A ChatGPT detector helps you answer a simple question: was this text written by a person, or generated with ChatGPT or another chatbot? Editors receive articles from freelancers, hiring managers read cover letters, website owners pay for content, and readers come across posts that feel strangely smooth. In all of these cases the detector above gives you a second opinion in a minute, before you decide whether to publish, pay or reply.
What the detector looks at
Chatbots write by choosing a likely next word again and again. The result is fluent, but it tends to be predictable: sentences of similar length, safe generic wording, neat transitions and a structure that sounds like a summary of everything and nothing in particular. The detector analyses the text fragment by fragment and estimates how likely each part is to have been produced this way. It does not search for the text on the web, and it cannot tell which tool, if any, was used.
How to check a text
- Paste the text into the field above or upload a .txt, .docx, .pdf or .odt file. The text needs at least 250 characters, and longer texts give the detector more to work with.
- Remove anything that is not the author's own writing, such as quotations, a brief copied into the document or a signature block.
- Start the check and wait for the result.
- Read the overall verdict first, then go through the highlighted fragments one by one.
How to read the result
The overall verdict has four levels:
- AI generation detected: a large part of the text shows the patterns typical of chatbot output.
- Uncertain result: clear generation is not confirmed, but a noticeable share of the text looks suspicious.
- Some suspicious fragments found: the text as a whole reads as human, with a few passages that look formulaic.
- No significant AI signs found: nothing in the text stands out as typical machine writing.
Below the verdict, each fragment is marked as likely AI, suspicious or human, and clicking it shows the confidence value. Pay most attention to long runs of likely AI fragments rather than to one isolated sentence. A single highlighted line in an otherwise human text usually points to a cliché, not to an answer copied from a chatbot.
When the result is hard to interpret
Many texts are neither fully human nor fully generated. Someone may draft with ChatGPT and then rewrite half of it, or write by hand and ask a chatbot to polish the grammar. Texts translated with automatic tools, very formal business writing and lists of short instructions can also look machine-like. In these cases expect an uncertain verdict or a mix of colours, and judge the text by its content as well: does it contain specific facts, personal experience, names, numbers and opinions that a generic answer would not have?
What to do next
- If someone else wrote the text, treat the result as a reason to ask questions, not as an accusation. Ask how the text was written, request a draft or notes, or compare it with earlier work by the same author.
- If the text is yours and some parts sound mechanical, rewrite them in your own voice: add concrete details, vary the length of your sentences and cut filler phrases. The result page also links to a humanizer tool that suggests more natural wording.
- If you need a record, download the PDF report with the verdict and the highlighted fragments.
Keep in mind that the detector estimates probabilities. It can mark a human text as suspicious and miss a generated one, especially after heavy editing, so a result on its own is never proof. The check is free without registration within a daily limit, and signed-in users and PRO subscribers get larger limits.