Check text for AI

What the check measures

The detector does not look for the signature of a particular model but for properties that generated text has in common: predictable word choice, an even sentence rhythm, worn phrases and neatly built paragraphs. The result is an estimate, not proof of who wrote the text.

With an account the text goes through seven detectors at once and each measures something different — from the deviation from your earlier texts through phrases typical of AI and machine-like sentence rhythm to hidden characters. We do not merge them into one number, because a disagreement between them is information in itself.

How to read the result

Each number is read on its own and says only what its detector measures. The marked sentences say more than the numbers: the colour map shows which passages each detector considers suspicious, and you can read them and judge for yourself.

When the detectors agree on the same passages, you have firmer ground. When each points somewhere else, the text is borderline and the decision has to rest on something more, such as a conversation with the author.

What to paste

Continuous text of at least 100 characters; a longer one gives a more reliable result. The full check automatically leaves out the reference list, quotations with a source, the table of contents, captions and other parts the author did not write in their own words. The demo without signing in evaluates the whole section you paste.

With an account you pick the text type before checking — essay, academic paper, e-mail and others. Some of the detectors judge the text differently depending on the type, because a cover letter is written differently from a blog.

Where the check has its limits

No AI text detector is one hundred per cent accurate. Reliability drops for short texts, translations, texts by non-native speakers and texts that AI only edited.

The result is therefore a basis for a conversation or for further checking, not a verdict. What usually decides is whether the author can explain the content and show how the text came about.

How many checks are free?

Without signing in, five demos per 30 days with texts up to 3,000 characters. A free account gives five checks per 30 days, texts up to 300,000 characters and all seven detectors.

Can the check tell which model wrote the text?

Not reliably. The detector measures how much the text resembles machine writing in general, and large models write in similar ways. Only a watermark would give away a specific model, and only the model's operator can verify it.

Which languages does the check work in?

Czech, Slovak, English and German; it recognises the language of the text on its own. For other languages some of the detectors have nothing to compare against and the result is indicative only.

Do you store the text I paste?

Without signing in we don't store the text — it goes through the analysis and we discard it. With an account it stays only in your history and is used only to make your own detection and humanisation more accurate.