Check if text was written by AI
Paste text or upload a document. Get a clear result with sentence-level signals.
0 words · Add at least 50 words for a more meaningful analysis.
AI detection is probabilistic and can be wrong. Text is sent to our detection provider for this check only and isn’t stored by us. How your text is handled
- Sentence-level marks
- Limits stated up front
- Text goes to Pangram to run the check. Privacy
How to read your result
You get a headline, a share, a confidence level and the text itself with marked passages. The headline is the quick answer; the marked passages are the evidence behind it.
- Mostly human-like means few passages looked like generated writing.
- Mixed signals means some passages looked human and others looked AI-like or AI-assisted. Edited AI drafts often land here.
- Likely AI-assisted means a large part of the text looked AI-like.
- Not enough text means the sample was too short to judge fairly.
“Flagged as AI-like” is the share of your text that was labelled AI-like. It is not a probability that the whole text was written by AI. More on reading the percentage.
Why sentence-level signals matter
A single score hides where the signal comes from. A document can be mostly the writer’s own words with one generated paragraph, or the other way round. Seeing which passages were marked lets you judge them directly: read them, compare them with the rest, and decide whether the pattern means anything in context.
The detector gives a label and a confidence for each passage. It does not explain its reasoning, and we do not invent reasons it did not give.
What AI detection can and cannot tell you
Detection looks at the writing, not the writer. It cannot say who wrote a text or how it was produced.
Formal, highly polished or non-native writing can look generated, and AI text that has been heavily edited can look human. Translated text and very short samples are less reliable still. Treat a result as a reason to look more closely, never as a verdict. Read the methodology for the full list of limits.
Check the kind of text you have
The checker is the same everywhere. These pages explain how to read results for specific situations.
How it works
Your text is cleaned up, then sent to Pangram’s detection model, which labels passages as AI-generated, AI-assisted or human-written. We turn those labels into the headline, share and highlighted text you see. Nothing is stored by us. See how AI detection works or the privacy details.
Frequently asked questions
Can this tool prove a text was written by AI?
No. It reports patterns that are commonly associated with AI-generated writing. That is a signal to weigh alongside context, drafts and conversation, not proof of authorship.
How much text do I need?
At least 50 words, and the more the better. Short passages carry little evidence, so results for them are weaker. You can check up to 5,000 words at a time.
What does "Flagged as AI-like" mean?
It is the share of your text whose passages were labelled AI-like. It is not the probability that the whole text was written by AI.
Can I upload a document?
Yes. TXT, DOCX and PDF files up to 10 MB work. Scanned or image-only PDFs contain no readable text, so paste the text instead.
What happens to my text?
We do not store it, and our logs never include it. It is sent to our detection provider, Pangram, to run the check, and Pangram’s own terms govern what it does with it. The privacy page explains the full data flow.
Does it work in languages other than English?
Detection is most reliable for English. Results for other languages, or for translated text, may be less accurate.