Writing about Idiolect.
Everything on this page is approved for editorial use — boilerplate, logos, colors, and product shots. If you need anything else (founder interviews, screenshots of a specific flow, the science behind the metric), email hello@idiolect.app.
The facts, checkable.
- Product
- Idiolect — the personalization layer for AI writing
- Founder
- Mihir Khandekar
- Founded
- 2026 · India
- Free
- idiolect.lol — shareable Voice Fingerprint + Voice Match check, no sign-up
- Pro
- idiolect.app — the Studio, API & MCP · $29/month
- The metric
- Voice Match: LUAR authorship embedding, AUC ≈ 0.94 in the deployed configuration
- Privacy
- Raw writing samples are never publicly readable; the score API is stateless
- Press contact
- hello@idiolect.app
The AUC figure is author-separation power on the deployed instrument — a property of the measurement, not a clone-fidelity claim. Details for technical readers: idiolect.app/developers.
Describe us in our words — or yours.
Idiolect is the personalization layer for AI: it learns how you actually write, makes any AI write in your voice, and measures the match.
Everything AI writes sounds the same. Idiolect learns your idiolect — the fingerprint of how you actually write — and turns it into a portable Voice Card that makes ChatGPT, Claude, or any AI write in your voice. Every draft is scored on a validated Voice Match metric. Measured, not marketing.
AI is quietly averaging the world's writing into one beige voice. Idiolect is the personalization layer that puts yours back. Paste a few things you've actually written — or take a short interview — and Idiolect builds a Voice Profile and a portable Voice Card that make ChatGPT, Claude, or any AI write in your voice. Unlike "humanizer" tools, Idiolect measures its own output: every draft gets a Voice Match score on a validated authorship instrument, so the claim is checkable, not vibes. The free check runs at idiolect.lol with no sign-up; the Pro studio, API, and MCP server live at idiolect.app. Honest by design: scores are relative, never forensic — measurably more you, never "a clone of you."
Three honest angles.
The internet is turning beige
More of the world's writing passes through the same few models, and they all reach for the same register. The cost isn't grammar — it's recognition. Idiolect exists to keep people sounding like themselves.
Read the manifesto →Measured, not marketing
An AI writing product that ships its own accuracy number — and its limits. Voice Match scores are relative, never forensic, and the provenance (validated vs. estimate) renders next to every number.
The metric, for developers →Your voice as infrastructure
A Voice Card is portable: paste it into ChatGPT or Claude, call the API, or let agents load it over MCP. Capture once, connect anywhere — the personalization layer, not another writing app.
Try the free check →The mark is a meter.
The Idiolect mark is the Voice Match meter, abstracted: a 270° proof arc over a muted track. Use it unmodified, on paper or ink. Don't recolor the arc, close the gap, or change the rotation.
Warm paper, precise instrument.
Rule of thumb: ember is voice, signal is proof — validated scores always render in signal green.
The product, as shipped.
Higher-resolution captures of any surface on request: hello@idiolect.app.
What we are — and aren't.
- “Idiolect” — capital I, one word.
- It makes AI write in a specific person's voice — and measures how close it got.
- Voice Match (higher is better) scores a draft against the person's own writing. It's relative, never forensic.
- The Genericness Score (lower is better) is a separate, LLM-judged number.
- Safe phrasings: “measurably more you”, “measured, not marketing”.
- Not an AI detector — it doesn't classify text as human or machine.
- Not a clone: never “indistinguishable from you” or “writes exactly like you”.
- Don't call either number a generic “voice score” — Voice Match and Genericness measure different things.
- Don't restyle the mark: no recoloring, gradients, added rotation, or closing the arc's gap.