Why does a prompt technique make a difference?
A prompt technique makes a difference because an AI model can only work with what is in your prompt. A loose question gives you a generic answer. A prompt with structure and context gives you an answer you can use.
The five techniques below work in any AI assistant, including Claude and ChatGPT. Each technique comes with an example prompt you can copy. Replace the text in square brackets with your own details. If you are not yet sure what a prompt is, first read what is a prompt.
1. Be specific about the desired format
State in your prompt what the answer should look like: length, form, tone and audience. The more of that you pin down, the less you have to correct afterwards.
Weak: "Write an email about a payment extension."
Strong: "Write an email of at most 150 words to a customer asking for a payment extension. Show understanding and be clear about the terms: one month's extension, then payment in two instalments."
Decide in advance which form you need. A list, running text, a table and a one-sentence summary each call for a different instruction.
Example prompt "Write [type of text] about [topic] for [audience]. Length: at most [number] words. Form: [running text, numbered list or table with columns X and Y]. Tone: [tone]. Start with the main message and leave out an introduction."
2. Give context
Tell the model who you are, what your goal is and who the output is for. Without that information the model assumes an average sender and an average reader.
Example: "I own a webshop in sustainable clothing. I want to send an email to customers who have not ordered in six months. Tone: warm and without pressure."
Context can also be a document. Paste your style guide, an earlier proposal or your customer's question and refer to it in your instruction.
Example prompt "Background: I am [role] at [type of business] with [number] employees. Our customers are [audience]. My goal with this text is [goal]. The reader already knows [prior knowledge] and does not yet know [what is new]. Task: [task]."
3. Use examples (few-shot)
Give two or more examples of the result you are looking for. The model recognizes the pattern in your examples and applies it to new input. This is called few-shot prompting.
This technique works well for tasks where style or form is fixed, such as product descriptions, social media posts and replies to customer questions. Your own earlier texts are the best examples, because they already carry your tone.
Example prompt "Below are three product descriptions from our webshop. Write a description for this new product in the same style and structure: [product name and features]. Example 1: [text]. Example 2: [text]. Example 3: [text]."
Choose your examples with care. The model also copies the weak points, so pick texts you are satisfied with yourself.
4. Have the AI reason step by step (chain-of-thought)
Ask the model to write out the intermediate steps first and only then give a conclusion. This is called chain-of-thought. It helps with questions that involve several steps, such as a trade-off, a schedule or a calculation.
The benefit for you: you see how the answer came about. If there is an error in an assumption or a calculation step, you will find it in the worked-out reasoning.
Example prompt "I am considering [decision], for example hiring a second employee for customer service. My situation: [figures and circumstances]. Work this out step by step. Step 1: list the assumptions. Step 2: calculate the costs and returns. Step 3: name the risks. Step 4: give your advice. Say so if you are missing information and do not fill in figures yourself."
Written-out reasoning makes errors visible. Always recalculate the figures yourself afterwards.
5. Iterate and refine
Treat your first prompt as a starting point and refine with concrete pointers. Read the answer, name what needs to improve and ask for a new version. The conversation remembers what you said earlier.
Concrete pointers work better than general ones. "Make it better" gives the model little to go on. "Delete the first paragraph and replace the second example with one from construction" gives you the version you are looking for.
Example prompt "This is a good start. Change three things. One: make the text a third shorter. Two: replace jargon with plain words. Three: add an example from [industry] after the second paragraph. Leave the rest unchanged."
If a prompt works well after a few rounds, work your pointers into the original prompt and save it. That way you get the right result straight away next time.
How do you combine the five techniques in one prompt?
You combine the techniques by putting them in a fixed order in one prompt: context, task, format, example and method. Refining happens afterwards in the conversation.
Example prompt with all techniques "Context: I own a painting company with eight employees and work for private customers. Task: write a follow-up email to a customer who received a quote two weeks ago and has not yet replied. Format: at most 100 words, running text, friendly and professional. Example of my writing style: [paste an earlier email]. Method: first think of two reasons the customer may have for waiting, then include one reassuring sentence per reason. Return only the email."
Save prompts that work well in a document with a short title per task. After a few weeks you have your own library for the work that keeps coming back.
When is a prompt not enough?
A prompt is not enough when you have to type the same context every day or when a task requires access to your systems. In that case you capture your instructions in a system prompt or have an AI agent do the work. Read the difference in AI agent vs chatbot.
If you are still at the beginning, start with AI for beginners and practise techniques 1 and 2 first. Those two bring the biggest improvement.
Frequently asked questions
What is the most important prompt technique? Giving context. An AI model knows nothing about your business or your goal until you tell it. Two sentences of background and a clear audience improve the answer more than any other technique.
What is few-shot prompting? Few-shot prompting is a technique where you put two or more examples of the desired result in your prompt. The model recognizes the pattern and applies the style and structure to your new task.
What is chain-of-thought prompting? Chain-of-thought prompting is a technique where you ask the model to write out the intermediate steps first and then give the conclusion. You see the reasoning and can check assumptions and calculation steps.
Do these prompt techniques work in ChatGPT and Claude? Yes. Stating the format, giving context, adding examples, asking for step-by-step reasoning and refining work in any AI assistant that works with language, including ChatGPT and Claude.
How do I save good prompts? Put prompts that work well in one document, with a title per prompt and the parts you fill in between square brackets. Update the prompt when you notice you keep giving the same pointer afterwards.
Want to keep learning?
If you want to apply these techniques to your own work with guidance, see the 1-on-1 AI training. If you want to keep sparring about prompts and tooling, read about AI coaching for entrepreneurs for €399 per month excl. VAT.
