AI that writes like you
By Jack Stovell X · Instagram · LinkedIn · YouTube · TikTok · published 2026-08-03 · updated 2026-11-17
Measure your voice, then draft against the measurements. Pasting "write like me" and a sample into a chat gives you a loose impression that fades fast. A profile built from writing you have already published gives the model something fixed to follow, and a score tells you whether the draft held to it.
Here's a free check you can run before anything else. Paste something you wrote and an AI draft, and see how far the draft sits from you, habit by habit.
The trouble with "write like me" prompts
You've probably tried it. You paste a few paragraphs of your own writing, type "write like me", and get back something that is polite, tidy and nobody in particular. Fair enough, that's what the prompt asked for.
So why does it fail? Two reasons. First, samples in a prompt are read once. By the third paragraph of the draft, the model has drifted back to its defaults, because nothing is holding it to what it saw. Second, you can't put most of your habits into words. Ask yourself how long your sentences run, how often you use a contraction, whether you ever touch a semicolon. Most people can't say.
What you can put into words is a description of your style. "Short sentences, dry, conclusion first" is something you can say, and something a model can follow. But the description only works if it is accurate, and a hunch about your own writing is usually a flattering one.
That's why the fix is measurement. Count what you actually do, write it down as constraints, and hold every draft against them. The rest of this guide shows you how to do that by hand, and where we fit if you'd rather not.
What measuring a voice means
Measuring a voice means counting habits in writing you already have. We track 45 attributes in 8 groups: lexical, syntactic, tone and register, rhetorical, punctuation and format, function words, content patterns, and quirks and cadence. You can browse all of them on the page covering the 45 attributes.
Three are worth doing by hand today.
Sentence rhythm and its spread. Take a page of your writing and count the words in each sentence. The average matters less than the spread. Some writers sit near twenty words every time; others jump from four to forty. That jump is rhythm, and a model left alone flattens it.
Contraction rate and punctuation. Count the contractions per thousand words, then check your commas and whether you use semicolons at all. Some writers never do, and that absence is as much a signature as heavy use. There's a page on the semicolon attribute if you want the detail.
Humor register and argument structure. Are you dry, warm, or straight? Do you state the conclusion first, or build up to it? These are harder to count, but you can read three pieces and answer honestly.
If you want the wider background, the post on stylometric analysis covers where this discipline comes from.
Build your profile, step by step
By hand, the job is tedious but possible. With us, it takes a few minutes. Either way the steps are the same.
- Choose three or more pieces nobody edited for you. This matters more than people expect. An editor's fingerprints are on anything that went through one, and you want yours. We recommend about 3,000 words in total, though one sample of 50 or more words is the minimum.
- Paste or upload them. You can add up to 20 samples.
- Wait. Building the profile usually takes under a minute.
- Read what comes back: a written summary, every attribute, and a confidence score.
- If it reads wrong, add more samples and re-analyze.
Step four is the one people skip. Don't. The summary is a claim about you, and you're the only person who can check it. If it says you hedge a lot and you know you don't, that's a signal your samples were thin or unrepresentative, or that one piece was an outlier.
Doing it by hand, the equivalent is a one-page note: average sentence length, how wide the spread runs, contraction habit, punctuation you use and avoid, humor, argument shape. Keep it short enough that you'd actually check a draft against it.
A thin profile is a sketch. A fuller one is a likeness. That's the whole reason to feed it more.
Draft, score, and read what moved
Now brief a draft. You give it a topic and a content type (there are 16 to choose from), pick British or American English, and the profile acts as the constraint while the draft is written.
Then comes the part a prompt can never give you: a score. Every draft is scored from 0 to 1 against your profile, which we call the voice match score. On-voice drafts typically land between 0.85 and 0.95. Partial matches sit between 0.5 and 0.8. A clearly different voice comes in under 0.45.
The number is only half of it. Each score returns deltas: for every habit, the draft's value beside your profile's. Here's an example, invented for illustration: your profile shows an average sentence length of 12 words, and the draft came back at 19. Now you know exactly what drifted, and you can see it without guessing.
You can run the same check by hand. Take your note from the last section, count the same things in the draft, and put the two columns side by side. It takes longer, but the principle is identical.
Scoring any text of 20 or more words against a profile is free. That includes text you didn't generate with us, so you can score a draft from any tool you like before you decide whether it sounds like you.
When a draft misses: polish and your own edits
Drafts miss. A score of 0.6 is information.
The first option is polish. It revises a draft toward a target voice match score (0.9 by default) over up to three measured passes, then returns the best version with before and after scores. It costs 1 credit for a draft up to 5,000 words, and it's free when the draft already passes. One thing worth knowing: polish changes how a thing is said, never what is said. It adds no facts and no first-person stories. So it can fix your rhythm, but it can't invent an anecdote you never lived, and you wouldn't want it to.
The second option is your own hand. Edit the draft until it reads like you, then save the edit. Edits you save feed the profile's learning, so the next draft starts from a slightly better place.
By hand, the habit is simple. When you fix a sentence, ask which measurable habit you just restored. Too long? Too formal? Too many hedges? Then add that to your note.
Use it inside ChatGPT, Claude and the rest
You don't have to leave the tools you already use. We make ChatGPT, Claude and the rest better at being you; we work with them.
The profile is read live through an MCP server (41 tools) and a REST API, both on every plan including Free. You can connect it in Claude, in ChatGPT (where custom connectors are on your ChatGPT plan), Cursor, Gemini CLI or any MCP client. The profile is read live each time, so there's nothing to export or paste.
Five tool names matter most:
- `get_profile` reads the measured profile.
- `generate_content` drafts with it (1 credit).
- `check_voice_match` scores any text against it (free, up to 200 scores an hour).
- `polish` revises toward the target.
- `save_edit` keeps your final edit (free).
One catch: the assistant only scores a draft when it calls the tool, so ask it to. "Draft this, then check the voice match" is enough. If you want the longer walkthrough for that setup, there's a full guide on making ChatGPT sound like you.
A voice is also what you have already said
Style is how you write. A voice is also what you've already said, and a draft that repeats your last post sounds like you in the worst way.
That's the job of the Voice Brain, which is free on every plan. It keeps the ideas, claims, stories and frameworks your writing has already covered, each with its exact quote. Before you write, a novelty check (`check_novelty`) scores your brief against that memory and tells you whether it repeats, builds on, or is new, and proposes angles. Ask `what_next` and it proposes five angles.
Once a voice has 10 indexed pieces, new drafts build on the memory and don't re-argue points you've already made. You can edit or delete any memory, so it stays yours. There's more on the Voice Brain if you want to see how it fits together.
By hand, the equivalent is a running list of your own claims. Keep one line per post. Before the next draft, check the list.
What it will not do
Let's be plain about the limits.
The score is a style check only. It never checks facts, so a person still reads every draft. A draft can score 0.93 and still be wrong about something, and the score won't tell you.
A thin profile is a sketch. One short sample will give you something, but it won't hold up across different content types. Feed it more.
We don't publish or schedule anything. A draft sits where you left it until you decide what to do with it.
And we promise nothing about AI detectors. Our aim is a draft that sounds like you, which is a different goal from beating a classifier. Those tools change, and we won't tell you otherwise.
On privacy: we never train AI models on your writing, and our terms with the AI providers forbid them from training on it. Your writing is measured, never trained on. It's stored in the EU (Ireland), and you can download or delete everything from Settings. We're a UK company.
Free to see it, paid to use it
Seeing your profile costs nothing. Using it for drafting is where credits come in. A draft up to 5,000 words is 1 credit; longer drafts are 2 or 3, up to 15,000 words. A credit is taken only after the draft is delivered. Scoring, outlines and the novelty check are free.
- Free: one voice profile, one full analysis of all 45 attributes, the API and MCP server, Voice Brain, and one free draft for the life of the account, no card.
- Writer, $17 a month: 1 profile, 40 credits a month.
- Operator, $40 a month: 3 profiles, 75 credits.
- Studio, $134 a month: 10 profiles, 400 credits, 3 seats.
- Agency, $405 a month: unlimited profiles, 1,500 credits, 10 seats.
Questions
How much of my writing do I need?
One sample of 50 or more words is the minimum, but that gives you a sketch. We recommend three or more pieces and about 3,000 words in total, and you can add up to 20 samples. Use writing nobody edited for you, because an editor's habits will blur your own.
Is a stored profile better than pasting samples into a prompt?
Yes, for two reasons. A prompt is read once and fades as the draft goes on, while a stored profile is the constraint every draft is written against. It also gives you a score and per-habit deltas, so you can see where a draft drifted instead of guessing.
Does it work when my voice changes?
It can, as long as you update it. Add newer samples and re-analyze, and the profile reflects how you write now. Edits you save also feed the profile's learning. If your voice has shifted a lot, build from recent pieces rather than old ones.
What does it cost to see my profile?
Nothing. The Free plan includes one voice profile and one full analysis of all 45 attributes, with no card. Scoring any text of 20 or more words against a profile is also free. You only spend credits when you ask for a draft or a polish.
Does my writing train a model?
No. We never train AI models on your writing, and our terms with the AI providers forbid them from training on it. Your writing is measured, never trained on. It's stored in the EU (Ireland), and you can download or delete everything from Settings.
Can I use it with ChatGPT or Claude?
Yes. The profile is read live through an MCP server and a REST API, both on every plan including Free. You can connect it in Claude, in ChatGPT (where custom connectors are on your ChatGPT plan), Cursor, Gemini CLI or any MCP client. Ask the assistant to call the scoring tool, or it won't.
Can it hold more than one voice?
Yes, depending on the plan. Free includes one voice profile, Writer includes 1, Operator 3, Studio 10, and Agency has unlimited profiles. Each profile is measured separately from its own samples, so a personal voice and a client's voice don't blur together.