Skip to content

Sign in / Sign up
ParaTrace

Unlimited free credit top-ups, for a limited time. Running low? Ask for more credits and we top you up, as often as you need. This offer ends soon, so catch it while it lasts.

About ParaTrace

How detection works, and where it falls short

ParaTrace estimates whether an English text was written by a person or generated by a language model. It gives a fast, explainable second opinion on authorship, from classrooms to newsrooms.

How detection works

  1. Cleaning. Quotes, dashes and Markdown markers are normalised and invisible characters are removed.
  2. Splitting. The text is split into overlapping windows of 512 tokens.
  3. Scoring. Each window is scored by a transformer classifier that we designed and trained in-house.
  4. Labelling. Each sentence takes the label of the windows it lies in, and the verdict follows from how much of the text is labelled AI. It is AI, human or mixed.
  5. Verdict. A cut-off labels each window as AI-generated or human-written. A text is reported as “likely AI-generated”, “likely partly AI-generated” or “likely human-written”.

Operating points

Checks on this site always use strict. The developer API offers both.

Strict
Flags text as AI only when confident. About 1 in 100 human texts is wrongly flagged.
Lenient
Catches more AI text, but about 1 in 20 human texts is wrongly flagged.

Accuracy and limits

  • English only. Texts in other languages are not supported.
  • At strict, 0.8 % of held-out human texts (about 1 in 120) were wrongly flagged as AI in our tests. At lenient, about 5 %. Other kinds of writing can differ.
  • Persuasive essays written by adults are a known weak spot. About 1 in 5 (18 %) is wrongly flagged at strict.
  • Texts under 50 words give less reliable verdicts. In our tests 2.8 % of short human texts were wrongly flagged.
  • Long texts are scored in overlapping chunks and labelled part by part, so a text can come out partly AI-generated. Where the AI part starts and ends is approximate.
This is a statistical estimate, not proof. Don't use it as the sole basis for decisions about a person.

For the full results, including detection by AI developer, false alarms by kind of writing, evasion attacks and speed, read the benchmark report.

Privacy

Texts you submit are stored on our server, in your history, until you delete them. Don't submit confidential or personal data.

Your checks stay in your history, with their texts and verdicts, until you delete them. Deleting a check removes it from our server. Texts sent to the developer API are scored and not stored.

Your sign-in is kept only for this browser tab (in session storage) and is sent with each request. It is removed when you sign out or close the tab.

The privacy policy has the details, including analytics, how long we keep data and your rights.

Getting access

Sign up with your Google account and start at once, with 100 free credits.