How to set up a custom AI writing style in under 10 minutes

2026-04-29·5 min read

Quick take

Setting up a custom AI writing style takes about 10 minutes. Gather your writing samples, upload them, and generate. This guide walks through each step with specific details on what works and what doesn't.

What you need before starting

You need 10-15 writing samples that represent your voice. These can be blog posts, newsletter issues, long emails, reports, or any substantial writing. Each sample should be at least 300 words. Longer is better.

The samples should be your own writing, not heavily edited by someone else. If a piece went through three rounds of corporate review, it's probably more "committee voice" than your voice. Pick pieces where you had editorial control.

Step 1: gather your samples (5 minutes)

Pull from wherever you publish or write regularly. Good sources:

Blog posts on your personal site or Medium
Newsletter archives from Substack, ConvertKit, or Mailchimp
Long-form social posts from LinkedIn or Twitter threads
Internal documents or memos you wrote
Email drafts from your sent folder (longer ones)

Copy the text into plain text files or have it ready to paste. You don't need perfect formatting. The voice analysis works on the writing patterns, not the layout.

Step 2: upload to UmanWrite (2 minutes)

Go to the voice training page. You'll see an upload area for your writing samples. Paste or upload each sample. The system accepts plain text, Markdown, and direct paste.

Once your samples are in, click "Analyze." The system processes your writing in about 30 seconds. It measures sentence length patterns, vocabulary preferences, paragraph structure, transition habits, and formality level.

You'll see a summary of what it found. Review this to make sure it matches your self-perception. If the system says you write with an average sentence length of 22 words and you expected shorter, you might want to add some of your more casual writing to balance the profile.

Step 3: generate a test piece (2 minutes)

Write a prompt for a topic you've covered before. Something like "write a 300-word section about [your expertise area]." Generate it with your voice profile active.

Compare the output to your original writing on the same topic. The sentence rhythm, vocabulary, and paragraph structure should feel familiar. If something's off, note what specifically doesn't match.

Step 4: refine if needed (1 minute)

If the output is too formal, add more casual samples. If it's too loose, add polished pieces. If it's missing a pattern you care about (like your habit of using rhetorical questions), add samples that demonstrate that pattern heavily.

Most users get good results on the first try. Light refinement gets you the rest of the way.

Step 5: humanize and verify

Run your generated content through the humanizer to catch any remaining AI patterns. Then check it with the AI detector. Voice-trained and humanized content typically scores under 15% AI on most detectors.

This three-step pipeline, generate with voice, humanize, detect, gives you content that sounds like you and passes scrutiny. For more on the humanization process, see how to humanize AI text.

Tips for better results

Quality over quantity in samples

Fifteen mediocre samples produce worse results than eight strong ones. Pick writing you're proud of. The AI will replicate whatever patterns you feed it, including bad habits from rushed first drafts.

Keep samples in one register

If you want a professional blog voice, train on professional blog posts. Don't mix in casual text messages or academic papers unless you want the AI to split the difference between all of them.

Update quarterly

Your writing evolves. Swap in fresh samples every few months so the profile stays current. This takes five minutes and keeps the output aligned with how you write now, not how you wrote a year ago.

FAQ

How much does custom voice setup cost?

Voice training is included in UmanWrite's paid plans. Check pricing for details on how many voice profiles each plan supports and word generation limits.

Can I have multiple custom styles?

Yes. You might want one profile for blog content and another for client communications. Each profile trains independently on its own set of samples.

What if I don't have enough writing samples?

Start with what you have, even if it's only 5-7 pieces. The results will be rougher than with 15 samples, but still significantly better than generic AI output. Add more samples as you create new writing.

Does this work for non-English writing?

Currently, UmanWrite's voice training is optimized for English. The pattern analysis relies on English-language structures. Other languages are on the roadmap.