A good result with Claude is not down to luck: it is down to the way you phrase your request. Most disappointments come from a prompt that is too vague, not from a limit of the tool. Here are the seven techniques that make all the difference, each illustrated with a before/after example you can reuse as-is in your own line of work.
1. Give context and a role
Claude knows neither your company, nor your client, nor your objective. Without those elements, it can only answer in generic terms. The first technique is therefore to set the scene: who is speaking, for whom, and to what end. Add a role to frame the tone and the level of expertise you expect in the answer.
Compare. A poor prompt: "Write an email to follow up with a client." A rich prompt gives an entirely different result, because it knows who it is speaking to and why.
You are the sales manager of a small HR consulting firm.
I need to follow up with a prospect I met at a trade show last week: the director of a 40-employee company, interested in our HR audit offer but silent since.
Goal: get a 20-minute meeting, warm and non-pushy tone.
Write the email.
2. Be explicit about the expected format
If you do not specify the form of the output, Claude chooses it for you. State the structure (headings, bullet list, table), the length (number of words or points) and the tone you want. An explicit format spares you three rounds of back-and-forth to "shorten it" or "turn it into a list".
Instead of "Summarise this meeting report", ask: "Summarise this meeting report in 5 points maximum, one per decision made, phrased in the imperative, with no introductory sentence." The result lands ready to use, with no rework needed.
3. Show examples (few-shot)
An abstract instruction leaves too much room for interpretation. A concrete example is worth more than a long description. That is the principle of few-shot prompting: you show Claude one or two cases of what you expect, and it imitates the model you have given it.
Rather than "write professional but warm replies to customer reviews", show it one:
Reply to customer reviews in this style.
Example review: "Fast delivery but the packaging was damaged."
Example reply: "Thank you for your feedback, and glad the delivery was quick! Sorry about the packaging: we are actually reinforcing our parcels this month. Drop us a DM and we'll make it up to you."
Now reply to this review: "Product as described, but customer service took 3 days to respond."
4. Structure with tags
As soon as a prompt gets longer — a context, several instructions, a document to process, examples — everything blurs together. The solution: separate each part with XML-style tags, such as <context>, <instructions> and <example>. Claude then identifies without ambiguity what is an instruction and what is material to process.
Claude is particularly comfortable with this structure: it was trained to recognise these markers, which makes it reliable on long, complex prompts.
<role>You are a marketing analyst.</role>
<context>
We are launching a subscription offer for project management software aimed at agencies.
</context>
<instructions>
Propose 3 campaign angles. For each one: a hook, the target audience, the recommended channel.
</instructions>
<format>
A table with 3 columns.
</format>
5. Ask for step-by-step reasoning
For analytical tasks — comparing options, making a judgement call, checking a calculation — explicitly ask Claude to reason through the problem before concluding. This is the technique known as chain-of-thought: "Detail your reasoning step by step, then give your recommendation." By laying out the steps, Claude avoids shortcuts and makes its reasoning verifiable by you.
A concrete example: "Compare these two supplier quotes on total cost, lead times and risk. Explain your analysis criterion by criterion, then conclude with a one-sentence recommendation." You get a reasoned decision, not an opinion out of thin air.
6. Iterate and correct
Prompting is not a one-shot exercise: it is a dialogue. Prompts that come out perfect on the very first try are rare. The good practice is to run a first version, read the result, then correct precisely: "It's too long, cut it by half", "The tone is too formal, make it more direct", "Add a numbered example". Each correction moves you closer to the result you want, and you keep the best version of the prompt for next time.
Tip: when a prompt works well, save it. You gradually build a reusable library that saves you considerable time on your recurring tasks.
7. The common mistakes
Three traps come up over and over with beginners. The vague prompt ("help me with my marketing"): Claude cannot guess your situation. The prompt that asks for everything at once (strategy + emails + posts + schedule in a single request): far better to break it into steps. And the prompt with no context, which produces an answer that is correct but unusable because it does not match your reality. Do the exact opposite of these three mistakes and you are already well above average.
| Weak prompt | Effective prompt |
|---|---|
| "Write a LinkedIn post." | "Write a LinkedIn post for an accounting firm, targeting small-business owners, topic: the 2026 year-end close, an educational tone, 120 words, a closing question." |
| "Summarise this text." | "Summarise this text in 4 bullets, one idea per bullet, for a reader in a hurry." |
| "Build me a complete strategy." | "Let's start with the 3 priority objectives for the quarter. Then I will ask you for the action plan for each one." |
| "Reply to this unhappy customer." | "You are the head of customer service. Here is the message from a customer whose order is 5 days late. Reply with: a sincere apology, a concrete solution, a calm tone." |
From technique to mastery
These seven techniques form the foundation of effective prompting. Taken together, they transform your exchanges with Claude: fewer generic results, more outputs you can use straight away. But the real acceleration comes from practising on your real cases — your emails, your meeting reports, your client relationships. That is where a structured method and a library of proven prompts save you weeks of frustrating trial and error.
To go further, discover how Claude helps sales teams day to day, or how to create an AI assistant with Claude for your repetitive tasks. And if you want the complete method applied to your line of work — with more than 50 ready-to-use prompts — the Claude training is built for exactly that.
Do you need XML tags to prompt Claude?
No, they are not mandatory. On a short prompt, a clear instruction is enough. Tags become useful as soon as the prompt gets longer: they separate the context, the instructions and the examples, which helps Claude keep everything straight. Save them for complex prompts.
What is few-shot prompting?
Few-shot prompting means showing one or two examples of what you expect, rather than describing it in the abstract. Instead of writing "use a professional but warm tone", you paste in an example of a message in that tone. Claude imitates the model far more faithfully than it follows a theoretical instruction.
How do you improve quickly at prompting?
By working on your real professional use cases, with a structured method and a library of prompts that work. That is exactly what our Claude training covers: the complete method and more than 50 ready-to-use prompts, applied to your line of work.
Does the same prompt always give the same result?
No, the responses vary slightly from one run to the next. That is normal. The goal is not a magic prompt that wins every single time, but a reliable prompt that produces a good result most of the time, which you then fine-tune over a few exchanges.
Move from theory to mastery
Book a free assessment: we identify your priority use cases and show you the prompting method applied to your line of work. See also our packages.
Book my free assessment