Better Results with Claude — Anthropic: Tasks and Practical Tips

How can you get better results with Claude?

Give Claude a clear goal, relevant source material, and specific requirements for the result. Condition what is certain, keep missing information visible, and provide targeted feedback; a useful assignment must above all be clear—not complicated or excessively long.

Better assignments for Claude start with the desired result

You’ll generally get more useful answers when you describe what you want to be able to do afterward. “Help with my customer service” leaves a lot open-ended. “Write a response to this customer inquiry based on the product terms below” provides a clearly defined task.

That clarity also helps when evaluating the outcome. You can check whether the question has been answered, the conditions have been met, and the text is ready to be sent. Without a clear goal, “good” remains mostly a feeling.

In its guide to writing better prompts, Anthropic recommends formulating prompts clearly, providing relevant background information, breaking down complex requests, and using feedback. You don’t need technical prompt language to do this.

A good prompt is a clear and understandable task description. If a colleague with no prior knowledge wouldn’t understand what you mean, there’s likely information missing for Claude as well.

The five components of a useful Claude prompt

A prompt is the question or task you give to Claude. For many everyday tasks, you can structure it into five parts. Not every part needs to be detailed; sometimes everything fits into just a few sentences.

Example of a complete order

Write a customer email addressing the installation question below. The customer has little technical experience. Use only the provided instructions. Do not include product features or promises. Start with the direct answer and conclude with the information we still need.

This assignment does not involve any complex formulas. However, it is clear who the reader is, what information is key, and what requirements the text must meet.

Include only context that makes a difference

Relevant background information is helpful. A complete company history is usually unnecessary when you’re looking to improve a single paragraph. Focus on providing information that influences the content, tone, or choice.

In a product description, the target audience and confirmed features are relevant. The names of internal employees often are not. Ask yourself: would the answer change because of this detail? If not, it can probably be omitted.

Turn vague wishes into reality

Words like “professional,” “strong,” and “human” can mean many things. For one user, “professional” means formal and comprehensive. For another, it means concise, accessible, and free of jargon.

Translate these requirements into tangible characteristics of the result. That way, you can assess afterward whether Claude followed your preferences.

An example can be helpful if the tone you’re looking for is difficult to describe. Show what you mean, but also make it clear which characteristics of that example should be included.

Use examples without including unwanted content

A short sample text can demonstrate how you want to structure a response. This is useful for recurring customer emails, reports, or product explanations.

Choose an example that truly fits. An enthusiastic sales pitch isn’t a good stylistic example for a sensitive complaint. Otherwise, Claude might end up adopting exactly the characteristics you want to avoid in this situation.

Example: same tone, different facts

Below is an example of the desired tone. Use the direct opening and simple wording. Do not copy any names, amounts, product features, or agreements from the example. Write the brand new text using only the information under “Brand New Situation.”

This distinction prevents confusion between a style example and a source of information. This is especially important when you use examples from previous customers.

Start with one well-chosen example. Only add more when you notice that a recurring distinction isn’t clear enough. More material doesn’t automatically make an assignment better.

Let Claude work with the right source

An answer can be clear yet still be based on incorrect information. For example, because an outdated manual was used, a general product feature was listed, or two versions were confused.

Therefore, specify which source is the primary one. Use recognizable file names, and if there are multiple documents, indicate their purpose.

Example: facts and writing style from various files

Use “Product Specifications—Approved” for all product facts. Use “Sample Texts” solely as a reference for tone and structure. If information is missing or sources contradict each other, include that in a list of questions.

This is much clearer than uploading five files and simply asking, “Make something good out of this.”

Ask for the rationale behind key conclusions

When analyzing a document, you can request a reference to a specific passage, table, or page. This makes it easier to verify the results.

Distinguish between what is literally stated in the source and what is an interpretation. For example, a report may show a decline in revenue without identifying the cause. The cause should not be included in the summary as a fact without proper attribution.

An upload does not guarantee that everything is usable

According to Anthropic’s file documentation, processing varies depending on the format and content. Embedded images, scans, and large documents may require extra attention.

Highlight the relevant sections and check that they are legible. A clear instruction cannot fix a missing page or an illegible table.

Make missing information visible

A well-written assignment leaves room for uncertainty. You don’t want Claude to fill in a missing amount just because the table would otherwise seem incomplete.

In its guidance on limiting incorrect answers, Anthropic mentions, among other things, allowing for uncertainty and basing claims on provided information.

Distinguish between confirmed facts, interpretations, and missing data. Do not fill in missing amounts, dates, or conditions yourself. Indicate which missing information could change the conclusion.

This way, an open issue becomes part of the result. That’s more useful than a seemingly complete overview that you later discover contains assumptions.

Ask only necessary questions

Sometimes Claude needs to know something first in order to proceed. For other tasks, a usable concept can be created with clearly identified open issues.

You can explicitly highlight that distinction:

First, ask a question if missing information would lead to a significantly different answer. If you can already work out the rest, do so and highlight what still needs to be added.

This helps you avoid both unnecessary interruptions and unwanted guesswork.

Schedule complex tasks when monitoring is needed on the go

A large order doesn’t always have to be placed all at once. Especially when later steps depend on earlier choices, it can be helpful to reassess your needs along the way.

Suppose you’re having a knowledge base article written. If the target audience and scope have been misinterpreted, it’s of little help to correct them only after a long final draft has been completed.

A practical guide for longer writing projects

1. Define the question: What should the reader understand by the end?
2. Organize the material: which sources and facts are relevant?
3. Review the structure: does it include all the necessary topics?
4. Revise the text: use the approved options.
5. Check the content and readability: improve the sections that need attention.

This sequence is particularly useful when working with dependencies. For shortening a short email, five separate rounds are usually excessive.

Describe the result, not every step of the process

You don’t need to tell Claude exactly what internal reasoning to follow. Instead, specify the outcome you need and the criteria that matter.

When making a comparison, for example, ask for the underlying assumptions, differences, and a brief justification. This provides verifiable information without confusing a lengthy line of reasoning with evidence.

Provide feedback that makes a difference

“Not good” doesn’t say much about what needs to be improved. Describe what you see, why it doesn’t work, and what you’d like to see instead.

A useful correction, for example, is: “The second paragraph assumes prior technical knowledge. Explain the concept first, and then include the installation step.”

Example: improving text that is too general

The text lists many general benefits but does not explain what the user needs to do. Replace the second and third paragraphs with specific offers. Keep the opening and the existing terms and conditions.

This way, the good parts are preserved. Plus, you can accurately assess whether the problem has been resolved.

Don’t keep having everything rewritten

A completely brand new version may unintentionally change other passages. Therefore, when requesting a minor correction, please ask that only the relevant section be modified.

If the starting point is incorrect, a more extensive revision may be necessary. In that case, explicitly condition which option is no longer valid and which brand new direction will be followed.

Use a single clear instruction for the current version. Otherwise, old and brand new instructions may remain side by side without it being clear which one takes precedence.

Practical Claude commands for everyday tasks

The examples below are standalone starting points. Replace the text in square brackets with your own information and include only the material necessary for the assignment.

Compose a customer email

Write a response to [customer inquiry]. Use this confirmed information: [facts]. The goal is [desired outcome]. Keep the tone friendly and clear. Do not include any discounts, refunds, delivery dates, or other commitments not specified in the details. Submit only the draft email.

Be sure to check the terms and conditions and commitments before placing your order. A small change from “may” to “will” can make the meaning stronger than intended.

Summarizing a document effectively

Summarize this document for [target audience]. Focus on [topic]. Highlight the key findings, relevant exceptions, and open questions. When discussing important conclusions, refer to the relevant passage. Do not provide a general summary of sections that are not relevant to the question.

This approach focuses on usability. Someone preparing to make a decision needs different information than someone who has to carry out a procedure.

Compare two offers

Compare these offers based on price, included features, exclusions, and terms. Only compare directly comparable items side by side. Mark missing information as “not listed.” Then list the questions needed for a fair comparison.

Don’t rush to declare a winner when essential conditions are missing. A well-designed questionnaire can yield the most valuable results at that stage.

Making an explanation easier to understand

Rewrite this explanation for someone with no prior knowledge. Retain the meaning and necessary technical terms, but explain them briefly. Use one concrete example. After the text, indicate which passage still needs clarification in terms of content.

Read the result as if you’re seeing the topic for the first time. Simplification is successful when the explanation becomes more accessible while retaining the important nuances.

Analyzing a data file

Check this file before drawing any conclusions. Identify the meaning of the columns used, missing values, possible duplicate rows, and different units. Then calculate [result] according to [definition]. Indicate which data you have excluded and why.

Accurate calculations start with the right definition. For example, revenue, payments received, and profit are different metrics. Define exactly what you want to measure.

Review a schedule

Evaluate this schedule for dependencies, available capacity, and outstanding decisions. Use only the specified personnel, deadlines, and resources. Identify bottlenecks and propose up to three adjustments. Make it clear which adjustments require a brand new agreement.

This way, you’ll receive a proposal that you can discuss, without extra people or extended deadlines being tacitly assumed to be available.

Up-to-date information requires a targeted search

When it comes to prices, product terms, and recent developments, clear writing guidelines alone aren’t enough. You also need up-to-date sources.

Claude can retrieve information and provide source references using available web search capabilities. Anthropic describes this in the web search guide.

Find the current genuine terms and conditions for . Use sources from the provider and specify the version, region, and date to which the information pertains. Point out any inconsistencies and do not list a current price if it cannot be verified.

Check the key sources yourself. A link may point to a legitimate page but still describe a different situation. Be sure to verify that the cited information supports the corresponding claim.

Avoid letting rigid instructions hinder your project

General preferences can be useful, such as for language and tone. But a rule like “always respond very briefly” may conflict with a request for a detailed analysis.

In its explanation of personalization, Anthropic distinguishes between general instructions, project-specific guidelines, and specific procedures.

Use this breakdown thoughtfully. Set a permanent language preference in your general instructions. Keep agreements for a specific client associated with the relevant project. Specify the exact length and format in the current assignment.

Check old rules in case of unexpected behavior

Do you keep getting bullet points when you’re asking for continuous text? Or is the nuance always missing because the answers are too brief? If so, check to see if a previous preference is still affecting the results.

Don’t add more rules right away. First, remove or clarify what no longer applies. A short, coherent set of instructions is easier to maintain than a list of exceptions.

What should you do if you’re not satisfied with the result?

First, try to determine what kind of problem you’re experiencing. The solution for an incorrect source is different from that for an incorrect pitch.

It’s best to change one key element at a time. That way, you’ll be able to better identify which adjustment makes a difference. If you change the source material, tone, structure, and purpose all at once, the comparison becomes unclear.

Evaluate the results using fixed criteria

An answer may be a pleasure to read but still be unusable. Therefore, choose a few criteria in advance that fit the task.

• Contents: Have the facts, terms, and exceptions been retained?
• Relevance: Does the result answer the actual question?
• Justification: Can key conclusions be traced back to sources?
• Format: Does the text or table fit the intended use?
• Follow-up work: how much correction is still needed?

If necessary, have Claude identify which components do not meet your requirements. This can be useful for a future version, but it is not an independent audit.

When it comes to important information, comparison with the genuine remains necessary. A second formulation of the same conclusion does not constitute brand new evidence.

Save tips that work in practice

If a task consistently yields a usable result, you can save it as a template. Clearly indicate which parts are fixed and which need to be filled in based on the specific situation.

For example, keep the purpose, structure, and checkpoints. Replace customer names, amounts, and old product data with recognizable placeholders. This prevents a reused prompt from carrying over outdated information.

Review a completed assignment when your approach or the model you’re using changes. A previous formulation may still work well, but that will only become clear from the results.

In its overview of prompt improvement, Anthropic emphasizes the importance of clear success criteria and evaluation. For everyday use, this doesn’t have to involve extensive research: a few representative tasks and consistent evaluation are enough to provide a solid foundation.

Terms for improved Claude prompts

Frequently Asked Questions

How do you write a good prompt for Claude?

Describe the desired outcome, provide relevant background information and sources, and specify the key constraints. Also indicate how you’d like to receive the answer.

Does a Claude prompt have to be long?

No. A simple assignment can be brief. Include only the information necessary to complete the task properly or to evaluate the result.

Should I assign Claude an expert role?

An outside perspective can be helpful—for example, that of an editor or a critical customer. Such a perspective does not replace source information and does not automatically make the answer expert or correct.

How can I get less generic text?

Please specify a concrete target audience, situation, and specific question. Include your own facts or examples, and indicate which general statements are of little value.

Can I ask Claude not to make up information?

Yes. Request that missing information be kept visible and that facts be separated from interpretations. This helps define the scope of the assignment but does not eliminate the need for verification.

Do I have to start a brand new conversation every time?

That depends on the task. Continue where you left off if the same sources and choices remain relevant. For a brand new topic, a brand new conversation may be clearer, with a brief summary of the necessary context.

Why doesn’t the same prompt always work just as well?

Source material, conversational context, settings, and models used may vary. Therefore, evaluate the result and make targeted adjustments to anything that is not sufficiently clear for the current task.

Key Insights

• Start by defining the result you need and clarifying your target audience, resources, and requirements.
• Translate vague requests into specific features and use an appropriate example when helpful.
• Keep missing information visible and verify key conclusions in the genuine source.
• Provide specific feedback and suggest changes only to the areas that need improvement.
• Use completed projects as a baseline and evaluate their quality based on your own work.