How Fíor works
There are two ways to check text with Fíor, and both are free: the browser extension for articles on the web, and the online Playground for any text you paste in. Both use the same model and give the same kind of result.
The browser extension
Add Fíor to your browser
Install the extension and sign in with your email. We send a 6-digit code, so there's no password to remember.
Open an article and press Run
Click the Fíor icon and press Run on this page only. Fíor takes the first 1,000 or so characters of the article's main text.
Read the result
A few seconds later you see Human or AI, with how likely each is. Every check is saved to your dashboard.
The online Playground
Sign up for free
Enter your email on getfior.com and type in the code we send you. Your account is ready straight away.
Paste your text
The Playground opens first on your dashboard. Paste anything from 300 to 1,000 characters: a paragraph from an email, a post, a document.
Press Check text
The result appears under the box: Human or AI, how likely each is, and the AI family it reads like when the model is sure.
Good to know
- Formatting is removed when you paste, so only the words are checked.
- You can run a new check every 15 seconds.
- Playground checks are kept apart from the pages you check with the extension.
What the result means
The extension and the Playground word results the same way. The wording tells you how sure the model is, not just which way it leans.
What's behind it, and how far to trust it
Fíor uses a RoBERTa language model that we trained on hundreds of thousands of news excerpts from published research datasets: human writing from CNN, the Daily Mail, the BBC and other outlets, and AI writing from models including GPT-4o, Llama 3, Mistral, Gemma and Qwen. No sites were scraped to build it.
On news text it hadn't seen before, it does well. Real pages are messier than any test, so treat a result as a strong hint rather than proof. It is tuned to be careful: it would rather miss some AI text than wrongly say a person's writing is AI, and it tells you when it isn't sure.
Where it's weaker
- The opening paragraphs of an article are harder than the middle, and that's the part Fíor reads.
- It was trained on English news. Other kinds of writing, and other languages, are less reliable.
- The newest AI models weren't in the training data yet, and AI text that a person has edited is harder to spot.
- Short text, lists and captions don't give it enough to go on.
Try the extension
A free browser extension for Chrome, Edge and Firefox.