See how AI detectors read your text.
Paste a text and see how likely an open detection model thinks it is that AI wrote it. Free with an account.
Your text
Log in to check a text. It is free: 30 checks a day.
About these scores
an estimate,
never a verdict
An open detection model estimates the probability that a language model wrote the text. A result of 70% AI means the model estimates a 70% chance that a language model wrote it, and a 30% chance that a person did. It is not the share of the text that was written by AI.
It is an estimate, not a verdict. Detectors make mistakes in both directions: they can flag plain human writing, writing by non-native speakers and formulaic text, and they can miss AI writing. Other detectors often give the same text a different score, and scores change as models change. No score proves who wrote a text, and we promise nothing about any detector.
Fakespot RoBERTa (fakespot-ai/roberta-base-ai-text-detection-v1, Apache-2.0). It runs on our own server. The exact model revision is shown next to each result.
It is scored and then deleted. We keep the scores, not the text, and never use checked text for training or evaluation.
Checks are free with an account: 30 a day, up to 2,000 words each. On a paid plan: 200 a day, up to 5,000 words. A check needs at least 80 words.
No. It is not for checking academic submissions, and a score is not a basis for accusing anyone of using AI. Candid Write is built for writers checking their own drafts.
Text reads like AI? Humanize it.
The Humanizer rewrites it in plain, natural language and keeps your figures, dates and quotes as you wrote them.