The best ChatGPT prompts share one shape: role, context, task, constraints, and the format you want back. Copy-paste prompt lists fail because they carry none of your context. This guide gives you prompts that work as templates, sorted by job, plus the pattern for writing your own. Every example below is written so you can swap in your own details in under a minute.
Prompting is not magic phrasing. It is briefing. If you would not hand that instruction to a new colleague and expect good work, the model will not manage either.
The Five-Part Prompt Pattern
Use this skeleton for anything that matters.
- Role. “You are a technical recruiter screening backend candidates.”
- Context. What you have, who it is for, what already exists.
- Task. One clear verb. Draft, compare, critique, extract, plan.
- Constraints. Length, tone, what to avoid, what must be included.
- Format. Table, bullets, 200 words, JSON, email with subject line.
Drop any part and quality drops with it. Format is the one people skip most, and it is the cheapest to add.
Why Prompt Quality Changes the Output So Much
A model is a text predictor. It continues what you started, in the style you established. Vague input maps to the most average possible answer, because that is the safest continuation. Specific input narrows the space to something useful.
That is the entire premise of prompt engineering: shaping the input so the likely output is the one you need. The developer documentation for these models makes the same point in practical terms, recommending explicit instructions, examples, and clear delimiters in OpenAI’s prompting guidance.
Two mechanics are worth knowing. A large language model has a limited context window, so very long pasted material can push earlier instructions out of view. And these systems produce hallucinations when asked for facts they do not hold, which is why prompts that supply your own source material beat prompts that ask the model to recall.
Categories of Prompts That Actually Earn Their Place
Writing and editing
Rewrite with constraints. “You are an editor. Rewrite the text between the triple quotes for a busy client. Keep every fact. Cut length by 30 percent. Use sentences under 20 words. Return the edited version only, no commentary. """[paste]"""”
Structure before drafting. “I need to write [document] for [audience] to get [decision]. Give me three possible structures as bullet outlines, each with a one-line rationale. Do not write the document yet.”
Critique my draft. “Act as a sceptical reviewer. List the five weakest claims in this draft, why each is weak, and what evidence would fix it. Do not rewrite anything.”
Business and work
Turn notes into actions. “Here are my meeting notes. Extract a table with columns: owner, action, deadline, dependency. Flag any action with no clear owner. Notes: """[paste]"""”
Pre-mortem a plan. “We plan to [plan] by [date] with [resources]. Assume it failed. List the eight most likely causes, ranked by probability, with one preventive step each.”
Objection handling. “You are a cautious buyer in [industry]. I will describe an offer. Give me your top five objections in your own words, hardest first, then the question behind each objection.”
Learning and research
Explain at my level. “Explain [concept] to someone who understands [what you already know] but not [the gap]. Use one analogy, then one worked example with numbers, then three questions to test whether I understood.”
Quiz me properly. “Quiz me on [topic]. Ask one question at a time, wait for my answer, then tell me what was wrong and ask a harder follow-up. Do not give the answer before I respond.”
Compare sources I supply. “Here are three documents. Build a table of where they agree, where they conflict, and what none of them addresses. Quote the exact line for each conflict.”
Personal organisation
Weekly triage. “Here is my task list and my calendar for the week. Group tasks into: do now, schedule, delegate, drop. Explain each drop in one line. Ask me for missing information before you answer.”
Decision support. “I am deciding between [A] and [B]. Ask me the six questions that would most change the answer, one at a time. Then summarise the trade-off, without recommending.”
Creative work
For teams whose creative output is mostly social content, our prompt patterns built for social media specifically extend the ideas below into platform-specific formats.
Constrained ideation. “Generate 15 ideas for [thing]. Five must be conventional, five unusual, five deliberately impractical. One line each. No explanations.”
Voice matching. “Here are three samples of my writing. Describe my voice in five specific attributes, then draft [new piece] in that voice. Point out where you were unsure.”
What Separates a Good Prompt From a Bad One
| Weak prompt | Why it fails | Stronger version |
|---|---|---|
| “Write a blog post about AI” | No audience, no angle, no length | “Write a 700-word post for small-business owners on choosing an AI tool, with three criteria and one example” |
| “Summarise this” | No purpose, no format | “Summarise for a board pack: five bullets, decisions only, no background” |
| “Make this better” | Undefined better | “Cut 20 percent, keep all figures, plainer verbs, keep the closing paragraph” |
| “Give me ideas” | No constraints, generic output | “Ten ideas, each testable in one week for under two hours of work” |
| “Is this true?” | Asks for recall | “Check this claim against the text I pasted. Quote the line that supports or contradicts it” |
The pattern is visible: named audience, named purpose, explicit limits, defined output.
A Worked Example, Step by Step
Take a common task: turning a customer call into a follow-up email.
Attempt one. “Write a follow-up email about my call.” Result: generic, wrong length, invents details.
Attempt two. Add role and context. “You are an account manager. Write a follow-up email after a call with a warehouse operations lead about stock-count errors.” Better, still generic.
Attempt three. Add constraints, format and source material. “…Use only the notes between the quotes. Under 150 words. Three sections: what we agreed, what I owe you, what you owe me. No adjectives like excited or thrilled. Notes: """[paste]"""” Result: usable with light editing.
Count the time. Attempt one costs 20 seconds and produces a rewrite of five minutes. Attempt three costs 60 seconds and produces a rewrite of one minute. Across 20 emails a week, the longer prompt saves roughly an hour. That is the whole argument for writing prompts properly, and you can run the same sum on any repeated task you own.
Tips for Crafting Your Own Prompts
- Give an example of the output you want. One good example beats three paragraphs of description.
- Use delimiters. Triple quotes or tags around pasted material stop the model confusing content with instructions.
- Ask it to ask. “Ask me any questions you need before answering” removes most guesswork.
- Set what not to do. Banned words, no preamble, no summary at the end.
- Iterate in one thread. Correcting an answer usually beats starting again.
- Save what works. A file of ten reliable prompts is worth more than a bookmarked list of 500.
- Split big jobs. Outline, then draft section by section. Long single prompts drift.
- Give it your material. Grounding in your own text is the single biggest accuracy improvement available.
Common Mistakes to Avoid
- Copying prompt lists verbatim. They contain someone else’s context. Adapt or skip.
- Asking for facts from memory. Ask it to work from text you supply, then verify anything numeric.
- Piling five tasks into one prompt. Output quality falls off a cliff. One task, then the next.
- Accepting the first answer. The second attempt, with one correction, is usually the good one.
- Pasting confidential material without checking settings. Client data, contracts and health details deserve a policy check first.
- Treating fluency as accuracy. Confident tone is free. Verification is not.
- Never saying what you disliked. “Too formal, cut the intro, keep the second paragraph” fixes more than a new prompt.
Real-World Applications Worth Copying
Support teams use extraction prompts to turn ticket threads into structured summaries, which makes handover fast and consistent, a pattern our prompt workflows built for customer support teams covers in more depth.
Analysts paste their own data descriptions and ask for the three charts that would answer the question, then build those charts themselves. Finance teams doing similar work may also want our prompt workflows built for accounting tasks.
Managers run pre-mortems before committing budget, because the model is genuinely good at listing failure modes without office politics.
Job seekers use critique prompts on their own CVs, then rewrite personally. That order keeps the writing theirs, which matters both for honesty and for the interview.
Teachers and trainers generate practice questions from their own material, then check every answer before use.
Product, Course, App and Platform Experience
If you’re still getting oriented, our full beginner’s guide to ChatGPT covers the basics before you start building your own prompt library. Prompting behaves slightly differently across assistants, but the pattern transfers. Longer documents favour tools with bigger context handling. Current events need a tool with live search. Anything involving your files works best where those files already live.
Two habits improve results everywhere. Keep a personal prompt file, because reuse beats reinvention. And keep a rejected pile, because knowing which prompts failed teaches faster than collecting new ones.
The skill itself is the durable asset here. Models change every few months; briefing clearly does not. If you want that built in a structured order rather than assembled from scattered tips, explore Coursiv AI lessons and practise on the tasks already sitting in your week.
Free prompt collections are everywhere, and some are genuinely good. What they cannot give you is sequencing, feedback on your own attempts, a deadline, or anyone checking whether your output actually improved. That gap is why most people bookmark hundreds of prompts and use four.
Decision Framework: Which Prompt Style to Use When
Pick by what you have and what you need.
- Do you have source material? If yes, always paste it and constrain the model to it. Accuracy rises immediately.
- Is the output for someone else? If yes, specify audience and format first. If it is for you, a rough prompt is fine.
- Is it repeated work? If yes, invest once in a template and save it. If it is one-off, do not over-engineer.
- Does it involve facts, numbers or law? If yes, use the model to organise, then verify every specific from a primary source.
- Is it creative? If yes, loosen constraints on content but tighten them on quantity and format.
If three answers point at repeatable work, write the template today. Everything else can stay ad hoc.
Frequently asked questions
What makes a ChatGPT prompt effective?
Should I use long prompts or short ones?
Do prompt libraries actually help?
How do I stop it inventing details?
Pick three tasks you repeat weekly, write one template for each using the five-part pattern, and keep them in a file you actually open. That takes an afternoon and improves every answer you get afterwards.