Why Does Claude Give Generic Answers?
Usually because the prompt left things for Claude to guess. Without a reader, a purpose or a standard, it can only write for a typical case. Add the missing details, and the answer has something specific to fit.
Before you answer, read my request below and find what's missing. Request: [your vague request] List the three questions whose answers would most change your response. For each one, give the default you'd assume if I don't answer. Then stop. Once I've replied, write the result for [the reader], in [the format], and meet this standard: [how you'll judge it].
- "Fix the login bug"no symptom, no place, no proof
- Add the symptomfails after session timeout
- Add where to looktoken refresh
- Add the proofa failing test, then the fix
Try it now
- 1Take a request that got a bland answer, and paste it into the prompt above in a new chat.
- 2Answer the three questions in a few words each, or accept the defaults.
- 3Compare the result to the first answer. You'll know it worked when you can't imagine the new answer being given to someone with a different situation.
Use it when
When the output is correct but useless, or when you're on the third "no, more like..." follow-up. Check for the three things most often missing: who it's for, what it's for, and how you'll know it's good.
Skip it when
When you're exploring. Anthropic's Claude Code guide says a loose prompt like "what would you improve in this file?" is fine, as long as you accept more correcting along the way. Open prompts also suit brainstorming, when you want range before you narrow things down.
Why it works
Anthropic's docs compare Claude to a brilliant new employee with no context on your norms, and they say the more precisely you explain what you want, the better the result. Anthropic's Claude Code guide adds that Claude tends to read a request conservatively, building the smallest thing that satisfies it, because it can't see how much you were hoping for. So "improve this" produces a safe, modest improvement. Length isn't the fix. Three specific constraints do more than three paragraphs of general description, as a rule of thumb. The guide's example shows this: it replaces "fix the login bug" with the symptom, the folder to look in, and a request for a failing test first.
Example
"Explain caching" could get a textbook paragraph, since nothing says who's asking. Try "Explain HTTP caching to a back-end developer debugging a stale API response. Focus on Cache-Control headers, in under 150 words, and end with what to check first." That has a reader, a problem, a scope and a length, and you can tell at a glance whether the answer fits.
Going further
When you already have the thing, hand it over instead of describing it. Anthropic's Claude Code guide suggests typing @ and a path to include a file, pasting a screenshot of the error, or piping in output with cat error.log | claude -p "what's going on here". Phrases such as "be thorough" and "think step by step" are ordinary text, and the same guide says the effort level is the main control. The phrase it does recommend is "be ambitious" followed by what you want, because that tells Claude about scope, which only you know.
Common mistakes
- Adding words without adding constraints. A long prompt can be just as vague as a short one.
- Leaving out the reader, then being surprised by the level of detail.
- Using "better" or "more professional" as the only standard. Say what would change, such as "shorter sentences and no jargon."
- Correcting one detail per turn for five turns. Put them all in one message, or start a new chat with a better prompt.
Sources
Checked 6 November 2026. Products change often, so confirm details with the vendor.