Last updated September 8, 2026
How Zonlo uses AI
Zonlo uses AI to check what you said and give you feedback. We want you to know exactly what that means, in plain words, before you lean on it.
What the AI actually does
When you finish speaking, Zonlo takes the text of what you said, weighs it against what the moment called for, and hands back feedback: whether your reply worked, a more natural way to say it, and a short note or two on grammar or politeness. That is done by an AI (Google's Gemini API) together with our own language checking code. Nobody reads what you say in order to mark it.
AI also writes conversations for you. You type a short description of something you want to be ready for, and the same Gemini API writes the scene, the model answer, and what a good reply has to do. Your first one is free and the rest come with Pro. The conversations that come with the app are not made this way. The difference is set out under where the person is below.
What the AI does not do
- It does not hear your voice. Your speech is turned into text on your phone. Only that text is sent to be checked, and the audio never leaves the phone. The Privacy Policy has the details.
- It is not a chatbot or an AI character. Zonlo has no tutor avatar, no companion, nothing pretending to be a person. It is a tool that checks a reply and hands it back.
- It does not make decisions about you. Passing or not passing only affects your own progress in the app. It has no bearing on any real qualification, exam, visa, or credential.
It can be wrong
AI makes mistakes. Zonlo's feedback can miss a mistake, flag something that was actually fine, or suggest wording that is a little off for your situation. We measure how often that happens and publish the numbers openly, on How accurate is Zonlo? When the checking fails or is unsure, the app tells you plainly that it couldn't check your reply rather than making up a result. Treat every suggestion as a helpful second opinion, not a final ruling, and please don't lean on it when something important is riding on the outcome.
A conversation you asked for can be wrong in the same way. It was written by AI and not read by a person before it reached you, so its wording can be off in a way the conversations that ship with the app are not. The numbers on the accuracy page measure the checking, not the writing.
Your words and AI training
Two things of yours reach Google's Gemini API, both as text, both over an encrypted connection, under Google's terms: the words you said when a reply is being checked, and the short description you type when you ask for a conversation to be written. Nothing else goes with either one. We do not use what you say or what you type to train our own models, we do not sell it, and we never store or share a recording of your voice. How long anything is kept, and who handles it, is set out in the Privacy Policy.
Where the person is, and where the AI is
The conversations that come with Zonlo are written by a person, the one who runs Zonlo: the scenes, the model answers, and what counts as a good reply. The conversations you ask for are written by AI instead. Those are checked automatically before you ever see one, including a check that the model answer can pass the very test the learner will be held to. A conversation that fails is refused rather than shown to you.
The feedback you get in the moment always comes from software, whichever kind of conversation you are in. The checking behind it is tested against answers labeled by humans before any change ships.
Honest claims
We describe what Zonlo does in the words above and nothing stronger. We do not claim the AI is a teacher, that it will make you fluent, or that it is never wrong. Any accuracy figures we publish come from our own testing plus an outside check on politeness, and we show our working on the accuracy page. If we ever put a review on this site, it will be a real person's real experience.
Questions
If anything about how Zonlo uses AI is unclear, email support@zonlo.app. This disclosure works alongside our Terms of Service and Privacy Policy.