How AI uses your data

Updated October 12, 2026. This guide explains how AI features work in Speaking Coach. For the full notice, read our Privacy Policy.

What we collect, how we collect it, and why

When you choose AI features, we collect the microphone audio you speak during live practice, recorded presentation and answer audio, their transcripts, custom situation descriptions, practice context and language, presentation slide text, title, audience, purpose, instructions, slide timings, questions and answers. If you attach a CV, job post or interview notes, the app reads the text on your device and includes that text in the live interview context. Personal details you speak, type or include in these materials are part of the data shared. We use this data to run conversations, transcribe recordings, check custom situations, generate audience questions and provide feedback.

Who receives the data

ElevenLabs: live microphone audio and practice context, including attached CV, job post and interview-note text, to run your AI voice partner; recorded practice and answer audio to create transcripts, including a recording of your side of each practice session, sent after the session to measure your pace, filler words and pauses. ElevenLabs operates the voice service using its service providers and subprocessors.
Hearing a line in your own voice (optional): only if you choose it and agree in the app, the recording of what you said in that session is sent to ElevenLabs to make a temporary copy of your voice, which reads one suggested line back to you and is deleted straight away. The copy is not kept or used for anything else. You can withdraw this permission in Settings → AI data sharing.
Google Gemini, through ElevenLabs: ElevenLabs sends conversation text and practice context, including CV, job post and interview-note text you provide, to Google Gemini to generate your voice partner's replies, using a second Gemini model if the first is unavailable. These are distinct from the OpenAI feedback requests described below.
OpenAI: transcripts, custom situation descriptions and practice context for evaluation and feedback; presentation slide text, title, audience, purpose, instructions, slide timings, questions, spoken-answer transcripts and presentation transcripts to generate questions and answer feedback. Speaking Coach sends this text through our backend. The current iOS app does not send raw microphone recordings directly to OpenAI.
Vercel and Supabase: our backend runs on Vercel and uses Supabase for authentication and stored account, practice transcript, report and activity data. Requests sent through our backend are processed by these infrastructure providers as needed to deliver the feature. PowerPoint files you import are uploaded to our backend for conversion to PDF; PDF slide text is extracted on your device and shared with OpenAI when you request AI presentation features.

Your choice comes before sharing

Before the first AI data transfer or PowerPoint conversion upload, the app displays a disclosure explicitly identifying the data sent and its recipients. It asks again after permission is withdrawn or the disclosure changes. Permission requires tapping Agree to AI data sharing. Tapping Not now or closing the disclosure does not grant permission and blocks the pending action. Permission is saved for the signed-in account on that device. The explicit disclosure in app version 2.0.1 build 24 requires a new agreement even if an older disclosure was accepted. Permission to use the microphone is separate from permission to share data with AI providers.

You can read the disclosure and withdraw permission in Settings → AI data sharing. Withdrawal blocks future AI requests until you grant permission again. It does not erase data already transmitted. You may still use non-AI parts of the app, including viewing saved material and managing your account. Remove document content you do not wish to share before starting an interview.

Protection, retention and deletion

We require third-party service providers receiving personal data to provide the same or equal protection as this privacy notice, including confidentiality, appropriate security, purpose limitations and assistance with applicable data rights. AI provider processing is subject to their applicable service terms and data-processing agreements: OpenAI Data Processing Addendum and ElevenLabs Data Processing Addendum. See also OpenAI business-data privacy and ElevenLabs privacy policy for provider practices. ElevenLabs' DPA requires its subprocessors to receive privacy and security obligations substantially as protective as its DPA; its subprocessor list and backup-model documentation describe onward processing. Data is transmitted over encrypted HTTPS or encrypted voice connections.

Saved practice transcripts and reports remain in your account for history and progress until deleted under our account-deletion process. Interview documents, imported slide decks and presentation recordings are stored on your device, and a practice session's recording only until you close that session; interview document text, presentation text and recordings leave the device for the purposes described above. The current ElevenLabs voice agents save call recordings and conversation transcripts with unlimited retention configured, until removed through a deletion process. Other providers may retain submitted content and service logs under their applicable terms and account configuration. Withdrawal alone does not remove those copies. Contact admin@sodhera.com to request access or deletion, including assistance with provider-held data. This notice does not promise zero retention or that all providers exclude all submitted content from model training.

Account, subscriptions and product analytics

We collect account and profile information through sign-in and the information you enter to save your settings and practice history. Apple and RevenueCat process subscription and entitlement information to provide and restore purchases. Supabase stores first-party app-usage events. PostHog receives screen, onboarding and fixed-choice usage events, app/device information and related identifiers for product analytics. Those analytics events do not include your practice speech, transcripts, attached document text or free-text responses. We use this information to understand feature use and improve the app; we do not use it for cross-company advertising tracking.