipping on her next order
- new delivery date: [date]

Format: the email body only, no subject line.

Ask me up to 2 questions first if anything is unclear.

The original was 11 words, and the check flagged What a good result looks like as missing, and Format and length as missing or could add, depending on the run. This rewrite came back with nothing missing and nothing to add. It is longer, and each added line gives ChatGPT a fact or a limit it would otherwise have guessed. The plant shop details are invented for the example, so swap in your own.

Common mistakes when improving prompts

  • Adding adjectives instead of facts. "High-quality, engaging, professional" makes a prompt longer without telling the model anything new.
  • Asking for everything at once. If you need research, a draft and a social post, ask for them in separate steps and review each one.
  • Regenerating instead of editing. A regenerated reply comes from the same prompt, so the same gaps get filled with guesses again. Fix the prompt, then regenerate.
  • Keeping the good prompt nowhere. When a prompt works, save it as a template and change only the details next time.

Check and improve your prompt for free

  • The prompt grader checks your prompt for the four items above and shows which ones are missing. It is free and needs no account.
  • The ChatGPT prompt generator turns a one-line task into a structured prompt you can edit.
  • The PromptMagicPro editor applies these steps to a rough prompt for you. You get 3 tries every 30 days with no account, and 25 a month with a free account.

If you want to understand why short prompts drift toward bland replies in the first place, read why ChatGPT gives generic answers.

Turn any idea into a working AI prompt.

Try the free generator