How to Write AI Prompts That Get Useful Answers: A Practical Guide
Two people ask the same AI assistant for help with the same job. One gets a generic page of advice and gives up. The other gets a draft they can use with ten minutes of editing. The model is identical. The difference is in the forty or so words each of them typed.
A good AI prompt tells the model five things: the task, the context it needs, who the answer is for, the format you want back, and any limits. Write it the way you would brief a capable new colleague who knows nothing about your situation. Then treat the first answer as a draft and refine it with follow-up instructions. Clear, specific prompts beat clever tricks, and they work across ChatGPT, Claude, Gemini and Copilot.
Why the prompt matters more than the tool
An AI assistant has no idea who you are, what you already know, what you tried yesterday or what a good result looks like to you. It only has the text in front of it. When that text is short and vague, the model fills the gaps with the most average, most widely applicable answer it can produce. That is why "write a blog post about email marketing" returns something that reads like every other article on the subject.
Give the same model a clear brief and it has far less to guess. The answer becomes specific because your request was specific.
This is good news for beginners. You do not need special vocabulary, secret phrases or a paid course. You need the habit of saying what you mean in enough detail, which is the same skill as writing a good email to a freelancer.
- The model sees only your prompt and the current conversation. It does not know your goals unless you state them.
- Longer is not better. Relevant detail helps; padding does not.
- Prompting is iterative. A second, corrective message is normal and often does more than a perfect first prompt.
- AI output can be confidently wrong. Checking facts stays your job.
The five parts of a strong prompt
Almost every prompt that works contains some version of these five parts. You will not need all five every time, but when an answer disappoints, one of them is usually missing.
What you want done
One clear verb and object: summarise this report, draft a reply, compare these two options, find the error in this formula.
The background
Your situation, the source material, what has been tried, what matters. This is the part people most often leave out.
Who it is for
A customer, a manager, a ten-year-old, a developer. The reader decides the vocabulary and the depth.
The shape of the answer
Bullet list, table, email, numbered steps, word limit. If you do not say, you get whatever the model defaults to.
What to avoid or include
Tone, things not to mention, sources to use, "use only the text below", "say if you are unsure".
A sixth element helps for some tasks: a role. Asking the model to act as an experienced copy editor or a patient maths tutor sets tone and level in a few words. It is useful, but it is no substitute for context. "You are a world-class marketer" followed by a vague request still produces a vague answer.
How to write a prompt, step by step
- Decide what "done" looks like. Before typing, finish this sentence: "I will be happy if I get ___." If you cannot, the model cannot either.
- State the task in one sentence. Lead with it. Put the instruction where it cannot be missed.
- Add the context a stranger would need. Who you are, what this is for, the relevant facts and figures. Paste source material if you have it.
- Name the reader. "For customers who are not technical" changes the whole answer.
- Specify the format and length. "A table with three columns", "under 150 words", "five bullet points".
- Add limits. Tone, exclusions, and how to handle uncertainty.
- Separate instructions from material. Put pasted text under a clear label or inside quotation marks or tags, so the model can tell your instructions from the content it should work on.
- Read the answer and reply to it. "Shorter", "more formal", "you missed the deadline in point 2". Refining is faster than starting again.
If you are not sure what context the model needs, ask it: "Before you answer, ask me up to three questions that would help you do this well." The prompt builder below has a switch that adds this line for you.
Free prompt builder
Fill in what you know. The builder assembles a structured prompt, tells you which parts are missing and gives you text you can paste into any AI assistant.
Prompt Builder
Free mini tool
A template tool that runs in your browser. It arranges what you type into a structured prompt; it is not an AI model and nothing is sent anywhere. Only the task is required.
Your prompt
Before and after: four real examples
The pattern is the same each time: the weak version leaves the model guessing, and the strong version removes the guesses.
| Job | Weak prompt | Stronger prompt |
|---|---|---|
| Write an email to a client about a delay | Draft a short email to a long-standing client explaining that their website launch moves from 3 March to 17 March because the payment provider has not approved the account. Apologise once, give the new date, say what we are doing meanwhile. Friendly, direct, under 120 words. | |
| Summary | Summarise this | Summarise the report below for a manager who has two minutes. Give five bullet points: the main finding, two risks, the cost and the decision needed. Use only information in the text. |
| Learning | Explain DNS | Explain how DNS works to someone setting up their first website. Use one everyday comparison, then the four steps that happen when a visitor types a domain. No jargon without a definition. Around 200 words. |
| Spreadsheet | Excel formula for totals | In Excel, column A has dates and column C has amounts. Write a formula that totals column C for the current month only. Explain each part in one line, and tell me what changes for Google Sheets. |
Notice what the stronger prompts do not contain: no "please act as the world's best", no capital letters, no threats or promised tips. They contain facts, a reader and a shape.
Six techniques worth knowing
Once the basics are in place, a handful of techniques cover most harder tasks. All of them are described in the model makers' own documentation.
| Technique | What you do | When it helps |
|---|---|---|
| Give examples | Show one to three samples of the output you want | Matching a style, a format or a labelling scheme |
| Ask for steps | Ask the model to work through the problem before giving the answer | Calculations, logic, planning, comparisons |
| Break it up | Split a big job into a chain of smaller prompts | Long documents, research, multi-stage writing |
| Supply the source | Paste the text and say "use only this" | Summaries, rewriting, anything where accuracy matters |
| Set a role | Name the expertise and the stance you want | Feedback, tutoring, tone control |
| Ask for a critique | "What is weak about this draft? Then fix it." | Improving your own or the model's work |
Examples beat adjectives
Telling a model to write in a "friendly, professional, concise" tone is open to interpretation. Pasting two paragraphs you wrote and saying "match this voice" is not. If the format matters, show a filled-in example of it.
Break large jobs into stages
Asking for a finished 2,000-word article in one prompt gives you the average of everything. Asking for an outline, correcting it, then drafting one section at a time gives you control at each stage. Our AI content workflow shows this staged approach end to end.
Keep the conversation tidy
Models work within a limited context window, and very long threads can drift or lose early instructions. For a new task, start a new chat and paste a short brief. Our guide to AI tokens and context windows explains why.
What good prompting can and cannot do
Advantages
- More specific, usable first drafts
- Fewer rounds of correction
- Consistent results you can repeat
- Works with any general-purpose assistant
Limits
- Does not make the model know facts it lacks
- Cannot guarantee accuracy
- Takes a minute longer to write
- Results still vary between models and versions
Who benefits most: people who use an assistant for writing, summarising, planning, learning or spreadsheets several times a week. Who should be cautious: anyone relying on AI for legal, medical or financial decisions. A well-written prompt improves the form of an answer, not the reliability of the facts in it.
Common mistakes
- One-line requests for complex jobs. The less you say, the more average the answer.
- Burying the instruction. If the task appears in the middle of three pasted paragraphs, it may be treated as part of the text.
- Asking for several unrelated things at once. Quality drops on each. Send separate prompts.
- Only saying what you do not want. "Don't be boring" gives no direction. Say what you do want.
- Trusting the first answer. Names, numbers, quotes and links can be invented. See how to fact-check AI content.
- Starting over when a follow-up would do. "Keep everything but make the opening shorter" is quicker than a new prompt.
- Pasting private data. Do not put passwords, customer records or confidential documents into a tool unless you know how that provider handles them.
- Copying "magic" prompts from social media. Long templates full of dramatic instructions rarely beat a plain, specific brief written for your own task.
Check your AI provider's data settings before pasting anything sensitive. Policies differ between free, paid and business plans, and they change. When in doubt, remove names and identifying details first.
Advanced tips
- Save your best prompts. Keep a simple document of prompts that worked, with blanks for the parts that change. That is a prompt library, and it is worth more than any downloaded pack.
- Use custom instructions. Most assistants let you store standing preferences, such as your role, your spelling convention and your preferred length, so you do not repeat them.
- Ask for options. "Give me three versions with different openings" is a cheap way to find a direction.
- Ask what is missing. "What would make this answer more reliable?" often surfaces the context you forgot.
- State how to handle uncertainty. "If the text does not say, write 'not stated'" reduces invented detail.
- Write prompts for briefs, too. A good prompt and a good content brief contain the same things. Our content brief template pairs well with this guide.
Prompt checklist
FAQ
A prompt is the text you give an AI model to tell it what you want. It can be a question, an instruction, pasted material, or all three together.
It is the practice of writing and refining prompts so that a model produces more useful, reliable output. For everyday users it mostly means being clear about the task, the context, the audience and the format.
As long as it needs to be to remove guesswork. A simple question can be one line. A drafting task often needs a short paragraph plus the source material. Length is fine when every sentence is relevant.
The principles carry across all general-purpose assistants. Results differ in style and detail between models and versions, so expect to adjust the wording slightly.
A role can set tone and level quickly, and it is useful for tutoring or feedback. It does not add knowledge the model lacks, and it works best alongside clear context and a clear task.
Language models include an element of randomness when choosing words, so repeated runs vary. A more specific prompt narrows the range of answers.
It reduces the risk, especially if you supply the source text and tell the model to say when something is not stated. It does not remove it. Verify anything important.
No. The model makers publish free prompting guides, linked below, and the fundamentals fit on one page. Practice on your own tasks teaches more than generic templates.
Final recommendation
Learning how to write AI prompts is mostly learning to brief clearly. State the task, give the context a stranger would need, name the reader, ask for a format and set limits. Send it, read the answer critically and refine. Use the prompt builder above until the five parts become a habit, then save the prompts that work for you. That small routine will improve your results more than switching tools.
Official references
- OpenAI: prompt engineering guide
- Anthropic: prompt engineering overview
- Google AI for Developers: prompt design strategies
- Microsoft Learn: Microsoft 365 Copilot overview
Related articles
AIHow to fact-check AI content
AIAI tokens and context windows explained
Next step
Build one prompt above for a task on your list today, and save it. If the task is turning an article into social posts, our content repurposing guide has a kit for that.
