Let AI draft. Keep your voice.
Personalect rewrites AI drafts the way you write. Give it two things you've written and it keeps everything the draft says, in your words and your rhythm instead of the model's.
Case study. An academic who answers questions online. Pessoa had never seen their writing before it read two of their posts. Pangram, the strictest AI detector, read the rewrite as human.
See all five writersStudents are adults and should be trusted to make their own decisions about attending class. In a lecture-based class, instructors shouldn't police students' reasons for missing class, as the idea of a "reasonable excuse" is meaningless. Attendance mandates are misguided and create unnecessary stress for both students and instructors. Dropping the requirement altogether would save time and energy. However, in a seminar, attendance may be more relevant. A better approach is to set a high limit on absences, such as 4 or 5, which treats students as adults and avoids bureaucratic conflicts. This policy allows for flexibility while still encouraging regular participation. The limit should be high enough to accommodate serious issues like illness or family emergencies. Instructors should avoid judging the validity of students' excuses and focus instead on creating a respectful and supportive learning environment.
Pessoa found that this change would better reflect your style. Is this correct or incorrect?
Thanks. Your answer helps tune your voice.
How It Works
Show Personalect how you write one time. After that, any AI draft is a click away from sounding like you, whether it's an email, a post or a report.
Writing Samples
Emails, posts, essays. Two pieces of about 300 words each are enough. Pick the kind of writing you want it to sound like.
Voice Analysis
How you open and close, how you hedge, the words you reach for and the length of your sentences. There's no training on you, so it works right away.
Rewriting
The facts, names and numbers stay. The voice becomes yours, and you can edit before you copy.
Availability: a web app, a Chrome extension for Gmail, Docs and LinkedIn, and inside Claude, Cursor and other AI tools that support MCP. One voice works in all three. Email and document integrations, an API and team controls come later.
Detection Results
Of 40 drafts written from someone's own notes, Pangram called none AI. Pangram is the AI detector built to catch rewritten text. We ran all 469 rewrites from our test set through it by hand. When the ideas are yours and the AI only drafted them, it almost never calls the result AI.
See the full resultsPangram's verdict on all 469 rewrites
Benchmark Comparison
We ran the published ways of rewriting in someone's style on the same 332 drafts. Prompting a large model with your samples keeps the meaning, but almost every draft still reads as AI. The best open style model gets past detectors only by breaking the text.
Editors like Grammarly push everyone toward one polished style. Brand-voice tools give a company one voice. Personalect writes as one person.
| Method | AI drafts still flagged | Meaning kept | Quality, of 5 |
|---|---|---|---|
| Pessoa | 35.5% | 0.75 | 4.51 |
| Few-shot prompting (STYLL), your samples as examples | 96.6% | 0.80 | 4.68 |
| Register-analysis prompting | 94.5% | 0.73 | 4.48 |
| TinyStyler, the best open style-transfer model | 49.3% | 0.40 | 1.28 |
| AI draft, unchanged | 97.8% | – | 4.66 |
Mean of six open AI detectors, each set to flag only 5% of real human writing. Prompting ran on a 30B model with the writer's samples. Meaning kept is the share of the draft's claims still in the rewrite; a writer's own text scores 0.75 by the same judge.
The Pessoa Model
Personalect runs on Pessoa, our own rewriting model, named for the poet who wrote as dozens of distinct authors. It was trained on whole documents, pairing AI drafts with the real writing they stood in for. It isn't a chat model with a prompt on top, and no outside AI service sees your text.
- Structure
- What you lead with, where the ask goes, how you close. Pessoa reorders the draft's own points to match.
- Style
- Your vocabulary, sentence length and rhythm, written fresh for every paragraph.
Percentile among the writers Pessoa learned from. In the rewrite: “I think many people…”, “(Especially if you're cooking to well-done.)”
Provenance. Every rewrite carries a watermark no reader can see. With our key, schools, publishers and platforms can check whether Pessoa wrote a text: four texts from one writer are enough to tell 98% of the time.
How the watermark worksFrequently Asked Questions
What does Personalect change in my draft?
The wording and the order. Pessoa rewrites the draft in your vocabulary and sentence rhythm, and rearranges its points the way you'd usually lay them out. What the draft says stays the same.
How much of my writing does it need?
Two pieces of about 300 words each is enough, and adding more barely changes the result. Pick the kind of writing you want modeled: work emails if you'll mostly rewrite emails, posts if you'll mostly write posts.
Will the rewrite get past AI detectors?
Often, but not always. In our tests six open detectors flagged 40% of rewritten AI drafts, down from 98%, and Pangram read 58% of rewrites as fully human. It works best when the ideas are yours and the AI only drafted them; when the AI came up with the ideas too, Pangram still flagged 85%. Pessoa was never trained to beat a detector. It gets there by sounding like you.
Can anyone tell that Pessoa wrote it?
Only with our key. Every rewrite carries an invisible watermark, so schools, publishers and platforms we work with can check whether a text came from Pessoa. Readers can't see it, and it doesn't change the quality of the writing or your voice. More on the watermark.
Could a rewrite change what I meant?
Pessoa writes several versions and keeps the one closest in meaning to your draft. It also shows what it changed, and you can edit the rewrite before you copy it, so nothing goes out unchecked.
Does it get better the more I use it?
Yes. When you mark a highlighted change as correct or incorrect, or edit a rewrite yourself, that feedback helps tune your voice.
What happens to the writing I give it?
It's used to build your voice, and only you can use that voice. You can delete a voice whenever you like. Our privacy policy lists exactly what we keep.
Can I build a voice from someone else's writing?
Only if it's yours or you have their permission. A voice is meant to sound like you, not to imitate another person. Pessoa is also built for AI drafts: rewriting one person's own writing into someone else's voice barely changes it.
Are these results independent?
Not yet. They're our own evaluations, run on writers Pessoa never trained on, with the bar for success set before each test. They haven't been peer reviewed. See the full results.
When can I use it, and what will it cost?
We're opening in stages, starting with the waitlist. Early users don't pay to get started. Plans come later, and people on the waitlist will hear about them first.