AI detector for German and English
Our Frontier AI detectors recognise AI text with an 87.8–94.4% detection rate* – free. One click, no account, detects AI-generated German and English texts.
Detect AI-generated text – instant, anonymous and privacy-compliant. German and English text, right in your browser, no sign-up.
Detect AI-generated texts
Our Frontier AI detectors recognise AI text with an 87.8–94.4% detection rate* – free. One click, no account, detects AI-generated German and English texts.
No US cloud providers. Your texts never leave the German infrastructure.
Detects text from ChatGPT, Gemini, Claude and others. The AI detection is updated regularly.
How does it work?
When you analyse hundreds of thousands of documents, you realise you can’t simply count emojis, em dashes or typical ChatGPT phrases:
No AI text detector is perfect. That’s why we keep retraining them – also to detect disguised or altered AI text. PlagiatGPT analyses over 700 features. A neural network we trained ourselves classifies them. This gives us leading benchmark figures for our calibrations, with detection rates of 87.8–94.4% and error rates of under 1%.*
No – discarded immediately, fully anonymous.
The free Quick Analysis uses the same detector as the Deep Analysis and delivers the same AI score - just as a rounded value from 1 to 10, without reasoning or file forensics. For a detailed picture, there is the Deep Analysis.
Yes, completely free and without signing up: up to 200 words per day. With a free account it is 5,100 characters a day (approx. 600 words) – plus 5 pages of Deep Analysis as a gift. For longer texts, whole documents, or up to 40 papers at once, use the Deep Analysis.
The free Quick Analysis without an account is calibrated for general texts - such as web copy, blog posts and everyday writing. With a free account you can set the type of text to "Education / Academic" for the Quick Analysis too, which uses the model for school, term papers and theses.
An overall score - a probability of whether the text is AI-generated. Without an account you see it as a rounded level from 1 to 10; with a free account as an exact percentage. A high level (8 to 10, or 85 to 100 % in an account) means the text was very likely generated by an AI model. A low level (1 to 3, or 0 to 30 % in an account) means the text was very unlikely to have been generated by an AI model, or was heavily altered.
The Quick Analysis and its score are enough for a first impression. However, when you need something you can act on, for example a suspected-cheating case, grading, or before publishing, use the Deep Analysis. It lets you not only analyse sentence by sentence but also upload files such as PDF, DOCX, TXT and Markdown with higher quotas.
Create a free account and analyse up to 600 words a day! Signed-in users see exact percentages instead of a 1-to-10 scale.
* The detection rates come from the out-of-distribution benchmark of our AI detector models (V3.1, July 2026), measured on unseen, external datasets at under 1% false-positive rate. Depending on the text category, the detection rate is between 87.8% and 94.4% (German/general 87.8%, English/academic 89.9%, German/academic 90.1%, English/general 94.4%). Quick Analysis and Deep Analysis use the same detector, so the figures apply to both. Note: the success of AI detectors depends heavily on the category of the text. Deliberate obfuscation and alterations to AI-generated texts can reduce the reliability of the AI classification. Highly technical texts as well as purely mathematical, abstract and unstructured texts are excluded from AI analysis.