These 40 Claude prompts for writing cover the full content workflow — drafting, editing, voice, fiction, email, headlines, repurposing, and research — and every one is copy-paste ready. They come from my own snippets file, built up through steady use of Claude as a daily writing partner, and each group includes a note on the Claude-specific technique that makes it work.
TL;DR:
• 40 prompts in 8 groups of 5, from first draft to final polish.
• The techniques that matter most for writing: paste examples of your voice inside XML tags, cap feedback so Claude prioritizes, and explain who the reader is and why the piece exists.
• Claude is a stronger editor than drafter — the highest-value prompts here are the revision ones.
How to Get Good Writing Out of Claude
Three habits separate usable output from generic AI prose. Show, don’t describe: two pasted paragraphs of your writing inside a <voice_sample> tag beat any adjective list. Give Claude the why — “this intro must survive a reader who will leave in ten seconds” produces a different opening than “write an intro.” And cap everything: three revisions, five headlines, one paragraph — uncapped requests come back diluted. These are the same core moves as our main Claude prompts library, tuned for prose; Anthropic’s prompt engineering guide explains the mechanics.
| Group | Prompts | Use when |
|---|---|---|
| Blog posts & articles | 1–5 | Drafting long-form from scratch |
| Editing & revision | 6–10 | You have a draft that isn’t working |
| Voice & style | 11–15 | Output sounds like AI, not you |
| Fiction & creative | 16–20 | Stories, scenes, characters |
| Email & newsletters | 21–25 | Messages that need replies |
| Headlines & hooks | 26–30 | Titles, intros, social copy |
| Repurposing & summaries | 31–35 | One piece, many formats |
| Outlines & ideas | 36–40 | Before the first word |
The 40 Claude Prompts for Writing
Blog posts and articles (prompts 1–5)
Prompt 1 — the grounded first draft:
You are a senior writer for [publication type]. Draft an article.
<topic>[topic and working title]</topic>
<reader>[who they are, what they already know, why they clicked]</reader>
<raw_material>[your notes, data, quotes, links — everything you have]</raw_material>
<voice_sample>[2-3 paragraphs of your own writing]</voice_sample>
Use only claims supported by the raw material — where you'd need a
fact I haven't given you, insert [NEEDS SOURCE] instead of inventing
one. Match the voice sample. Open with the answer, not a wind-up.
The [NEEDS SOURCE] instruction gives Claude a legal way to not know something, which stops it fabricating statistics. The voice sample does more for tone than any instruction can.
- Write the opening 100 words of this article three ways: one starting with a scene, one with a claim, one with a question. Same facts each time.
- Here’s my outline. Flag the sections where I’m asserting instead of showing, and tell me what evidence or example each one needs.
- Expand this bullet list into prose without padding: every sentence must carry information from the bullets or connect two of them. No filler transitions.
- Write the conclusion for the draft below. It must earn its place: one synthesis the reader couldn’t have made at the start, one next action. No summary of what they just read.
Editing and revision (prompts 6–10)
Prompt 6 — the prioritized edit:
You are a developmental editor. Read the draft and give me only
the three highest-impact revisions — structural, not line edits.
<draft>[paste draft]</draft>
<goal>[what this piece must accomplish, for whom]</goal>
For each: what's weak, why it costs me readers, and the fix
demonstrated on my own sentences. Then stop. Do not line-edit.
Capping at three is the whole trick — uncapped, Claude returns twenty flat suggestions. “Demonstrated on my own sentences” converts advice into usable material.
- Line-edit this passage for economy only. Cut 25% of the words without losing a single fact or softening a single claim. Show the before/after word count.
- Read this draft as a hostile skimmer: quote every sentence where you’d stop reading, and say why. Don’t fix anything yet.
- Check this piece for AI tells and stock phrases — hedge words, empty intensifiers, “in today’s world” openings. List each with a plainer replacement.
- My editor said this draft “lacks a spine.” Identify the actual argument buried in it, state it in one sentence, then reorder my sections to serve it.
Voice and style (prompts 11–15)
Prompt 11 — the voice extraction:
<samples>
[paste 3-5 pieces of your best writing]
</samples>
Analyze these and write a style guide for imitating this writer:
sentence rhythm, vocabulary register, how they open and close,
what they never do. Be specific enough that another writer could
follow it. I'll use this as a system prompt for future drafts.
Run this once, then paste the result into a Project’s instructions or a system prompt — our Claude system prompts guide shows exactly where it goes. Every future draft starts in your voice instead of default-Claude.
- Rewrite this passage in my voice, using the style guide below. Where my guide conflicts with “good writing,” follow my guide.
- Take this paragraph up two notches in formality without making it stiff, then down two notches without making it sloppy. Three versions side by side.
- This draft sounds like everyone. Find the three most generic sentences and rewrite each with a concrete detail from my notes below.
- Compare these two drafts of mine — one readers loved, one they ignored. What does the loved one do at the sentence level that the other doesn’t?
Fiction and creative writing (prompts 16–20)
Prompt 16 — the scene that shows:
Write this scene. Think about what each character wants and is
hiding before you draft.
<scene_brief>
Characters: [who, and what each wants in this scene]
Setting: [where/when]
What must happen: [the plot event]
What must NOT be stated directly: [the subtext]
</scene_brief>
Rules: reveal the subtext only through action and dialogue.
No character says what they feel. End one beat earlier than
feels comfortable.
The “must not be stated” field forces subtext, which is exactly where flat AI fiction fails. The thinking trigger helps Claude hold character motives steady across the scene.
- Interrogate my protagonist: ask them 10 questions a skeptical detective would ask, and answer as the character. Flag any answer that contradicts my character notes below.
- Here’s my chapter. Mark every place where I summarized an emotion instead of dramatizing it, and rewrite the worst one as a scene.
- Give me 5 complications for this plot situation that arise from character flaws I’ve already established — no new characters, no coincidences.
- Rewrite this dialogue so the two characters stop answering each other directly — people talk past each other. Keep the information transfer intact.
Email and newsletters (prompts 21–25)
Prompt 21 — the email that gets a reply:
Write an email for me.
<to>[who they are, our relationship, how busy they are]</to>
<outcome>[the ONE thing I need them to do]</outcome>
<context>[what they already know; what's new]</context>
Under 120 words, because they read on a phone between meetings.
The ask appears in the first two sentences AND as the final line.
No "hope this finds you well."
- Rewrite this email I’m about to send while angry. Keep every substantive point, remove everything I’d regret, and tell me which sentence was the most expensive.
- Draft the newsletter issue from my notes below in my usual structure: one idea, one story that proves it, one action. Voice sample attached.
- Write a follow-up to someone who ignored my last email — reference it without guilt-tripping, add one new piece of value, re-ask smaller.
- Turn this rambling update into a status email execs will actually read: headline verdict first, then three bullets — done, blocked, needed.
Headlines and hooks (prompts 26–30)
Prompt 26 — headlines with a kill rule:
Generate 10 headlines for the article below, then act as a jaded
editor and kill 7 of them, saying why each died. For the surviving
3, name the specific reader who clicks it.
<article>[paste draft or summary]</article>
Constraint: every headline must be true to the article — no
promises the piece doesn't keep.
Generate-then-kill beats “give me your 3 best” — the critique pass applies a different, harsher standard than the generation pass, and you get to see the reasoning.
- Write 5 opening sentences for this piece that make deleting the tab feel like a loss. No questions, no “imagine,” no statistics I haven’t given you.
- Turn this article’s core insight into one social post per platform below — native to each, not cross-posted. The insight must survive intact.
- My title promises X but my draft delivers Y. Either give me 5 titles that honestly sell Y, or tell me what the draft needs to earn X.
- Write the meta description for this post: under 155 characters, contains [keyword], and makes a promise the first paragraph immediately keeps.
Repurposing and summaries (prompts 31–35)
Prompt 31 — the faithful summary:
<source>[paste the long piece]</source>
Summarize this at three lengths: one sentence, one paragraph,
one page. Rules: preserve the author's actual argument, including
its qualifiers — do not flatten "usually" into "always." Quote the
single most important sentence verbatim in the one-page version.
- Turn this webinar transcript into a blog post: cut the verbal filler, keep the speaker’s phrasing for key ideas, and mark timestamps where I should embed clips.
- Break this 2,000-word guide into a 7-part email course — each email one idea, one example, one task, and a reason to open the next.
- Extract every quotable line from my draft below — sentences that stand alone without context. Rank them by how likely someone shares each.
- Rewrite this technical explainer for a smart 15-year-old: same facts, no jargon without an immediate plain-language gloss, examples from everyday life.
Outlines and idea generation (prompts 36–40)
Prompt 36 — the outline that argues:
I'm writing about [topic] for [reader]. Before outlining, think:
what does this reader believe that's wrong, and what do they need
to believe by the end?
Build the outline as an argument, not a topic list — each section
header is a claim, and each section earns the next. Flag the
section where I'm most likely to lose them.
- Give me 10 article angles on [topic] that assume the obvious take is already written. For each: the counterintuitive claim and the evidence I’d need.
- Here are my last 10 published pieces. What themes am I circling without committing to? Propose the piece that commits.
- Interview me for this article: ask one question at a time, follow up on vague answers, and stop when you have enough for an outline. Then show the outline built only from my answers.
- I have these three half-ideas below. Find the piece they’re all secretly part of, or tell me honestly that they’re three separate pieces.
Common Mistakes With Writing Prompts
- Asking for the whole piece in one shot. Draft, then critique, then revise — three short prompts beat one long one every time.
- Describing your voice instead of showing it. “Conversational but authoritative” means nothing; two pasted paragraphs mean everything.
- Skipping the reader. Claude defaults to writing for everyone, which reads as no one. Name the reader in every drafting prompt.
- Trusting generated facts. Statistics and quotes in a Claude draft need verification unless you supplied them. The [NEEDS SOURCE] pattern from prompt 1 makes this systematic.
For the workflow around these prompts — Projects, document uploads, revision loops — see our Claude for writing guide. If your writing is commercial, the marketing prompts list picks up where this one stops.
FAQ
Which Claude model is best for writing?
Sonnet 5 handles most drafting and editing well. For long manuscripts, subtle voice matching, or structural editing of a full book chapter, Opus 4.8 or Fable 5 noticeably outperform it. Haiku 4.5 is fine for summaries and quick rewrites.
How do I make Claude write in my voice?
Paste three to five samples of your best writing and ask Claude to extract a style guide, then reuse that guide as a system prompt or Project instruction. Showing samples always beats describing your style with adjectives.
Can Claude write a whole article for me?
It can produce a full draft, and with good raw material the draft is usable. The reliable workflow is still human-led: you supply facts and voice samples, Claude drafts, you cut and verify, Claude polishes. Facts you didn’t supply need checking before publishing.
Why does Claude’s writing sound generic, and how do I fix it?
Generic output usually means a generic prompt: no reader, no voice sample, no stakes. Name the specific reader, paste a voice sample in an XML tag, explain why the piece exists, and cap the scope. Then run a second pass asking Claude to hunt down its own stock phrases.
Are these prompts useful for fiction as well as nonfiction?
Yes — prompts 16 through 20 are fiction-specific. The key fiction technique is constraint-based prompting: tell Claude what must not be stated directly, and force subtext through action and dialogue. Claude follows structural constraints in creative work unusually well.
Next Steps
Start with prompt 11 — extract your voice once, install it as a system prompt, and every other prompt on this page improves. Then browse the curated 100 best Claude prompts of 2026 for the strongest picks across every category, or head back to the prompts section for the specialized lists.
