Claude Prompts for Writing with Fable 5, Opus 5, and Sonnet 5

Claude prompts for writing work when you name the Claude tier, wrap the brief and source in XML tags, and demand one artifact: a brief, a draft section, or a line edit. You leave with paste-ready wrappers for Claude Fable 5, Claude Opus 5, and Claude Sonnet 5, plus stop rules that keep Claude from inventing quotes or rewriting the whole piece. This page is for writers who chat in claude.ai or the Anthropic API. ChatGPT craft lists, human-voice cut passes, and Claude coding prompts live elsewhere. Soft tip: PromptMake /text can turn a rough writing ask into a labeled scaffold at https://promptmake.net/text before you paste into Claude.

Who Claude writing prompts help

You write for readers and use Claude as a desk for briefs, section drafts, and edits. The patterns fit newsletter authors who dump notes into a tagged source block, editors who need a line pass against a locked draft, client writers who must match a house voice sample, and researchers who turn interview transcripts into a report outline. Students who outline a paper from notes they own land here too.

Skip this page if you want ChatGPT genre craft, a human-voice rhythm kit, or Claude prompts for coding. Those jobs live in other guides. Stay here for Claude-specific paste shapes: XML tags, Projects that pin voice, long source dumps, and tier routing for Fable 5, Opus 5, and Sonnet 5.

Treat Claude as a contractor with a written ticket. You supply genre, audience, length, source notes, a voice sample, and a ban on invented facts. Claude proposes a brief, a section, or an edit. You reject quotes, citations, and events that do not sit in SOURCE.

Claude prompts for writing: the XML paste

A strong set of claude writing prompts leads with tagged blocks before any request for polished prose. Put SOURCE first when the paste is long: interview notes, a chapter, a research dump, or the current draft. Put BRIEF, AUDIENCE, OUTPUT, and RULES after the source so Fable 5 and Opus 5 see the job after a long dump. You wrap blocks in XML tags so they do not mix. Claude reads <source> as evidence and <brief> as the assignment. ALL-CAPS headers work on short pastes. Tags help more once you paste more than a page of notes. Demand one artifact. A brief plus a full draft plus a line edit in one message muddies length and voice. Pick brief, outline, one section draft, or one edit pass. Put the invention ban in a closing reminder so it survives a long paste.

Vague "write a great essay about X" chats produce smooth empty pages because Claude has no fact pool and no output cap. Fable 5, Opus 5, and Sonnet 5 all follow labeled blocks. They invent helper quotes if you omit RULES. They rewrite the whole piece if you omit OUTPUT. Keep a short writing kit outside the chat: working title, audience, length, voice sample, banned claims. Reuse that kit. One artifact per thread. Mixing a brief, a draft, and an edit in the same message hides the decision you needed.

Tags, source, and one artifact

Copy this wrapper and fill the brackets. Keep the tags if a block is short:

ROLE: You are a writing assistant for [genre]. You draft and edit from SOURCE only. You do not invent facts, quotes, citations, statistics, or events.

<audience>[who reads this and what they should leave with]</audience>

<brief>[genre, length, thesis or story goal, tone]</brief>

<source>[paste notes, quotes you may use, research bullets, or the current draft]</source>

<voice>[paste 2 short paragraphs in the target voice]</voice>

<output>[brief | outline | one section draft | line edit]. Cap at [word count]. Return the artifact only, with no preamble.</output>

<rules>Stay inside SOURCE. Mark gaps with [NEED FACT]. Match VOICE for diction and sentence rhythm. Prefer concrete nouns and active verbs. No em dashes. No empty adverbs. No fake citations.</rules>

REMINDER: Output the artifact named in OUTPUT. Do not rewrite the whole piece unless OUTPUT asks for a full draft. Do not echo your reasoning.

You add that reminder to cut two Claude habits. Fable 5 can survey options you did not ask for. Sonnet 5 can return a complete essay when you asked for a 120-word brief. Your rule forces the named artifact.

Projects, long dumps, and [NEED FACT]

Pin ROLE, VOICE, and RULES in a Claude Project so you do not retype them. Paste a fresh <source> and <brief> per piece. Pin house style sheets and a living banned-claim list as Project files. A Project does not replace SOURCE for the piece in front of you. Claude will mix last week's interview into this week's draft if you leave old notes in the same thread.

For a long dump, put <source> first and <output> last. Repeat the artifact name in REMINDER. Add a stop rule: "If SOURCE cannot support a claim, write [NEED FACT] and stop that sentence. Do not guess a quote, a date, or a statistic." Fable 5 and Opus 5 follow this when you put it in RULES and repeat it in REMINDER.

Give Claude the reason for the job in one sentence inside <brief>: who the reader is, and what the page must enable. Fable 5 uses that intent to pick detail. Skip "think step by step" slogans on Fable 5 and Opus 5. Goal, constraints, and format steer them. Asking Claude to transcribe its reasoning can trigger a refusal on Fable 5; demand the artifact, not a thought log.

Brief, draft, and edit templates

You split Claude writing chats into three jobs: a brief before you draft, a section draft from locked source, and a line edit that keeps your facts. Run them as separate messages or separate threads. A combined brief-plus-draft request for 1,200 words of prose hides the decision you needed: thesis, length, audience, and gaps. The templates below share the XML paste from the last section. You swap TASK and OUTPUT. You keep SOURCE and RULES stable so Claude does not invent a new voice mid-thread. Store one wrapper per series or client so you do not rewrite ROLE and RULES each morning.

Pick the Claude tier after you pick the job. Sonnet 5 fits first-pass briefs and short section drafts when SOURCE is short. Opus 5 fits long source dumps, client copy, and edits you will ship. Fable 5 fits tangled arguments, long reports from mixed notes, and edits where claims must match SOURCE line by line. All three fail if you omit SOURCE and OUTPUT. Cap each request at one section so you catch invented facts on the first pass.

Brief and outline prompts

Brief pattern (run this before any draft):

TASK: Propose a writing brief for [working title]. Do not draft body prose yet.

OUTPUT: 1) Goal in 2 sentences. 2) Audience in 1 sentence. 3) Thesis or story goal. 4) Section list with one sentence of intent per section. 5) Gaps marked [NEED FACT]. 6) Word budget per section. Cap the brief at 250 words.

Read the brief out loud. Cross out sections you will not write. Fill [NEED FACT] from your notes. Then send the outline lock:

TASK: Lock the outline from the approved brief. OUTPUT: section headers and one sentence of intent each. Flag any section that needs facts missing from SOURCE. Do not draft body text.

Fable 5 handles a long SOURCE plus a tight brief in one paste. Sonnet 5 handles a short newsletter brief if SOURCE is a page or less. If Claude starts drafting in the brief pass, you forgot "Do not draft body prose yet."

Draft and line-edit prompts

Section draft pattern:

TASK: Draft the [section name] only. Stay inside SOURCE. Match VOICE. OUTPUT: the section as one copyable block. Length: [N] words. Open with a concrete claim or scene, not a definition. If a fact is missing, write [NEED FACT] in place.

RULES: No whole-piece rewrite. No new sections. No citations that do not appear in SOURCE. Lead with the outcome the reader came for.

Line-edit pattern (second message after you accept the section):

TASK: Edit DRAFT against SOURCE and VOICE. Keep meaning and all proper nouns. Cut invented facts. Cut empty adverbs. Cut throat-clearing openers. Cut binary contrast pivots. Vary sentence length. Return the full revised section only. List leftover [NEED FACT] markers after the section.

Honesty pass for public claims (third message, stronger tier):

TASK: Audit DRAFT against SOURCE. Mark any claim missing from SOURCE with [NEED FACT]. Do not rewrite for style in this pass. Return the draft with markers inline.

Paste the section, not the whole article, unless the edit target is the full piece. Opus 5 and Fable 5 match claims to SOURCE with more care than a vague "make this better." Sonnet 5 works if the section is under a few hundred words and SOURCE is short. Fill markers offline from your notes. Publish after you, not Claude, own the claims.

Mistakes that wreck Claude writing chats

Mistake 1: Pasting a topic and writing "write a great essay" with no SOURCE. Claude fills gaps with tidy fake research. You spend the next hour hunting quotes that do not exist. Add SOURCE, OUTPUT, and [NEED FACT].

Mistake 2: Asking for a brief, a full draft, and a line edit in one message. Claude mixes length caps and hides the thesis decision. Split: brief first, one section second, edit third.

Mistake 3: Leaving OUTPUT empty. Claude returns a complete essay plus a survey of angles you will not use. Name the artifact. Cap the word count. Repeat the name in REMINDER.

Mistake 4: Routing a 40-page report rewrite to Sonnet 5 and accepting the first pass. Sonnet 5 is the daily driver. Fable 5 or Opus 5 fit the ship-risk pass on long SOURCE.

Mistake 5: Padding prompts with "think step by step" on Fable 5 and Opus 5. Give goal, constraints, format. Drop CoT slogans unless you measured a gain on that task. Do not ask Claude to paste its reasoning.

Mistake 6: Treating a Project as a junk drawer. Old interviews leak into new drafts. One Project per series or client. Fresh <source> per piece.

Mistake 7: Dumping ChatGPT custom-instruction walls into a Claude Project without XML or OUTPUT. This article stays on Claude paste. ChatGPT writing craft and human-voice cut passes belong on their own pages.

Mistake 8: Trusting Claude on secrets. Strip client names you cannot share, unpublished numbers, and private emails from pastes. Ask for structure on notes you redacted.

Fable 5, Opus 5, and Sonnet 5 notes (mid-2026)

As of mid-2026, Claude Fable 5 is the public top tier for hard reasoning. Claude Opus 5 is the strong enterprise tier for careful long work. Claude Sonnet 5 is the volume model for daily chats and the default on many Free and Pro plans. Haiku 4.5 fits cheap classification, not ship-ready drafts. Mythos 5 stays limited or trusted-access; skip it in public prompts unless your org has it.

Sonnet 5: first-pass briefs, short section drafts, and line edits of a few hundred words. Keep prompts short: ROLE, tagged SOURCE, OUTPUT, RULES. Few-shot a house paragraph if voice slips.

Opus 5: long source dumps, client copy, and honesty audits. Give goal, constraints, format. Repeat [NEED FACT]. Use a separate style pass and a separate fact pass so Claude does not "fix" a gap by inventing a tidy sentence.

Fable 5: tangled arguments, mixed-note reports, and ranked claim checks on long drafts. Paste more context (two source docs, the current draft, the brief). Demand one OUTPUT. Add a brevity line: lead with the outcome; drop options you will not pursue; skip a survey of angles. A Fable 5 rewrite of an entire book chapter is as costly as a Sonnet 5 rewrite if you forgot the cap.

Prompting split that holds: fast Sonnet chats get RTF plus a short voice sample. Fable 5 and Opus 5 get goal plus constraints plus format, with [NEED FACT] and no CoT padding. None of them replace a human read-aloud. Claude cannot see your style guide unless you paste it or pin it in a Project.

Build writing prompts with PromptMake

Write the rough ask in plain words: genre, brief vs draft vs edit, length, and whether SOURCE is notes or a current draft. Open https://promptmake.net/text and generate a structured prompt once. Expect labeled sections you can wrap in XML and fill with BRIEF and SOURCE.

Edit audience, length, and banned claims yourself. PromptMake cannot see your notes. Paste the filled prompt into Sonnet 5 for drafts or Opus 5 / Fable 5 for ship-risk edits. Keep free-tier runs for scaffolding. Guests get about three runs per day; free accounts get about five.

Workflow that sticks: writing kit → PromptMake scaffold → wrap tags and fill SOURCE → Sonnet 5 brief or short draft → Fable 5 or Opus 5 for the hard edit or honesty pass → fill [NEED FACT] from notes → human read-aloud. Store one wrapper per series so you do not rewrite ROLE and RULES each morning.

FAQ

What are the best Claude prompts for writing in 2026?

The best Claude prompts for writing lead with tagged SOURCE, BRIEF, OUTPUT, RULES, and a REMINDER that names one artifact. Demand a brief, an outline, one section draft, or a line edit. Add [NEED FACT] so Claude stops instead of guessing. Match Sonnet 5 for volume and Fable 5 or Opus 5 for ship-risk work on long notes.

How do I use claude writing prompts for a first draft?

Lock a brief first, then paste SOURCE, AUDIENCE, and VOICE with OUTPUT set to one section and a word cap. Ban body prose in the brief pass and draft the approved section in a second message. Fill [NEED FACT] from your notes before you ask for the next section. Sonnet 5 fits a short first draft; move to Opus 5 or Fable 5 if SOURCE is long or claims will go public.

Should I pick Claude Fable 5 or Claude Opus 5 for writing?

Pick Fable 5 when the argument crosses several source docs or the edit needs a line-by-line claim check. Pick Opus 5 when you work a long client piece with a pinned Project and a strict honesty pass. Pick Sonnet 5 for short briefs and daily section drafts. All three need the same XML paste; tier choice does not replace SOURCE.

How do XML tags change Claude prompts for writing?

You use tags so a long dump does not blend with the assignment. Put <source> first on long pastes and <output> last so the job sits after the notes. Fable 5 and Opus 5 follow <rules> and REMINDER with more care than a wall of unlabeled prose. Short chats can use ALL-CAPS headers; switch to tags once notes pass a page.

How do I stop Claude from inventing quotes and facts?

Ban invention in ROLE and again in REMINDER. Tell Claude to mark gaps with [NEED FACT] and to stop that sentence. Run a separate honesty pass that audits DRAFT against SOURCE with no style rewrite. Fill markers yourself; a smooth fake claim costs more trust than a clumsy true sentence.

Do I need a Claude prompt generator for writing work?

A generator helps you scaffold ROLE, OUTPUT, and RULES once. Writing quality comes from SOURCE, VOICE, and the artifact you name. PromptMake /text at https://promptmake.net/text can shape a rough ask; fill notes and length yourself. A generator without your facts yields generic essay advice.

How do I start with free Claude and PromptMake?

Write audience, length, and one SOURCE block in a notes file. Open PromptMake /text, generate a writing scaffold, wrap the blocks in XML, and paste into Claude Sonnet 5 (or the Claude tier your plan includes). Guests get about three PromptMake runs per day; free accounts get about five. Spend those runs on the wrapper, then iterate inside Claude.