Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Claude is more likely to deliver useful work when your prompt makes the task, context, and desired result explicit. Anthropic’s current Prompting best practices guide says Claude responds well to clear instructions, but it does not publish a canonical list of 19 mistakes. The checklist below is an editorial audit of outdated or unhelpful prompting habits drawn from that guidance—not a ranked set of proven causes of workflow failure.
Start with a clear task and a definition of success
When you ask, “Can you help with this?”, Claude has to infer what kind of help you want. Anthropic’s guidance is to be direct and explicit. A prompt should tell Claude what to do and what a successful result looks like.
1. Asking vaguely instead of naming the task
Replace “Look at this” with an action: summarize, compare, classify, rewrite, extract, or diagnose. Specify the material Claude should work on.
2. Leaving success criteria implicit
Define what a good answer must accomplish. For example, ask for a summary that preserves the decision, supporting evidence, and unresolved questions rather than simply requesting “a useful summary.”
#1 Best Overall
3. Failing to explain why the work matters
Relevant motivation helps Claude tailor its response. Explain the audience, use case, or decision the output will support; omit background that does not affect the answer.
4. Asking for an unspecified level of detail
Say whether you want a short answer, a detailed explanation, or a step-by-step procedure. If length matters, state a practical limit or target instead of expecting Claude to guess.
Specify the output before Claude starts
A sound answer can still be inconvenient if it arrives in the wrong shape. State the format and constraints as part of the task, especially when the result will be reused in a report, workflow, or application.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →5. Leaving the format open
Name the form you need: a table, numbered steps, an email, JSON, or prose. If the output must follow a schema, include the required fields and explain whether extra text is allowed.
6. Forgetting audience and tone
Describe who will read the result and how it should sound. “For a new support agent, calm and plain-spoken” gives more useful direction than “make it professional.”
7. Omitting important constraints
State requirements such as what to include, what to exclude, which terminology to preserve, or whether Claude should use only the supplied material. Keep constraints relevant and specific rather than piling on contradictory demands.
Separate instructions from material to analyze
When a prompt mixes commands, background, examples, and source text together, it can be harder to see which material plays which role. Anthropic recommends using XML tags to distinguish parts of complex prompts. Tags are organizational labels, not a substitute for clear instructions.
8. Blending the task with the source material
Label the input and the requested action separately. For example, place source text inside <document>...</document>, then state outside that tag what Claude should extract from it.
9. Using one undifferentiated block for everything
For a complex prompt, use descriptive tags such as <instructions>, <context>, <examples>, and <input>. This can make boundaries clearer when the prompt contains several kinds of information.
10. Making reusable prompts hard to update
Keep stable instructions distinct from information that changes between runs, such as a customer message or report. Anthropic’s guide covers prompt templates and variables; clearly marked input fields make it easier to replace one case’s data without accidentally changing the task.
Use examples that match the real job
Examples can show the desired format, tone, and level of specificity more concretely than abstract instructions alone. Anthropic recommends using 3–5 examples for best results. That is guidance, not a guarantee or a universal measured optimum.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errors11. Providing no example when the target is hard to describe
If the output has a distinctive structure or voice, include a representative example and state which qualities Claude should follow. An example is particularly useful when “good” depends on details that are awkward to express as rules.
12. Choosing examples that do not reflect the use case
Use examples drawn from the actual kind of work Claude will perform. An example for a casual product description may not steer a formal policy summary in the right direction.
13. Showing only one narrow case
Where inputs vary, include examples that reflect meaningful variation or edge cases. This gives Claude a better picture of which features are constant and which should change with the input.
Arrange long documents so the question is easy to find
For long-context tasks, placement and organization matter. Anthropic’s guide advises putting source material near the top, structuring documents and metadata, and placing the query after the context. It identifies 20k+ tokens as the scope for its long-context recommendations; this does not mean every prompt above that size needs identical treatment.
Recommended Free Tools
14. Putting a long source after the question
For a task involving long documents, provide the source material first and ask the question afterward. Anthropic reports up to a 30 percent improvement in response quality in tests, especially with complex, multi-document inputs, when queries come at the end of long context. This is a qualified finding from those tests, not a result to expect from every prompt.
15. Sending multiple documents without structure
Separate documents and label them with useful metadata, such as title, date, or source, when those details matter. Clear boundaries help Claude identify which source supports which part of the answer.
16. Hiding the relevant question in a long prompt
After the source material, make the question and requested output easy to identify. If there are several questions, number them and specify whether Claude should answer all of them or prioritize one.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Say whether Claude should advise or act
A request for ideas is not necessarily a request to implement them. If Claude has access to tools in the workflow, say explicitly whether it should use them and what action to take. The available tools and permissions depend on the product or integration in use.
17. Asking for a fix when you only want a diagnosis—or vice versa
Clarify whether you want an explanation, a recommendation, or a change to be made. For example: “Identify the bug and explain the cause; do not edit the files” is different from “Apply the smallest safe fix and show the changes.”
Best Value
18. Expecting a tool action without requesting it
If the workflow supports a tool and you want Claude to use it, say so directly and specify the intended outcome. A vague request for suggestions can produce advice rather than implementation.
Retire blanket rules and check guidance when models change
Prompt advice is not all equally portable. Anthropic’s documentation includes general recommendations as well as model-specific guidance and migration considerations. Avoid treating every prompt rule as universal or permanent.
19. Reusing old model-specific rules as permanent law
When changing model generations, check Anthropic’s current guide for advice tied to the model or its behavior. Also be cautious with blanket demands that encourage overtriggering or overengineering in some model contexts: replace them with targeted instructions about the result you actually want. Test important prompts in the model and workflow where they will run rather than assuming one prompt is optimal everywhere.
A practical prompt audit
Before reusing a prompt, check that it identifies the task, success criteria, relevant context, output format, and any constraints. For complex inputs, separate the instructions from the source material; add representative examples when they clarify the target; and place the question after long context. If the task requires action, specify whether Claude should use an available tool or only advise. These checks reflect Anthropic’s guidance, not a guarantee of a particular result.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
