Build a Writing Assistant
Train a model that writes in your voice and style.
This guide walks through creating a fine-tuned model that matches your writing style — for blog posts, emails, documentation, or any other content.
What you'll need
- 10+ writing samples in your voice (blog posts, essays, emails, articles)
- Samples should be text you actually wrote — not edited by someone else
- More variety is better (different topics, same style)
Step by step
Gather your writing
Collect samples that represent how you want the model to write. Good sources:
- Blog posts — export from your CMS or copy-paste into
.mdfiles - Emails — export significant emails (not one-liners)
- Documents — reports, proposals, memos
- Social posts — longer-form threads or posts
Aim for at least 10 samples, ideally 50+. Each sample should be a meaningful piece of writing (a few paragraphs minimum).
Prepare your files
Save your writing samples as .txt or .md files. One sample per file, or combine them into a single file with clear separators.
You can also use a single .jsonl file:
{"text": "Your first writing sample goes here..."}
{"text": "Your second writing sample goes here..."}
{"text": "Your third writing sample goes here..."}Don't worry about cleaning — Commissioned handles formatting.
Create the fine-tune
- Upload your files to Commissioned
- Use a specific description:
"Create a writing assistant that matches my personal writing style. I tend to write in a [casual/formal/technical] tone with [short/long] sentences. The model should be able to draft blog posts, emails, and documentation in my voice."
- Select a base model:
- GPT-4.1 for the best writing quality
- GPT-4.1 Mini for a faster, lighter option
Test and iterate
Once the model is ready, test it with prompts like:
- "Write a blog post introduction about [topic]"
- "Draft an email to a client about [subject]"
- "Rewrite this paragraph in my style: [paste text]"
Pay attention to:
- Does it match your tone? (formal vs casual)
- Does it use similar sentence structures?
- Does it capture your vocabulary and phrasing?
If something's off, add more samples that demonstrate the missing pattern and create a new fine-tune.
Tips for better results
- Include a variety of topics — this helps the model learn your style independent of subject matter
- Include both short and long pieces — the model learns your natural rhythm at different lengths
- Remove other people's writing — if a document was co-authored, include only your sections
- Be patient with iteration — fine-tuning style typically takes 2–3 rounds to get right
A fine-tuned writing assistant is not a ghostwriter that perfectly mimics you. It's a tool that drafts in your general style, which you then edit and refine. Think of it as a first-draft accelerator.