Terms of AI

Transcription. Your voice note is turned into text by a speech recognition model (Whisper), so you do not have to type. You always see the text and can correct it before it is published.

Moderation. A language model reads the text and the transcription looking for insults, threats, sexual content and spam. A vision model looks at photos to tell a real photo from an advert or a screenshot, and to spot explicit content.

Tone. A language model notes whether a review sounds positive, neutral or negative. It is never shown publicly and never changes a ranking. We use it to find reviews worth reading, like a five-star rating written as a complaint.

All of these models run on Cloudflare Workers AI.

It never deletes anything and never publishes anything on its own. When moderation is unsure, the review is held and a person decides. When a model fails to answer, the review goes through and is flagged for a person to check, so a technical fault never silences you.

It never writes reviews. No review on Rabbit is generated. The rankings start empty on purpose.

We do not train AI models on your voice, your words or your photos, and neither do the providers we use for these tasks.

Transcription makes mistakes, more with accents, noise and place names. That is why you check it before publishing. Moderation also makes mistakes both ways: it can hold a fair review or miss a bad one. If it holds yours unfairly, it comes back as soon as a person looks. If it misses one, use the report button.

You do not have to use your voice: you can publish with a photo and type your review. You can also skip the location check. You can ask us to delete anything you published through the contact form.