How to Get Better Results from Claude: A Practical Prompting Guide

Claude can write an email, analyse a report, research a competitor, build a plan, or help you think through a business decision.

But there is a catch: Claude cannot read your mind.

If you ask:

Write a marketing email.

Claude has to make a remarkable number of decisions for you. Who is the email for? What are you selling? What does the audience care about? How long should it be? Should it sound polished, playful, direct, urgent, reassuring? What should the reader do after reading it?

Claude will fill in those blanks. The problem is that its guesses may have very little to do with your business.

Better prompting is not about learning secret commands or finding a magical combination of words. It is about reducing the amount Claude has to guess.

For most everyday business tasks, that comes down to three things:

Instructions. Context. Examples.

Get those right, and you can dramatically improve the quality of what comes back.

The Three Ingredients of a Strong Prompt

A useful prompt answers three basic questions:

What do you want Claude to do?

What does Claude need to know to do it well?

What does a good result look like?

Those correspond to instructions, context, and examples.

You will not need all three for every request. If you ask Claude to convert 25 kilometres to miles, you do not need to explain your business model first.

But the more subjective, creative, strategic, or business-specific the task becomes, the more these three ingredients matter.

1. Instructions: Tell Claude What You Actually Want

Start with the task.

This sounds obvious, but vague instructions are responsible for a lot of disappointing AI output.

Compare:

Write a launch email for our new project management tool.

with:

Write a launch email for freelance designers who currently track projects in spreadsheets. Keep it under 200 words and use a direct, practical tone rather than sounding overly promotional. Lead with the problem of missed deadlines caused by scattered project tracking. End with one call to action: start a free trial.

The second prompt gives Claude far fewer decisions to make on your behalf.

A useful instruction can specify things such as:

• the task
• the intended audience
• the format
• the length
• the tone
• the goal
• important constraints
• the action you want the reader to take

You do not need to specify every one of these every time. Include the ones that affect what a good answer looks like.

Anthropic’s own prompting guidance recommends being clear and explicit about the desired output rather than expecting Claude to infer what you want. Claude Platform Docs

Be specific about the result, not just the activity

There is a subtle difference between telling Claude what to work on and telling it what you want back.

For example:

Analyse these survey responses.

tells Claude the activity.

But:

Analyse these survey responses and identify the five complaints that appear most often. Group similar complaints together, estimate how frequently each theme appears, and give me three issues worth investigating first.

tells Claude what a useful result looks like.

That distinction becomes increasingly important as the task gets more complicated.

2. Context: Give Claude What It Cannot Know

Instructions tell Claude what to do. Context tells it what it needs to know about your situation.

Suppose you ask:

How do I make more money with my business?

There is no useful universal answer to that question. Claude does not know whether you run a restaurant, an accounting firm, an ecommerce store, or a software company. It does not know whether you have 200 employees or work alone.

Now compare it with:

I run a two-person bookkeeping firm that currently gets most clients through referrals. I want to develop one repeatable marketing channel without hiring another employee. Suggest three options suited to a small service business with limited time, and explain the trade-offs of each.

The question has not become complicated. It has become grounded.

Useful context might include:

• what your business does
• who your customers are
• what you have already tried
• what has already been decided
• your budget
• your deadline
• your team size
• your constraints
• information from an earlier stage of the project

The trick is not to give Claude everything.

Give it the information that changes the answer.

Anthropic describes a similar principle in its guidance on context engineering: effective context is not simply about providing more information, but about supplying high-signal information that helps produce the desired result. Anthropic

A simple test is:

If I removed this piece of information, could it reasonably change Claude’s answer?

If yes, it is probably useful context.

If not, you may not need it.

3. Examples: Show Claude What “Good” Looks Like

Some things are surprisingly difficult to describe.

“Professional.”

“Friendly.”

“Punchy.”

“Premium.”

“Not too salesy.”

Two people can interpret each of those completely differently.

Claude has the same problem.

Instead of spending a paragraph explaining the style you want, sometimes the fastest approach is simply to show it.

For example:

Make this professional but approachable.

is open to interpretation.

Compare that with:

Match this tone: “Quick heads up — your renewal is coming up next week, so I wanted to make sure you have everything you need.”

Now Claude has something concrete to work from.

Examples are particularly useful for:

• tone of voice
• email structure
• headlines
• product descriptions
• reports
• formatting
• recurring content
• brand copy

Anthropic describes examples as one of the most reliable ways to steer Claude’s format, tone, and structure. For more complex or repeatable prompting, its guidance recommends using several relevant and varied examples rather than relying on abstract descriptions alone. Claude Platform Docs

You do not need five examples every time you ask Claude to write an email. But when consistency matters, examples can do something adjectives often cannot: show the target.

Put the Three Together

Here is a deliberately weak prompt:

Summarise this contract.

Claude can do it. But it has to decide what matters.

Now give it instructions and context:

Summarise this vendor contract for our operations manager, who needs to prepare for Friday’s budget review. Focus on the renewal date, cost changes, cancellation terms, and any new obligations compared with last year’s agreement. Use bullet points, keep it under 150 words, and flag anything that may need closer review.

Same contract. Same AI.

Very different task.

The improvement does not come from clever prompt engineering. It comes from telling Claude what the summary is for.

Stop Trying to Write the Perfect Prompt

This is where prompting advice often becomes more complicated than it needs to be.

You do not need to spend ten minutes crafting a masterpiece before asking Claude anything.

Start with a reasonably clear request. Look at what comes back. Then correct it.

If Claude gives you something too formal:

Make this warmer and more conversational. Keep the information unchanged.

If it is too long:

Cut this by about 40%. Keep the examples and main argument.

If the answer feels generic:

This is too broad. Focus specifically on a two-person service business with no dedicated marketing team.

If you like one part but not another:

Keep the opening and first two sections. Rewrite the final section so it is more practical and less repetitive.

That conversation is part of prompting too.

You are not failing because the first response was not perfect. You are giving Claude additional information about what “good” means.

Six Prompting Mistakes That Make Claude Less Useful

Mistake	What happens	Better approach
Being vague about the task	Claude has to decide what you meant	State the result you want
Assuming Claude knows your situation	You get generic advice	Provide the relevant business context
Asking for “professional” or “engaging” copy	Claude interprets subjective words for you	Show an example where possible
Giving lots of irrelevant background	The important information becomes harder to identify	Include context that materially affects the answer
Bundling several unrelated jobs together	The response may become shallow or disorganised	Separate unrelated tasks
Treating the first answer as final	You discard useful work and start again	Tell Claude specifically what to improve

For Bigger Prompts, Structure Helps

You do not need special syntax for ordinary Claude conversations. Natural language works perfectly well.

But once a prompt contains instructions, background information, examples, documents, and several requirements, structure becomes useful.

You might write:

Task: Compare these three proposals and recommend what we should investigate further.

Context: We are a five-person consultancy. Price matters, but implementation time is our biggest constraint.

Criteria: Cost, setup time, reporting, support, and contract flexibility.

Output: Give me a comparison table followed by the three biggest trade-offs.

Important: If the documents do not contain enough information to compare something, say so rather than assuming.

Nothing there is technically sophisticated. It is simply easier for Claude to understand.

For more complex prompts, Anthropic also recommends clear structure and supports XML tags to separate instructions, context, examples, and other inputs. Claude Platform Docs

For most business owners, though, headings and clearly separated sections are more than enough to start.

Tell Claude What to Do, Not Only What to Avoid

There is another small change that can improve results.

Compare:

Don’t make this long. Don’t make it corporate. Don’t use lots of bullet points.

with:

Keep this concise and conversational. Write mainly in natural paragraphs and use a short list only where it makes the information easier to scan.

The second prompt gives Claude a positive target.

This is particularly useful when you find yourself writing a long list of things you hate about AI-generated content.

Instead of only saying what the response should not look like, describe what should replace it.

Anthropic specifically recommends this approach for controlling output and formatting. Claude Platform Docs

Use a Role When It Changes the Perspective

You will often see prompting guides telling you to begin every request with:

You are an expert…

You do not need to do that automatically.

A role is useful when the perspective itself changes the answer.

For example:

Review this budget as a sceptical CFO. Identify assumptions that look unusually optimistic and questions you would want answered before approving it.

Or:

Review this landing page as a first-time visitor who has never heard of the company. Tell me what is unclear before suggesting any rewrites.

Those roles give Claude a lens through which to approach the task.

“Act as a world-class expert copywriter” is much less useful if you have not explained the audience, offer, objective, or constraints.

When the Task Is Complex, Ask for Better Analysis — Not Hidden Reasoning

For difficult decisions, analysis, or problem-solving, you can ask Claude to examine the problem carefully before giving you its recommendation.

For example:

Compare these options carefully against the criteria above. Identify important assumptions and trade-offs before giving me your recommendation.

Or:

Before finalising the answer, check it against the requirements I gave you and flag anything that remains uncertain.

That is generally more useful than asking Claude to reveal a long internal chain of thought.

Anthropic’s current guidance also notes that its newer models have stronger reasoning capabilities and recommends giving them clear goals and appropriate thinking effort rather than over-prescribing every reasoning step. Claude Platform Docs

A Prompt Formula You Can Reuse

When you are staring at an empty Claude box and do not know what to write, use this:

I want you to: [task]

The situation: [relevant context]

The result needs to: [requirements/constraints]

Use this as a reference: [example, if useful]

For example:

I want you to: Draft a follow-up email after a sales call.

The situation: The prospect liked the product but is worried that switching from their current provider will take too much staff time. We agreed I would send our migration guide.

The result needs to: Be under 150 words, sound helpful rather than pushy, mention the migration concern, and suggest a 30-minute technical call next week.

Use this as a reference: Our emails are direct and conversational. We avoid phrases such as “just checking in” and “don’t miss out.”

That is a strong prompt without being complicated.

The Practical Takeaway

Better prompting is not about learning how to “speak AI.”

It is about communicating the task clearly enough that Claude does not have to invent the missing pieces.

Before you send a prompt, ask:

1. Instruction: What exactly do I want Claude to do?
2. Context: What does it need to know that it cannot reasonably guess?
3. Example: Would showing what I want be easier than describing it?

Then look at the response and refine it.

If the result is generic, add context. If the structure is wrong, specify the output. If the tone is off, show an example. If most of the response is good, tell Claude exactly what to keep and what to change.

That is prompting in practice: less guessing, better context, clearer direction, and useful iteration.

FAQ

Do I need special commands or symbols to prompt Claude effectively?

No. Claude responds well to clear, natural language. Special syntax is not necessary for everyday prompting. More structured formatting can become useful when a prompt contains several different types of information.

Is a longer prompt always better?

No. A longer prompt is useful only when the extra information helps Claude perform the task. Ten lines of relevant context can improve an answer; ten lines of irrelevant background can simply add noise.

What should I do if Claude’s response is close but not quite right?

Keep working in the same conversation and tell Claude specifically what needs to change. You can preserve the useful parts rather than starting from scratch.

Should I give Claude examples?

When tone, structure, style, or format matters, examples can be extremely effective. For simple factual questions, they are usually unnecessary.

Should I tell Claude to act as an expert?

Only when the role gives Claude a useful perspective or changes how the task should be approached. A role does not replace the need for clear instructions and relevant context.

Does more context always produce a better answer?

No. Relevant context helps. Irrelevant context does not. Focus on information that could materially change what a good answer looks like.

What if I don’t know how to write the prompt?

Tell Claude what you are trying to achieve and what information you have, then ask it to help structure the request. You do not need to know the perfect prompt before you begin.

Do these principles work with AI tools other than Claude?

Yes. Clear instructions, relevant context, useful examples, and iterative refinement are broadly useful when working with generative AI systems.