How to write better ChatGPT prompts: a 5-part framework with examples

To write better ChatGPT prompts, give the model five things: a role, a clear task, the context it cannot guess, the format you want back, and the constraints it must respect. Most weak answers come from prompts that skip two or three of these. Below you will find each part with a before/after example, the follow-up tactics that fix a mediocre first draft, and a master template you can copy today.

Why most prompts get generic answers

ChatGPT fills every gap in your prompt with the most average assumption available. Ask for "a blog post about saving money" and it has to guess the reader, the length, the tone, the angle and the purpose. It picks the safest middle option for each one, and the result reads like every other article on the topic.

Better prompts are not longer for the sake of it. They remove guesses. Every sentence you add should answer a question the model would otherwise answer for you, badly.

The five parts below are a checklist for removing those guesses:

1. Role: who should be answering
2. Task: what exactly you want done
3. Context: the facts about your situation
4. Format: what the output should look like
5. Constraints: the rules and limits

You do not need all five for a quick question. For anything you plan to use, send, publish or decide on, you usually do.

Part 1: Role

A role tells the model which expertise, vocabulary and standards to apply. "You are a pediatric nutritionist" produces different advice than "you are a fitness influencer," even for the same question.

Before:

What should I eat before a half marathon?

After:

You are a sports dietitian who works with amateur runners. I am running my first half marathon on Sunday at 8 a.m. What should I eat the evening before and the morning of the race?

Why it works: the role sets the level of expertise and the audience in one line. The model now answers like a professional advising a beginner, not like a general search summary. Keep roles specific and relevant. "You are a world-class genius" adds nothing; "you are a hiring manager at a 40-person SaaS company" adds a point of view.

Part 2: Task

The task is the verb. Vague verbs like "help with," "look at" or "talk about" leave the model guessing what done looks like. Strong verbs like "rewrite," "compare," "list," "draft," "critique" or "summarize in" give it a finish line.

Before:

Help me with my cover letter.

After:

Rewrite my cover letter so the first paragraph explains why I want this specific job, and cut it to under 250 words. Here it is: [paste cover letter]

Why it works: "help" could mean proofreading, restructuring, praise or a full rewrite. The after version names the action, the priority and the size limit. One prompt, one main task. If you need three things, ask for them in order or in separate turns.

Part 3: Context

Context is everything true about your situation that the model cannot know: your audience, your goal, what you already tried, your industry, your deadline. This is where most prompts are thinnest and where the biggest gains are.

Before:

Write an Instagram caption for our new product.

After:

Write an Instagram caption for a new product launch.

Context:
- Business: Crumb & Co, a neighborhood bakery in Portland
- Product: a sourdough cinnamon roll, available Saturdays only, $4.50
- Audience: locals aged 25 to 45 who already follow us for weekend specials
- Goal: get people to pre-order through the link in our bio
- Last week's caption that performed well: "Croissant drop at 9. They were gone by 10:40."

Why it works: the model now knows the product, price, audience, goal and the voice that already resonates. Notice the format: short labeled lines are easier for you to write and for the model to use than one long paragraph. Pasting a real example of your past work is one of the fastest ways to transfer tone, which we cover in depth in how to make ChatGPT write in your voice.

Part 4: Format

If you do not specify a format, you get the default: a friendly intro, some bolded headers, bullet points and a summary paragraph. That is rarely what you need.

Tell the model the shape of the answer: a numbered list, a table of options, an email with a subject line, three variations, a JSON object, a script with timestamps, plain paragraphs with no headers.

Before:

Give me ideas for a team offsite.

After:

Give me 6 ideas for a one-day team offsite for 12 people. For each idea, give:
1. A name
2. One sentence describing the activity
3. Rough cost per person
4. Why it helps a remote team that rarely meets in person

No introduction and no closing summary.

Why it works: you get a fixed number of ideas, comparable details for each, and nothing to delete. "No introduction" is one of the most useful phrases you can add to almost any prompt.

Part 5: Constraints

Constraints are the rules: length, tone, words to avoid, reading level, what to leave out, what sources to use, what not to assume. They keep a good answer from drifting into something you cannot use.

Useful constraints include:

• Length: "under 150 words," "exactly 5 bullets," "one paragraph"
• Tone: "plain and direct, no hype," "warm but professional"
• Audience level: "explain for someone with no finance background"
• Exclusions: "do not mention pricing," "avoid jargon," "no emoji"
• Honesty: "if you are not sure, say so instead of guessing"

That last one matters. Asking the model to flag uncertainty, and to say when it is missing information, makes answers more trustworthy and tells you what context to add next.

Iterate: the second prompt matters as much as the first

Even a well-built prompt rarely gives the perfect answer on the first try. Treat the first response as a draft and steer it. These follow-ups do most of the work:

• Be specific about what is wrong: "The second paragraph is too formal. Rewrite only that paragraph, keep everything else."
• Ask for options: "Give me 3 more versions of the headline, each with a different angle: curiosity, urgency, specificity."
• Push for depth: "Point 4 is too general. Give a concrete example with numbers."
• Ask for a critique: "Act as a skeptical editor. List the three weakest points in this draft and fix them."
• Lock what works: "Keep this structure and tone for the rest of our conversation."

Ask it to ask you questions

When you are not sure what context matters, let the model interview you first. This is the single most underrated prompting habit.

I want you to help me [goal]. Before you answer, ask me up to 5 questions about anything you need to know to do this well. Ask them all at once, then wait for my answers.

You answer in one message, and the model writes with exactly the context it asked for. It works especially well for plans, resumes, business documents and anything personal.

If you want to see the habits that quietly sabotage your results, read our list of common prompt mistakes.

The master template

Copy this, fill in the brackets, and delete any line you do not need.

You are [role with relevant expertise and point of view].

Task: [one clear action verb + what you want produced].

Context:
- Audience: [who will read or use this]
- Goal: [what success looks like]
- Background: [key facts, what you already tried, relevant details]
- Example or reference: [paste a sample of the style or quality you want, if you have one]

Format:
- [structure: list, email, table, sections, number of items]
- [length]

Constraints:
- Tone: [tone]
- Avoid: [things to leave out]
- If anything is unclear or missing, ask me before answering.

Here it is filled in for a real-feeling task:

You are a customer success lead at a B2B software company.

Task: Draft a reply to a customer who is angry that a promised feature was delayed.

Context:
- Audience: Dana, operations manager at a 60-person logistics firm, customer for 2 years
- Goal: keep the account, be honest about the new date, offer something concrete
- Background: the bulk export feature moved from September to November because of a data migration issue. We can offer a manual export done by our team every Friday until then.

Format:
- Email with a subject line
- Under 180 words

Constraints:
- Tone: accountable and calm, no corporate filler
- Avoid: blaming other teams, the phrase "we apologize for any inconvenience"
- If anything is unclear, ask me before answering.

This template works the same way in Claude, Gemini, Copilot and other assistants, with small differences covered in ChatGPT vs Claude vs Gemini prompting.

A shortcut when you prompt every day

Filling in five parts by hand is worth it for important work, but it gets tedious for the twentieth prompt of the day. If that is you, a prompt generator such as Promptay builds this role, task, context, format and constraints structure from one sentence, tuned for ChatGPT, Gemini or Mistral, and you can open the result directly in ChatGPT, Claude or Gemini. The free plan includes 3 prompt generations a day with no credit card.

Whether you use a tool or not, the skill is the same: notice what the model would have to guess, and tell it instead.

Frequently asked questions

What is the best way to write a ChatGPT prompt?

Give it a role, a clear task with an action verb, the context it cannot guess, the output format you want, and any constraints such as length or tone. You do not need all five for simple questions, but anything you plan to use or publish benefits from each of them. Then refine the first answer with specific follow-up prompts instead of starting over.

How long should a ChatGPT prompt be?

As long as it needs to be to remove guesswork, and no longer. A quick factual question can be one line, while a prompt for a business email or a content plan often runs 80 to 200 words once you include context and format. Length is not the goal; every sentence should answer something the model would otherwise assume.

Does giving ChatGPT a role actually help?

Yes, when the role is specific and relevant to the task. A role like "a hiring manager at a small SaaS company" changes the vocabulary, priorities and level of detail in the answer. Generic roles like "an expert" or "a genius" add little, so describe the role the way you would describe a real person you would ask for help.

How do I get ChatGPT to stop giving generic answers?

Add concrete context about your audience, goal and situation, and paste an example of the style you want. Ask for a specific format and say what to leave out, such as introductions or summaries. If the answer is still generic, tell it exactly which part is too vague and ask for a concrete example with real details.

Should I ask ChatGPT to ask me questions first?

It is one of the most effective habits for complex or personal tasks. Ask it to list up to five questions it needs answered before it starts, answer them in one message, and the result will be built on your real situation. It also shows you which context you tend to forget, which makes your next prompts better.