No, and we are explicit about this. AI detection is an indirect measurement — a strong indicator, not legal proof. A high score provides a solid, evidence-based starting point for a conversation with the student, but sanctions should always be based on the full integrity report, that conversation, and your institution's established process. For stronger evidence, writing process data from the Version Scan or Writing Journal is the decisive complement.

Mentafy's AI detector is model-agnostic. It does not try to identify which specific tool — ChatGPT, Gemini, Claude, Llama, or others — was used. Instead, it detects the statistical patterns that AI-generated text characteristically leaves behind, regardless of the underlying model. This makes it robust as new models emerge, since the approach does not depend on knowing the tool in advance.

The detector works across many languages. Importantly, Mentafy takes a language-specific approach rather than running every text through an English-centric model — a common weakness of generic detection tools. German is our first fully language-specific model, trained on German-language samples to deliver fairer and more reliable results. Spanish is next on the roadmap. English is well-covered throughout. Read more about our language-specific approach →

100 words is the safe minimum — shorter than that and results become less reliable. 50 words can work in some contexts, but the type of text matters too: a short email behaves differently from a poem or an essay introduction. As a rule of thumb: the more text you provide, the more reliable the result. For academic submissions, the full document always gives the most accurate picture.

All AI detectors produce some false positives, and we take this seriously. To protect students, Mentafy applies a conservative threshold: a paragraph is only classified as AI-generated when our model assigns it a probability above 90%. This means we deliberately let some borderline cases pass undetected in order to reduce the risk of falsely flagging human writing. No system is perfect, which is precisely why we present AI detection as one signal among several — not as a standalone verdict. For a broader discussion of false positive challenges across the field, see this overview from the University of San Diego.

Mentafy analyses text at paragraph level, not as a single whole-document score. If some paragraphs are AI-generated and others are authentically human-written, the report reflects exactly that — flagging the specific paragraphs that exceed the 90% likelihood threshold and leaving the others unmarked. This granularity is important: it allows educators to see precisely where AI was used, rather than receiving a single percentage that obscures the pattern.

They detect fundamentally different things. The AI detector looks for statistical patterns in the text itself — the characteristic fingerprints that language models leave behind. The plagiarism checker (and specifically Mentafy's S³ Semantic Source Search) looks for intellectual source overlap — whether the ideas, arguments, or structure were taken from an existing source, even if every word has been rewritten by AI. Both tools are complementary: a student could use AI to write entirely original-sounding text (caught by AI detection) or could use AI to paraphrase someone else's work (caught by S³). Using them together gives a much fuller picture.

Humanizers make detection harder — we won't pretend otherwise, and any tool that claims to be 100% reliable is misleading you. However, Mentafy's approach is specifically built to be robust against this: by detecting statistical patterns at the model level rather than surface-level stylistic features, many humanizer outputs still carry detectable traces. More importantly, Mentafy does not rely on AI detection alone — the Version Scan analyses whether the document was actually written organically over time, which no humanizer can fake.

Yes. Mentafy is designed from the ground with GDPR/CCPA principles in mind (data minimization, purpose limitation, user controls). Learn more >

Yes. The AI detector on this page is free to use directly. For institutional use — including batch processing of submissions, full integrity reports, and integration with your LMS — see our pricing or contact us for a demo.

AI detection is a post-hoc analysis of the finished text — it works on any document, even if Mentafy was not involved during writing. Writing process analysis (via the Version Scan or Writing Journal) records how the document was actually created over time. The two approaches are complementary: text analysis tells you what the finished document looks like; process analysis tells you how it came to exist. Together they provide the strongest possible basis for an integrity assessment.

No. Mentafy does not store the content of submitted documents beyond what is necessary for generating the result. The tool is fully GDPR- and CCPA-compliant. Your text is processed to produce the classification result and is not used to train models or shared with third parties. Full details are in our Privacy Policy.