What we hold, who processes it, how long we keep it, and how you delete all of it yourself — the complete list.
Last updated: September 2026
In shortThis policy covers IELTS AI — the practice platform — and every piece of data you create by using it.
IELTS AI is a practice platform that simulates the full IELTS exam — Listening, Reading, Writing, and Speaking — and scores your attempts against the official band descriptors. This policy explains, in plain English, what data the platform handles, who else touches it, how long it is kept, and how you delete it yourself.
Questions about anything here can go to contact@svanai.com. A person reads every message.
In shortYour account, your practice (written answers, spoken recordings, scores), your preferences, and a few helpers stored in your own browser. That is the whole list.
We keep this list deliberately complete. If data is not on it, we do not hold it.
| Data | What it is | Where it lives |
|---|---|---|
| Account record | Your email address and sign-in identifiers. | Supabase Auth |
| Practice responses | Written answers, selected answers, and transcripts of spoken responses. | Supabase Database |
| Speaking recordings | The audio you record during the Speaking module, used to produce your evaluation. | Supabase Storage (speaking-recordings bucket) |
| Scores and reports | Criterion-level band estimates, per-question review, and improvement plans. | Supabase Database |
| Preferences | Target band, test category (Academic or General Training), onboarding answers. | Supabase Database |
| Local helpers | Sign-in state, unfinished-attempt resume flags, and experiment assignments — stored in your browser's local storage, not on our servers. | Your browser (see the Cookie Policy) |
We do not collect payment data (no paid plans exist yet), and we do not collect advertising identifiers.
In shortRecordings are transcribed by a speech-to-text service; written answers and transcripts are evaluated by large-language-model providers; scores are stored back in your report.
When you complete a module, the following happens:
Evaluation requests are made from our servers, not your browser. We send them the minimum needed to score the response: the question, your answer, and the marking criteria.
In shortA short, real list: Supabase, Deepgram, our evaluation-model providers, and our audio-voice providers. No one else receives your practice data.
| Processor | Role | What they receive |
|---|---|---|
| Supabase | Account sign-in, database, and file storage. | Account record, practice data, recordings. |
| Deepgram | Speech-to-text transcription of Speaking recordings. | Your speaking audio. |
| OpenRouter / Groq / OpenAI | Large-language-model evaluation of responses. | Question text, your answer or transcript, marking criteria. |
| DeepInfra / Inworld | Text-to-speech voices for Listening audio. | Passage text only — no personal data. |
This list is accurate as of the date above. If we add a processor, we update this table before the change goes live. We do not sell your data, and we do not share it with advertisers.
In shortFor as long as your account is active. Delete an attempt and its data goes with it; delete your account and everything goes, automatically, right away.
In shortYou can delete your account yourself, right now, from your Account page. It runs automatically and removes your recordings, reports, and history — we hold no copies afterwards.
This is a promise about a feature that exists today, not a request you must email us to fulfill:
We deliberately do not ask you to email us to exercise this. If the self-serve flow ever fails you, write to contact@svanai.com and we will treat it as our error to fix.
In shortScores here are criterion-level estimates from automated evaluation against the official band descriptors. They are not official IELTS results, and they can be wrong.
Every score you see in IELTS AI is produced by automated evaluation against the official band descriptors, criterion by criterion. These scores are estimates. They are not issued, endorsed, or recognised by IELTS, Cambridge Assessment English, the British Council, or IDP, and they carry no official standing of any kind.
Automated evaluation can misjudge — a transcript can garble a word you actually said, a model can weigh a criterion differently than a human examiner would. When a module cannot be evaluated properly, your report says so instead of showing an invented number.
In shortEncrypted connections everywhere, access limited to what the product needs, and the smallest practical set of data shared with processors.
In shortIf what we hold or who processes it changes, this page changes with it — dated, plainly.
We update this page whenever our data practices change, and we change the date at the top so you can see at a glance whether anything is new. Material changes — a new processor, a new kind of data — will be called out on this page, not buried.