What you will be able to do
- Explain why a prompt has to carry all the context Claude needs, and use the colleague test to find gaps
- Paste the actual source material into a request, kept separate from your instructions, instead of describing it
- Give the reason behind an instruction so Claude can apply it to cases the rule does not name
Key concept
The prompt carries all the context — Claude knows only what your request tells it about you, your audience and your task. A good prompt fills that gap on purpose: it states what you want, supplies the material to work from, and explains why, the way you would brief a capable new hire.
1.Brief Claude like a capable new hire
Most weak prompts fail for one reason: the writer knows things the prompt never says. You know the client, the house style, what last quarter's report looked like, and which details matter. Claude knows none of that unless you put it on the page. Anthropic's prompting guide uses an image for this: think of Claude as a brilliant but new employee who lacks context on your norms and workflows. The Claude help centre says much the same thing, describing Claude as a newly-hired contractor with no context about you, your task or your organisation, and advising you to give it very specific instructions.
The image is useful because it tells you what not to cut. A capable new hire does not need you to explain how to write a sentence. They do need to know who the document is for, what it will be used for, what format your team expects, and what is off limits. Leave those out and even a strong writer has to guess, and the guess is usually generic. The guide says it directly: Claude responds well to clear, explicit instructions, and being specific about the output you want improves results.
The documentation gives a simple way to check a draft. Its golden rule is to show your prompt to a colleague who has minimal context on the task and ask them to follow it. If they would be confused, Claude will be too. The test works because it puts you in the reader's position. Words like "the usual format", "the client" or "keep it short" feel complete to you but give a stranger nothing to work with.
Almost everything a newcomer would need: who "everyone" is (staff, visitors, tenants?), the dates and hours of the works, what readers should do differently, where the notice will appear (email, poster, intranet), how long it should be, and the tone. Each of these is something the manager knows and Claude cannot. Adding them turns a generic notice into one the manager can actually publish.
2.Paste the material instead of describing it
Once you accept that Claude starts with no context, the next step follows: give it the actual material. The help centre's advice for prompts that aren't producing helpful answers is to make sure the prompt contains as much context as possible, and to write as if you were instructing someone with no background knowledge of what you are asking. A summary of the policy, the notes or the dataset is your interpretation of it, and anything you leave out of that summary is gone. If you paste the document itself, Claude works from the same source you would.
That doesn't make longer always better. Anthropic's latency guidance asks you to be clear but concise and to avoid unnecessary details or redundant information. It also reminds you that Claude lacks context on your use case and may not make the leaps of logic you expect. So the rule is to include what the task depends on, such as the source text, the audience and the purpose, and to leave out background that changes nothing.
When a request combines instructions with pasted material, keep the two visibly apart. The prompting guide recommends XML tags for this: they help Claude read complex prompts unambiguously, especially when a prompt mixes instructions, context, examples and variable inputs. Put each kind of content in its own tag, such as <instructions>, <context> or <input>. If you are pasting several documents, nest them inside a <documents> tag and give each one its own <document index="n"> tag. Claude can then tell which text is the handbook it should rely on and which text is your request about it.
A technical writer's system prompt says "NEVER use ellipses" but Claude occasionally still uses them in unusual contexts. The writer wants Claude to generalize the rule reliably. What should they add to the instruction?
Correct answer: A — Explain that a text-to-speech engine cannot pronounce ellipses, so the response should avoid them
- A. Correct. Providing the context or motivation behind an instruction helps Claude understand the underlying goal, so it can generalize the rule correctly even to cases the writer didn't explicitly anticipate.
- B. Adding emphasis through punctuation or capitalization does not explain why the rule exists, so it does not help Claude generalize to edge cases.
- C. Listing more banned punctuation marks addresses symptoms rather than the underlying reason, and still would not cover every unanticipated case.
- D. A separate post-generation check adds an extra step but does not address why Claude keeps producing ellipses in the first place.
3.Say why, not just what
A bare instruction covers only the case it names. An instruction with a reason covers every case the reason applies to. Anthropic's guide says that providing context or motivation behind your instructions helps Claude understand your goals and give more targeted responses. Its standard example compares a flat ban with the same ban plus its cause:
| Less effective | More effective | What the reason adds |
|---|---|---|
| NEVER use ellipses | The response will be read aloud by a text-to-speech engine, which cannot pronounce ellipses | Claude understands the real goal is speakable text, so it can avoid other things a speech engine would stumble on |
| NEVER use bullet points | I prefer natural paragraph form because I find flowing prose easier to read | Claude understands the goal is readable prose, so it can make related formatting choices the same way |
Your response will be read aloud by a text-to-speech engine, so never use ellipses since the text-to-speech engine will not know how to pronounce them.The guide sums up the effect in one line: Claude is smart enough to generalize from the explanation. Anthropic's prompting blog describes the bullet-point example the same way. The version with a reason lets the model understand the thinking behind the rule, so it makes better decisions about related formatting. It also lists the kinds of context worth giving: the purpose or audience of the output, why a constraint exists, how the output will be used, and what problem you are trying to solve.
Notice what the effective version does not do. It does not repeat the rule, use capital letters, or add "this is very important". A reason gives Claude something to reason from. Extra emphasis on the same rule only makes it louder.
A prompt for an AI physician's assistant includes patient symptom notes, medical records, and appointment history, and Claude's diagnostic summaries sometimes reference irrelevant details buried in the documents. What technique reduces this noise?
Correct answer: A — Ask Claude to first quote the relevant passages in a dedicated tag before writing its summary
- A. Correct. Asking Claude to quote relevant parts of long documents before carrying out the task grounds the response in the source material and helps it cut through noise from irrelevant sections.
- B. Ignoring the patient records removes the exact source information needed for an accurate, grounded diagnosis and would make the task worse, not better.
- C. A one-word summary would discard nearly all of the clinically relevant detail rather than helping Claude focus only on what matters.
- D. Reading the documents only after answering means Claude's initial answer isn't grounded in the source material at all, which is the opposite of the intended fix.
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.Writing a rule in capitals, or repeating it, is the most reliable way to make Claude follow it.Why is that wrong?
The documented improvement is to explain why the rule exists. With the reason, Claude can apply the rule to related cases, which extra emphasis does not help with.
Covered in Say why, not just what
2.A short description of the source document is enough, because Claude can fill in the rest.Why is that wrong?
Claude has no background on your task, so anything missing from your description is simply unavailable to it. Include the context itself, such as the pasted document.
Covered in Paste the material instead of describing it
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practicesOfficial docs
“Think of Claude as a brilliant but new employee who lacks context on your norms and workflows.”
↩︎ Brief Claude like a capable new hire“Golden rule: Show your prompt to a colleague with minimal context on the task and ask them to follow it.”
↩︎ Brief Claude like a capable new hire“Claude responds well to clear, explicit instructions. Being specific about your desired output can help enhance results.”
↩︎ Brief Claude like a capable new hire“XML tags help Claude parse complex prompts unambiguously, especially when your prompt mixes instructions, context, examples, and variable inputs.”
↩︎ Paste the material instead of describing it“Providing context or motivation behind your instructions, such as explaining to Claude why such behavior is important, can help Claude better understand your goals”
↩︎ Say why, not just what“The more precisely you explain what you want, the better the result.”
↩︎ Key concept“Claude is smart enough to generalize from the explanation.”
↩︎ Exam trap 1 - 2.
“Think of Claude as a newly-hired contractor.”
↩︎ Brief Claude like a capable new hire - 3.https://support.claude.com/en/articles/7996857-my-prompt-isn-t-giving-me-a-helpful-answerOfficial docs
“Pretend you are giving these instructions to someone with no background knowledge about what you are asking.”
↩︎ Paste the material instead of describing it“Make sure that your prompt contains as much context as possible.”
↩︎ Exam trap 2 - 4.https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-latencyOfficial docs
“Avoid unnecessary details or redundant information, while keeping in mind that Claude lacks context on your use case”
↩︎ Paste the material instead of describing it - 5.https://claude.com/blog/best-practices-for-prompt-engineeringSecondary source
“The second version helps the model understand the reasoning behind the rule, which allows it to make better decisions about related formatting choices.”
↩︎ Say why, not just what“Describing how the output will be used”
↩︎ Say why, not just what