The best Claude prompts share one trait: they treat Claude like a capable collaborator with a specific job, not a search box. Claude tends to produce longer, more structured, more carefully reasoned output than other models by default — which makes it especially good at two things people use it for constantly: long-form writing that needs to sound human, and code work that needs to be correct, not just plausible-looking. This guide collects the best Claude prompts for both, organized so you can copy, fill in the brackets, and go.
Claude's defaults lean toward thoroughness and structure — it tends to think through edge cases in code and hold a consistent voice across longer pieces of writing. That's a strength if your prompt gives it a real target to aim at (a voice to match, a bug to find, a spec to follow), and a weakness if your prompt is vague, because "thorough" can turn into "long-winded" without clear constraints. Every prompt below exists to give Claude that target.
Here are 2-3 short samples of writing in the voice I want you to match:
[paste sample 1]
[paste sample 2]
Now write [content type, e.g. "a 300-word blog intro"] about [topic] in
that exact voice — same sentence rhythm, same level of formality, same
kinds of word choices. Don't summarize the samples back to me; just
write in that voice.
Pasting real samples works far better than describing a voice in adjectives ("casual but professional") — Claude is good at pattern-matching tone from examples, less reliable at reverse-engineering it from a description.
I'm writing [content type] about [topic] for [audience]. Target length:
[word count]. Before writing the full draft, give me a section-by-section
outline with one sentence per section describing what it covers. Wait
for my approval, then write the full draft following that outline.
Splitting outline and draft into two steps catches structural problems before you've spent tokens on prose you'll have to rewrite anyway.
Edit the following for clarity and concision without changing my voice
or the substance of what I'm saying. For each significant change, add a
short bracketed note explaining why. Don't rewrite sentences that are
already clear just to sound different.
[paste your draft]
Rewrite the following so it reads like a person wrote it, not an AI:
vary sentence length, cut generic transition phrases ("in today's fast-
paced world," "it's important to note"), remove unnecessary hedging, and
keep specific details instead of vague generalizations.
[paste text]
Turn the following rough notes into a structured [document type, e.g.
"one-page product brief"] with clear headings. Fill obvious gaps with
reasonable placeholders marked [NEEDS INPUT] rather than inventing facts.
[paste notes]
Review this [language] code as a senior engineer would. Flag actual
bugs and correctness issues first, then readability and maintainability
issues, then nice-to-haves. For each issue, explain the failure
scenario in one sentence and suggest a fix. Don't comment on style
choices that don't affect correctness or clarity.
[paste code]
Explain what this code does, section by section, as if onboarding a new
engineer. Call out anything that looks fragile, undocumented, or
non-obvious in its intent before I make any changes to it.
[paste code]
I need to build [feature/change]. Before writing any code, give me a
short implementation plan: what files/functions are affected, the order
of changes, and any edge cases or existing behavior I should be careful
not to break. Wait for my go-ahead before writing code.
Asking Claude to plan first is one of the highest-leverage coding prompts there is — it surfaces wrong assumptions before they turn into wrong code, and it's much cheaper to correct a plan than a finished diff.
This code is producing [exact error / wrong output]. Expected:
[expected behavior]. Actual: [actual behavior]. Walk through the likely
causes in order of probability before proposing a fix, and tell me what
you'd check first to confirm which one it is.
[paste code and relevant error/output]
Write tests for the following function that cover: the normal case,
at least two edge cases, and one case that should raise an error or
return an explicit failure. Use [testing framework]. Briefly explain
what each test verifies.
[paste function]
One thing that separates people who get consistently good results from Claude from people who don't: knowing when to keep pushing in the same thread and when to start over. If Claude's first draft is close but needs adjustment — shorter, more formal, missing a section — stay in the same conversation and iterate; it already has the full context of what it wrote and why. But if you've gone back and forth several times and the output still isn't landing, that's usually a sign the original prompt was underspecified rather than that Claude needs more nudging. In that case, it's often faster to write a new, more specific prompt from scratch than to keep patching a conversation that started on the wrong foot.
The same logic applies to code. If a fix doesn't work, resist the urge to just say "that didn't work, try again" — paste the actual new error or behavior. Claude reasons much better from concrete evidence of what went wrong than from a vague signal that something is still broken.
Want these as reusable templates instead of retyping brackets every time? PromptiFex's prompt library has a fill-in-the-blank version of prompts like these, including a dedicated collection of Claude prompts — see the voice-matched writing prompt pack, the senior developer code review prompt, and the productivity and systems prompt pack for ready-made versions of the ideas above.
Free to start, and the same prompts work whether you're pasting them into Claude, ChatGPT, or Gemini on web, iOS, or Android.
Are these prompts specific to Claude, or do they work with ChatGPT and Gemini too? The underlying structure — outline before drafting, plan before coding, give real examples instead of adjectives — works across ChatGPT, Claude, and Gemini. Claude just tends to follow multi-step instructions and maintain a consistent voice over long outputs particularly well by default, which is why it's a strong default choice for the writing and coding tasks above.
What's the single best Claude prompt for coding? If you only take one: ask Claude to produce an implementation plan before writing any code (prompt #8 above). It's the prompt most likely to catch a wrong assumption before it becomes a wrong diff, and it costs almost nothing extra to include.
How do I stop Claude's writing from sounding like AI wrote it? Give it real writing samples to match instead of describing the tone, and explicitly ask it to cut generic transition phrases, vary sentence length, and keep specific details over vague generalizations (prompt #4 above). Voice-matching from examples consistently outperforms voice-matching from adjectives.
Ready to try these prompts yourself?
Browse 468+ ready-made prompts you can customize and use with ChatGPT, Claude, and Gemini, free to start.