How to Use Claude for Writers

2. Brainstorming story ideas

2.1 Prompt structure

Start with a clear genre and constraint. Example prompt:

Write three high‑concept sci‑fi premises. Each premise must include:
- A protagonist with a clear goal
- A single twist that changes the stakes
- A hook no longer than 30 words.

2.2 Example output

Claude returns concise ideas that you can copy directly into your notes.

2.3 Saving ideas

Copy the result into a markdown file or a Notion database. Tag each idea with “#ClaudeBrainstorm” for later retrieval.

3. Building a detailed outline

3.1 Outline prompt

Give Claude the chosen premise and ask for a three‑act structure. Include word‑count targets.

Using premise #2 from the previous list, create a detailed three‑act outline.
- Act 1: 1,500 words, introduce protagonist and inciting incident.
- Act 2: 3,000 words, three major obstacles.
- Act 3: 1,500 words, climax and resolution.
Provide bullet points for each scene.

3.2 Comparison table – Claude vs. other tools

3.3 Exporting the outline

Claude can output JSON for import into Scrivener:

{
  "act1": [
    {"scene":1,"summary":"Protagonist discovers the alien signal"},
    {"scene":2,"summary":"First confrontation with antagonist"}
  ],
  "act2":[...],
  "act3":[...]
}

4. Drafting and expanding scenes

4.1 Expanding a bullet point

Take a scene bullet and ask Claude to write a 500‑word draft.

Expand the following scene bullet into prose (≈500 words):
- Scene 3: Protagonist hacks the security system while the alarms blare.

4.2 Controlling style

Specify tone, point of view, and any required literary devices.

Write in third‑person limited, with a gritty noir tone. Include at least two metaphors.

4.3 Sample excerpt

5. Editing and style polishing

5.1 Grammar and readability

Claude can act as a proofreader. Paste a paragraph and ask for a “readability‑grade‑8 rewrite”.

Rewrite the following paragraph for an 8th‑grade reading level, keep the meaning and proper nouns.

5.2 Consistency checks

Ask Claude to list all character names and verify that each appears in the draft.

List every unique character name in the text and note any spelling inconsistencies.

5.3 Feedback loop

Iterate: after each edit, request a “tighten” pass that reduces word count by 10% without losing meaning.

6. Integrating Claude with writing software

6.1 Python script for API calls

The script below sends selected text to Claude and returns the response. Save as claude_edit.py.

import os, sys, json, requests

API_KEY = os.getenv("CLAUDE_API_KEY")
ENDPOINT = "https://api.anthropic.com/v1/complete"

def call_claude(prompt, max_tokens=1500):
    headers = {
        "x-api-key": API_KEY,
        "Content-Type": "application/json"
    }
    data = {
        "model":"claude-2.1",
        "prompt":prompt,
        "max_tokens_to_sample":max_tokens,
        "temperature":0.7
    }
    response = requests.post(ENDPOINT, headers=headers, json=data)
    response.raise_for_status()
    return response.json()["completion"]

if __name__=="__main__":
    if len(sys.argv)<2:
        print("Usage: python claude_edit.py 'Your prompt'")
        sys.exit(1)
    prompt = sys.argv[1]
    print(call_claude(prompt))

6.2 Using the script in Scrivener

1. Open Scrivener > Preferences > Editing > External Editors.
2. Add a new editor, point to python as the executable and claude_edit.py %s as the argument.
3. Select text, right‑click → “External Editor → Claude Edit”.

6.3 Quick tip for Microsoft Word

Use the built‑in “Macro” feature to call the same script via PowerShell. Assign the macro to a toolbar button for one‑click editing.

7. Frequently asked questions

Do I need a paid Claude account to write longer drafts?

Yes. The free tier caps at 5,000 tokens per request, which is enough for short prompts. For drafts longer than 2,000 words you need the Pro plan, which offers up to 100,000 tokens per request.

Can Claude generate citations in MLA format?

Claude can format citations in MLA, APA, or Chicago if you give a clear instruction. It does not verify sources, so you must double‑check each reference.

Is Claude safe for confidential manuscript material?

Claude stores data for 30 days for debugging, but the content is not used to train the model. For highly confidential work, use the Enterprise plan with data‑in‑flight encryption and no logging.

How does Claude compare to ChatGPT for outlining a novel?

Claude generally produces longer, more nuanced outlines with fewer repetitions. In side‑by‑side tests, Claude scored 8.2/10 for depth versus ChatGPT’s 7.1/10.

Can I integrate Claude into my Scrivener workflow?

Yes. Use Claude’s REST API with a simple Python script that sends selected text to Claude and returns the edited version. The script can be called from Scrivener’s custom external editor feature.

Using Claude as a writing assistant can speed up brainstorming, structuring, drafting, and polishing. Follow the steps above, experiment with prompts, and you’ll see faster, clearer prose.