The Prompt Framework That Fixes a Vague Claude Answer
6 min read
The reply arrives quickly, it is well written, and it is no use to you. It says that pricing depends on many factors, that you should consider your audience, and that testing is important. Nothing in it is wrong. Nothing in it is about your business either.
That is the shape of a vague answer, and it is almost always a response to a vague request. This article gives you a three-part check you can apply in under a minute, then adds the habits Anthropic recommends for working with Claude specifically. If you are new to Claude, the Complete Beginner's Guide to Claude is the right first stop.
The test Anthropic suggests
Anthropic's prompting guidance offers a simple test. Show your prompt to a colleague who knows nothing about the task and ask them to follow it. If they would be confused, Claude will be too. It describes Claude as something like a brilliant new employee who lacks context on your norms and workflows: the more precisely you explain what you want, the better the result.
That framing does most of the work. A new employee handed "write something about pricing" would come back with questions. Claude, by default, does not ask. It fills the gaps with the most common version of your situation, which is rarely yours.
Three gaps to close: context, task, format
Context is what Claude cannot guess: who the output is for, what situation you are in, what has already been decided, and what the constraints are. Anthropic adds a useful twist: explain why. Its example contrasts "never use ellipses" with "your response will be read aloud by a text-to-speech engine, so never use ellipses because it will not know how to pronounce them." Claude can generalize from the reason, so a single motivation covers cases you did not list.
Task is one instruction, stated plainly. Anthropic recommends being explicit about what you want, and if you want Claude to do more than the task strictly requires, to say so rather than hoping it infers it. When the order or completeness of steps matters, give them as a numbered list.
Format is the shape of what comes back: length, structure, whether you want a recommendation or options, and whether you want the reasoning shown. Anthropic's advice here is to say what to do instead of what not to do. "Write in flowing prose paragraphs" works better than "do not use bullet points."
- 1
Write the context in three facts
The three facts that would change the answer if they were different: audience, situation, constraint. If you cannot name three, the request is still too abstract to answer well.
- 2
State one task
If the word "and" is carrying a lot of weight in your sentence, it is two tasks. Send the first, then the second.
- 3
Describe the format you want, positively
"Three options, each in two sentences, then your recommendation." Not "keep it short."
A before and after
A vague request:
The same request after closing the three gaps:
I run a four-person bookkeeping firm. From January, our monthly plan goes from $120 to $145 for existing clients; new clients pay $145 from today. Most clients have been with us three or more years and many are not very comfortable with software. Write the email announcing this. It will be read by the client's owner, usually on a phone. Be direct about the new price in the first two sentences, give one honest reason, mention what is not changing, and keep it under 150 words. Give me two versions: one warm, one more formal.
”The second has context (who you are, who reads it, where), one task (the announcement), and a format (two versions, length, order of information). The difference in the reply is not subtle.
Two Claude-specific habits
Put the long material first, the question last
If your request includes a long document, Anthropic's guidance is to place it near the top, above your question and instructions. It reports that placing the query at the end can improve response quality by up to thirty percent in its tests, especially for complex inputs with several documents. When you paste more than one document, label each one so Claude can tell them apart and cite the right one.
Document 1 (the signed agreement): [paste] Document 2 (the proposed amendment): [paste]
Using only these two documents, list every change the amendment makes to payment terms. Quote the exact words from each document for every change.
”Show examples, not just rules
Anthropic calls examples one of the most reliable ways to steer format, tone and structure. It recommends three to five, relevant to your real use, varied enough that Claude does not pick up an unintended pattern, and separated clearly from your instructions. If you cannot describe the tone you want, paste two paragraphs you like and say "match this."
Tip
You can ask Claude to improve your examples too: "Are these three examples varied enough? What pattern might you copy that I did not intend?"
What this framework isn't
Common mistake
Treating length as the fix. A longer prompt that repeats the same thin context produces a longer vague answer. Add information, not words.
It is also not a reason to write an essay for every question. A quick factual question needs no framework. The check is for the moments you are about to be disappointed: when the reply is generic, when you have rewritten the same prompt twice, when the stakes are high enough to be worth ninety seconds of setup.
A one-minute checklist
Would a new colleague know what I want, from this message alone?
Does the context include who it is for and why it matters?
Is there exactly one task, and did I say what format I want?
Did I put any long document first and the question last?
Would an example of the result I want save me a paragraph of description?
The same framework works in other assistants, with different quirks: The Prompt Framework That Fixes Vague ChatGPT Answers, Gemini, Copilot and Grok each cover theirs. If you keep falling into the same traps, The Biggest Claude Prompting Mistakes Beginners Make lists them.
Official sources
Checked on October 24, 2026. Features, plans and names change often, so the vendor's own pages are the final word.