AI Detector

Check whether a text was written by Fruitly. Every answer from Fruitly carries an invisible watermark.

Checked texts are neither stored nor logged.

For authorities, media, research and organisations with their own access. With a key, the page’s daily limits don’t apply.

What the result means

Written by Fruitly

The watermark is clearly detectable. A false alarm happens in fewer than one in 10,000 checks, and usually far more rarely for longer texts.

Signs of Fruitly

There are traces of the watermark, but no certain proof – for example with short or heavily edited texts.

No watermark detected

The text isn’t from Fruitly, is too short or was changed too much. That doesn’t prove a person wrote it.

How we mark Fruitly’s texts

So that AI text can be recognized as such, we mark all text Fruitly writes in a machine-readable way. This is how we implement the transparency obligations of Article 50 of the EU regulation on artificial intelligence.

Invisible

A Fruitly text looks like any other text. It contains no hidden characters and no markers.

No effect on answers

The watermark doesn’t change what Fruitly writes: quality, style and variety stay the same.

Always on

Every answer carries it. It can’t be turned off in the app or in the settings.

Only we can check it

It can only be detected with our secret key – that’s why you check texts here on this page.

What’s checked is the text of an answer as Fruitly wrote it. If formatting such as bold text or bullets gets lost when copying, that hardly matters.

How reliable the detector is

How often the detector recognized Fruitly text in our tests.

100%
for whole answers: 92%
for half an answer (about): 38%
for a quarter of an answer (about): 0
false alarms on human-written texts

The longer a text, the more certain the result. Parts of an answer can be checked too – just less certainly.

Limits of checking

No watermark is perfect. Keep this in mind:

Length

It becomes reliable from about 150 to 200 words. One or two sentences can’t be checked with certainty.

Editing

Small changes usually stay detectable. Paraphrasing, translating or rewriting with another program can remove the watermark.

Predictable text

Lists of numbers, set phrases and quotes carry little watermark.

Fruitly only

The detector doesn’t recognize texts from other AI providers – nor Fruitly texts written before the watermark.

No proof of the opposite

“Not detected” doesn’t mean a person wrote the text.

Larger checks

Checking is free for everyone. So that the detector stays fast for all, each internet connection can make up to 50 checks a day for texts under 500 words and up to 200 for longer texts. If many people share a connection, such as at a school or company, that runs out sooner – then write to us.

Authorities, media and research

Market surveillance and other authorities, law enforcement, media, fact-checkers, trusted flaggers, research, schools and universities, and civil society organisations get their own access – free and without a volume limit, for automated checks too.

Businesses

If you want to check larger amounts regularly, for example for a platform or a service of your own, you get access with a daily allowance. We’ll make you a fair offer.

Request access

Tell us briefly who you are and what you need access for.

If you publish AI text

If you create and publish texts with Fruitly to inform the public on matters of public interest, you may have to disclose them as AI-generated yourself (Art. 50(4) AI Act). This doesn’t apply if a person has reviewed the text editorially and takes responsibility for it.

Questions

We answer questions about the marking or the detector by email.

info@fruitly.net

Last updated: October 2026