Voice test · no account
Does this text still sound like you?
Give Kaneme 3 texts that sound like you, then a new text to test, all in English or all in French. You get a probability from a named comparison, not an arbitrary score out of 100.
1 · Text to test
50 words minimum; 150 for a stronger measurement.
0 / 50 wordsAdd the references required for the verdict
2 · Your reference writing
Use texts you know are yours, in the same language as the one you test. Vary the context—a post, an email and a note—so the measurement sees an author rather than one format. At least 100 words each.
0 / 100 words
0 / 100 words
0 / 100 words
3 · Your question
Temporary processing
Your texts are processed temporarily by the server, without server storage. They are not saved, linked to an account, or reused. A local copy stays in this browser for up to 48 hours so you can resume.
0,989voice separation from generic AI
0,869your writing versus another person
The gap shows which question is harder. 1 is perfect; 0.5 is chance. The production reading is validated in English and in French, each with its own engine.
Nothing is stored
Your texts pass through the server only during the request because the measurement layer cannot run in a browser. No saved writing, account or reuse.
The right to abstain
Outside the validated language domain, outside covered registers, when the candidate overlaps the references or when the deep layer fails, Kaneme reports no probability and names the reason.
A signal, not proof
A probability cannot support an accusation. Edited machine text can make the upper range overstate certainty; this limitation is measured and published.
Why three texts instead of one?
Authorship is a comparison problem. With no reference, there is nothing to compare. Three texts of 100 words provide enough material to avoid measuring a single format. Vary the contexts so the system sees the person behind the writing.
The candidate needs at least 50 words. A longer candidate improves confidence. Samples and candidate must share a language, English or French: across two languages, the verdict abstains.
Read the protocol and observed limits on the method page, including what voice measurement covers, and what it does not.
You have the verdict. What an account adds is memory.
Without an account, you paste three samples every time. With one, Kaneme remembers the voice, measures each new text in one click and creates a shareable certificate.