Practical techniques for getting better results from AI chatbots: giving context and a clear task, specifying format, using examples, breaking work into steps, iterating, and checking outputs, with before-and-after examples.
The quality of what you get from an AI chatbot depends heavily on what you ask. You don't need special tricks, just clear communication, much as you'd brief a capable colleague who knows nothing about your situation. These techniques work across most tools.
1. Be specific about the task
Vague prompts get generic answers. Say exactly what you want.
- Weak: "Write about our product."
- Better: "Write a 150-word product description for our reusable water bottle for an online shop. Highlight that it keeps drinks cold for 24 hours and is dishwasher safe."
2. Give context
The model doesn't know your audience, purpose or constraints unless you say. Useful context includes:
- Who it's for: "for first-time buyers with no finance background".
- Why: "to persuade our manager to approve the budget".
- Background: paste relevant notes, data or documents, if it's appropriate to share them.
- Tone: "friendly but professional, in British English".
Some people assign a role, such as "You are an experienced copy editor". This can help set tone and focus, but clear context and instructions usually matter more.
3. Specify the format
Say how you want the answer: a table, bullet points, a short email, numbered steps, a maximum length, or headings. "Give me three options in a table with columns for cost, pros and cons" is far more useful than "What are my options?"
4. Show examples
If you want a particular style, include one or two examples. This is sometimes called "few-shot" prompting. For instance, paste two product descriptions you like and ask for a third in the same style. Examples often work better than describing a style in words.
5. Break complex tasks into steps
For bigger jobs, either list the steps in your prompt or work through them in separate messages: first an outline, then each section, then a review. Asking the model to think through a problem step by step before answering can also improve results on reasoning tasks.
6. Separate instructions from material
When you paste a long document, mark it clearly, for example with headings like "Instructions" and "Text to summarise", or by putting the text between quotation marks or tags. This reduces confusion about what's an instruction and what's content.
7. Iterate
Treat the first answer as a draft. Follow up with specific feedback: "Make it shorter", "Use simpler language", "The second point is wrong because...", or "Give me five more alternatives". It's often quicker to refine than to rewrite the prompt from scratch.
8. Ask it to check itself, then check it yourself
You can ask the model to list its assumptions, flag uncertainty, or review its answer against your requirements. But AI tools can still produce confident errors, including made-up facts and references. Always verify anything important against reliable sources, especially figures, legal, financial or medical information.
A before-and-after example
Before: "Help with my CV."
After: "I'm applying for a junior data analyst role. Below is my CV and the job advert. Suggest five specific improvements to make my CV match the advert better, explaining each in one sentence. Keep British English spelling. Don't invent experience I don't have."
Protect your data
Don't paste confidential, personal or sensitive information into an AI tool unless you know how the provider handles it and your organisation allows it. Many workplaces have rules on which tools staff can use.
Quick checklist
- Is the task clear and specific?
- Have I given the context it needs?
- Have I said what format and length I want?
- Would an example help?
- Will I check the result?
Sources and further reading
- OpenAI: Prompt engineering guide
- Anthropic: Prompt engineering overview
- NCSC: ChatGPT and large language models, what's the risk?
Facts checked on 20 October 2026. Spotted an error? Tell us and we will correct it.
Frequently asked questions
What should you know about 1. Be specific about the task?
Vague prompts get generic answers. Say exactly what you want.
What should you know about 2. Give context?
The model doesn't know your audience, purpose or constraints unless you say.
What should you know about 3. Specify the format?
Say how you want the answer: a table, bullet points, a short email, numbered steps, a maximum length, or headings.
What should you know about 4. Show examples?
If you want a particular style, include one or two examples. This is sometimes called "few-shot" prompting.