In short: A practical guide to writing clearer prompts for AI chatbots, with before-and-after examples covering context, role, format and iteration, plus a short checklist. Last reviewed: 22 November 2026. Written by the AI Decoded desk.

A prompt is simply the instruction you give an AI tool such as a chatbot. The words you choose make a real difference to what comes back. You do not need special tricks or technical knowledge to get better answers. Most of the gains come from saying clearly what you want, giving the background the AI needs, describing the shape of the answer, and refining as you go. This guide draws on the published advice of two of the biggest AI developers, OpenAI and Anthropic, and turns it into plain-English habits with examples.

Why clear prompts matter

A large language model cannot read your mind. It only has the words in front of it. Anthropic suggests thinking of the AI as "a brilliant but new employee who lacks context on your norms and workflows". The more precisely you explain what you want, the better the result.

Anthropic also offers a simple test: show your prompt to a colleague who knows little about the task and ask them to follow it. If they would be confused, the AI probably will be too.

Be specific about what you want

Vague requests get generic answers. OpenAI's prompting guidance recommends being "specific, descriptive and as detailed as possible about the desired context, outcome, length, format, style, etc".

Before: "Write a poem about the seaside."

After: "Write a short, cheerful poem about a family day at the seaside in late summer, in four rhyming verses, suitable for reading aloud to young children."

The second version tells the AI the mood, the subject, the length, the structure and the audience. Each detail removes a guess.

It also helps to swap woolly wording for exact numbers. OpenAI's own example replaces "fairly short, a few sentences only, and not too much more" with "Use a 3 to 5 sentence paragraph".

Finally, if you want the AI to go further than the bare minimum, ask for it directly. Anthropic notes that its models respond well to explicit requests rather than being left to infer what you want from a vague prompt.

Give context: explain the why

Context is the background the AI cannot know unless you tell it: who the answer is for, what it will be used for, and why a rule matters. Anthropic says explaining the reason behind an instruction gives the model more information about your goals and helps it deliver more targeted responses.

Its example is a formatting rule. "NEVER use ellipses" is less effective than explaining that the response will be read aloud by a text-to-speech engine, which will not know how to pronounce them. With the reason, the AI can apply the same thinking to other situations it was not told about.

Before: "Summarise this report."

After: "Summarise this report for our board, who have five minutes before a meeting and are not technical. Focus on costs and risks."

When you paste in text for the AI to work on, OpenAI suggests putting your instruction at the beginning and separating it from the text with markers such as ### or triple quotation marks ("""), so the AI can tell the instruction from the material.

Give the AI a role

A role is a short description of who the AI should act as. Anthropic says setting a role focuses the model's behaviour and tone, and that "even a single sentence makes a difference". Its example is: "You are a helpful coding assistant specializing in Python."

Before: "How should I reply to this complaint?"

After: "You are an experienced customer service manager at a small online shop. Suggest a polite, helpful reply to this complaint."

Developers often place the role in a system prompt, a set of standing instructions the AI reads before each conversation. In everyday chat tools, you can simply put it at the start of your message.

Describe the format you want

If you need the answer in a particular shape, say so, or better still, show it. OpenAI says models "respond better when shown specific format requirements". Its example asks for company names, people's names, topics and themes, then lays out the exact headings the answer should use.

Before: "Pull out the key details from this email."

After: "Pull out the key details from this email in this format: Sender: ... Deadline: ... Action needed: ..."

Both developers also advise telling the AI what to do rather than only what not to do. Anthropic gives the example of replacing "Do not use markdown in your response" with "Your response should be composed of smoothly flowing prose paragraphs." OpenAI makes the same point with a customer support example: rather than shouting "DO NOT ASK USERNAME OR PASSWORD", it describes what the agent should do instead.

Anthropic adds that the style of your prompt can influence the style of the reply, so a prompt full of bullet points may get bullet points back. When order matters, it suggests giving instructions as numbered steps.

Show examples

Examples are one of the most reliable ways to steer an AI's format, tone and structure, according to Anthropic. Giving a few is known as few-shot prompting. Anthropic suggests three to five examples, and says they should mirror your real use, vary enough to cover unusual cases, and be clearly separated from your instructions.

OpenAI recommends starting with a plain instruction and no examples (zero-shot), then adding a couple of examples if that does not work. Only if neither works does it suggest fine-tuning, which means further training a model on your own data, a step aimed at developers.

Iterate: treat the first answer as a draft

Good prompts are rarely perfect first time. Treat the first reply as a draft, see what is missing or wrong, and adjust. Each approach above gives you something to change: add context, tighten the format, give an example or set a role.

You can also ask the AI to review its own work. Anthropic describes a common pattern of generating a draft, having the model review it against criteria, then having it refine the draft based on that review. It also suggests ending a prompt with something like "Before you finish, verify your answer against [your criteria]."

A good prompt improves your chances but does not guarantee accuracy. AI tools can still produce confident mistakes, known as hallucinations, so check anything important before you rely on it.

A quick prompt checklist

The bottom line

Better prompting is less about clever phrasing and more about clear communication. Say what you want, explain why, describe the shape of the answer, and refine as you go. The same habits that help a new colleague do good work help an AI tool too.

Sources

Follow AI Decoded on Substack

Weekly plain-English AI news and verified UK events.