A good prompt is one that clearly tells an AI tool what you want, why you want it and in what form you expect the result. The prompt writing guides that Anthropic and Google have published for their own models agree on the same points: reduce ambiguity, give context, show examples when needed and improve the prompt based on the first response. The suggestions below are based on these guides as of October 2026 and work with all chat assistants such as ChatGPT, Claude or Gemini.
What should a good prompt include?
Before you write your prompt, think about these five parts. Not every prompt needs all of them, but the more of them are in place, the more accurate the response will be:
- Task: What do you want done? Summarise, write, compare, list, correct.
- Context: Who is the text for and what will it be used for? What details does the model not know?
- Constraints: Length, language, tone, topics to avoid.
- Output format: A paragraph, a bulleted list, an email or a table?
- Example: One or a few examples similar to the result you expect.
Anthropic's guide suggests a practical test: show your prompt to a colleague who knows nothing about the subject. If that person struggles to understand what to do, the AI will struggle too.
How do you write a prompt step by step?
- Write what you want in one sentence: "Write a short note telling the neighbours about the lift maintenance."
- Add context: "It will be written on behalf of the building management, and most of the readers are elderly."
- State the constraints: "Use simple, polite language and keep it to no more than five sentences."
- Say the format: "Leave space for a heading and for the date and time."
- If needed, give an example or paste in an old note you liked.
- Read the response and say what is missing in the same chat: "Make it shorter and remove the second sentence."
A prompt written in this order gives a far more useful result than a one-word request such as "write an announcement".
Why does giving examples work?
An example is the shortest way to explain the format and tone you expect. Google's guide explains that prompts with examples are often used to shape the format, wording and scope of a response, but warns that if too many examples are given, the model may produce responses that resemble the examples too closely. Anthropic also recommends that examples should be close to your real need, varied and kept separate from the instructions.
In practice, watch out for this: if you give more than one example, write them all in the same layout. If one is a bulleted list and another is a paragraph, the model cannot tell which one to follow.
What should you write instead of saying "don't"?
Saying what you want gives better results than only setting prohibitions. In Anthropic's example, instead of saying "Do not use Markdown formatting in your response", it recommends saying "Write your response in flowing paragraphs". The same guide says that explaining the reason for an instruction also helps the model understand the goal. For example, when you say "This text will be read on a phone screen, so keep it short" instead of "Write briefly", the model understands your goal and applies it to similar situations too.
What should you watch out for when working with a long text?
If you are going to paste in a report, a contract or lecture notes and ask a question:
- Put the long text at the start of the prompt and your question and instructions at the end. Anthropic states that with long documents this order improves response quality.
- If there is more than one document, separate each with a heading and write its name.
- Ask it to quote the relevant sections first and then comment; that way you can check that the response is based on the text.
- As in Google's guide, you can limit the model from going beyond the text by saying "Answer only on the basis of the text I have given".
What should you do when you do not like the result?
Writing prompts is a matter of trial and correction. Google's guide suggests these approaches:
- Rewrite the same request in different words.
- Switch the task to a similar but easier one; for example, turn an open-ended question into a multiple-choice one.
- Change the order of the parts in the prompt.
- Break a complex job into steps; make the output of the first step the input of the second.
Anthropic suggests a similar chain: first have it write a draft, then ask it to assess the draft against criteria you set, and finally have it revise the draft based on that assessment.
Which mistakes should you avoid?
- Assuming the model knows your situation; you need to write every detail it does not know.
- Cramming many unrelated jobs into a single prompt.
- Writing identity numbers, passwords or other people's personal information into the prompt.
- Using the response without checking it; even a good prompt does not completely prevent incorrect information.
For more guides on using AI tools in everyday life, take a look at the AI in Everyday Life section.
Frequently asked questions
What does prompt mean?
A prompt is the instruction or request you write to an AI model. It can be a question, a task description or a text you want completed; the model's response depends largely on how clear this prompt is.
What should a good prompt be like?
It should be clear and specific: you should write what you want, who it is for and why, constraints such as length and tone, and the format of the result. If needed, also add an example similar to the result you expect.
Do you need to add examples to a prompt?
Not always, but they help a great deal when format and tone matter. Google's and Anthropic's guides recommend that examples be close to the real need and written in the same layout; too many examples, however, can make the response resemble the examples too closely.
Should a prompt be written in Turkish or English?
Writing in the language you want the response in is usually the easiest way. To be sure, you can add a clear instruction such as "Reply in Turkish" at the end of the prompt.
What should I do if the AI does not give me the response I want?
Rewrite the same request in different words, point out what is missing in the same chat, change the order of the parts in the prompt or break the job into small steps. Google's guide recommends these ways of trying and correcting.
How do you ask a question about a long document?
Put the document at the start of the prompt and your question and instructions at the end. Ask it to quote the relevant sections first and then answer; that way you can check whether the response is based on the document.
Editorial desk
Stories signed by the Cultvia editorial desk follow our editorial policy; every story lists its sources at the end and errors are fixed under our corrections policy.
Last updated:
What changed?
- : First published
Your follow list is kept only on this device; when the page is updated you'll see it in Following.
Get email updates
Sources
- Prompting best practices — Anthropic
- Prompt design strategies — Google AI for Developers

