What happened
A developer figured out how to clone your writing style with AI without touching a GPU, without renting cloud servers, and without the usual month-long fine-tuning process. Instead of training a custom model on someone's old posts, he built two lightweight "skills" — small instruction packages — that plug into everyday AI coding assistants like Claude Code or Codex. One skill reads a batch of a person's existing writing and measures dozens of habits: how long their sentences run, how often they use dashes versus commas, whether they talk directly to the reader, how many emoji they drop in, and more. The second skill takes those measurements and rewrites any bland AI-generated draft so it sounds like that person wrote it.
The results are measurable, not vibes-based. A generic AI-written text was compared against a large Telegram channel's usual style across 31 separate writing habits. Before the rewrite, it differed from the channel's normal style on 14 of those 31 markers. After running it through the two skills, only 2 markers were still off — and real posts written by the channel's own author naturally land 2 to 5 markers off from their own average, so 2 is well within normal human range. The whole process takes one evening, and because the style rules live in a plain text file rather than inside a trained model, the same profile works with any AI model going forward.
What it means for you
You don't need to run a media company or manage a brand's Telegram channel for this to matter. Fine-tuning has always been the thing people mention and then don't do, because it needs a dataset, cloud GPU time, and technical setup most people don't have. A text-file style profile removes all three barriers. Here's what that looks like in practice.
Home
Writing a heartfelt letter, a eulogy, or a long birthday message and want it to actually sound like you instead of like a chatbot? Feed the AI a handful of your old messages or emails first, and ask it to match your tone before it drafts anything new.
Work
If you're ghostwriting for a manager or executive who needs to sound consistent across emails, LinkedIn posts, and internal memos, you can build one style profile from their past writing and reuse it for every future piece, instead of guessing at their voice each time.
Business
Brand voice consistency is usually the first thing that breaks when a company scales content production — different freelancers, different tones, no unified voice. A saved style profile means every new writer or every new AI-generated draft gets checked against the same measured rules instead of a vague style guide nobody reads.
Study
Students preparing personal statements, essays, or long-form reports can use their own earlier papers to keep their natural voice intact even when AI helps structure or tighten the draft, instead of ending up with text that reads suspiciously polished.
Creativity
Writers experimenting with a recurring character, a pen name, or a fictional narrator can lock in that voice once and apply it across chapters or episodes without re-explaining the tone every single time.
Income
Freelance copywriters and ghostwriters can turn this into a service: measure a client's existing content once, sell a reusable "voice profile" as a deliverable, and charge for every additional piece written in that voice — without spending money on model training for each client.
How to try it right now
You don't need a technical background to test the free path first, though the exact skill-based workflow described above currently requires an AI coding assistant such as Claude Code or Codex, since the two skills are built to run inside those tools.
1. Collect a sample. Gather 20–50 pieces of writing from the person or brand whose style you want — social posts, emails, blog entries. More consistent samples give more reliable measurements.
2. Ask an AI assistant to build a style profile. Instead of vague instructions like "write energetically," have it extract concrete patterns: average sentence length and its normal range, punctuation habits, use of direct reader address, paragraph length, and similar countable traits. Save this as a simple text file so it's reusable.
3. Apply the profile to any draft. Take a plain, generic AI-written text and instruct the assistant to rewrite it strictly according to the saved profile's ranges, not just its overall "vibe."
4. Check the result against a few real samples. If your rewritten text differs from the real writing on more markers than the real writing differs from itself, refine the profile rather than trusting it blindly.
If you don't use a coding assistant and just want quick AI writing or rewriting help without any setup, you can start with free, no-signup tools like Mykreatool to draft or clean up text before applying a style pass on top of it.
Upsides and what changes
The biggest shift is cost and portability. Fine-tuning a model on a writer's voice has real, documented prices: one public case used 51,000 examples, a rented 40GB A100 GPU, and 10.5 hours of training — and the author still described the output as unstable. Fine-tuning through an API is cheaper but still not free: roughly $100–300 for 5,000 examples and around $700 for 24,000, on top of GPU rental starting near 200 rubles (about $2) per hour if you train it yourself. A text-based style profile costs none of that, takes an evening instead of days, and — because it's just a document, not model weights — moves instantly to a new AI model whenever a better one comes out. It's also transparent: you can literally open the file and see which rule produced which behavior, which a trained model never lets you do.
Limitations
This approach isn't magic and it isn't fine-tuning's equal in every way. A research paper covering more than 400 real writers and over 40,000 AI-generated samples found that simply showing a model more examples of someone's writing barely improves style matching on its own — the gain comes from explicitly measuring and stating the patterns, not from volume. The method also depends entirely on having enough clean writing samples to measure from; a person with only a handful of short posts won't give reliable ranges. And it still won't produce a perfect match — even the original author's own real posts land a few markers outside their personal average, so a small gap between the rewritten text and the "true" style is normal and expected, not a bug to chase away completely.
Conclusion and one action for today
Cloning a writing style with AI no longer requires a GPU, a dataset, or a training budget — a measured, reusable text profile beats a costly one-off fine-tune for most real use cases. Today's action: pick one person or brand voice you need to match, pull together 20–30 of their past texts, and ask an AI assistant to turn those into a concrete style profile before you write anything new in their voice.
Comments 0