The same ten questions for everyone
Each product is read against one codebook: audio, transcript, training, vendors, human review, advertising, retention, deletion, transfers, roles. LingoSide answers the same questions in the same column.
Back to LingoSide
Lesson data, compared
What happens to what you say in a lesson?
We read the current privacy policies of five language apps and put them beside LingoSide, on the same ten questions. Every answer paraphrases the official text, links its source and shows the day we checked it. There is no score and no verdict on anyone's intentions.
- 5 language apps
- 10 questions
- Every answer sourced and dated
How we compared
One codebook, official sources, no score.
Official documents only, with dates
Every cell links the privacy policy or official product page it paraphrases, shows the date the document prints, where it prints one, and the day we checked it. Nothing is taken from news, reviews or app-store listings.
No score, no verdict on intent
The colour of a cell reads the stated practice against one fixed rule: does it limit or expand what happens to your data? Nothing adds up to a score, and nothing says what a company cares about. Where a policy handles something well, the table says so.
The table
Ten questions, six products.
Open any cell for the scope, the longer paraphrase and the dated source. On a phone, each question becomes its own card.
| Question | LingoSideLive lesson with a human teacher; read from LingoSide's privacy policy and FAQ. | DuolingoVideo Call and other AI features; product-wide where noted. | BusuuSpeaking Practice and Conversations; product-wide where noted. | BabbelAI conversation features; product-wide where noted. | MemrisePronunciation mode, Communicate and in-app conversations. | SpeakSpeaking and pronunciation features; one product-wide notice. |
|---|---|---|---|---|---|---|
| Raw audioIs a recording of your voice created or kept? | Limits your exposureNo audio recording is created or saved.Past its review date. Read the current policy at the source.Scope and sourceAudio from the source selected for the lesson is streamed to a speech-to-text provider to produce the transcript, and that provider is used in its EU region. LingoSide does not create or keep a recording. Scope: the live lesson in the LingoSide app.
|
Limits your exposureRecordings go to transcription and are deleted right after it.Past its review date. Read the current policy at the source.Scope and sourceFor conversational features, recordings are passed to AI providers solely for transcription and removed once transcribed. Pronunciation-review recordings stay within the specific lesson.
|
Not establishedMicrophone audio scores pronunciation and is converted to text; retention of raw recordings not separately stated.Past its review date. Read the current policy at the source.Scope and sourceThe app asks for microphone access to deliver the service. Audio is used to score pronunciation and to convert speech to text. Whether raw recordings are kept is not separately stated.
|
|||
| TranscriptIs what you said kept as text? | Limits your exposureTranscripts, translations and suggestions stay in your lesson archive until you delete them; archiving can be switched off.Past its review date. Read the current policy at the source.Scope and sourceTranscripts, translations and AI suggestions are stored in the lesson archive, on the device and on LingoSide's server, until you delete them. With session archiving enabled, which is the default, finalised content is checkpointed locally during the lesson; a user setting can disable archiving.
|
Not establishedNot stated for Speaking Practice or Conversations.Past its review date. Read the current policy at the source.Scope and sourceNo statement covers storing transcripts of Speaking Practice or Conversations. The AI Coach section lists text inputs as processed, and community written exercises and messages are collected.
|
PartlyThe transcript may be kept and reused in later conversations.Past its review date. Read the current policy at the source.Scope and sourceFor conversational features, the transcript may be stored and used in later conversations. No separate retention term for transcripts is given.
|
Not establishedNot stated.Past its review date. Read the current policy at the source.Scope and sourceThe notice mentions turning audio into text but says nothing about storing transcripts or conversation contents, and gives no separate retention rule for them.
|
||
| Own model trainingDoes the company train or improve its own models on it? | Limits your exposureLingoSide does not use lesson data to train AI models.Past its review date. Read the current policy at the source.Scope and sourceThe privacy policy and the FAQ state that lesson data is not used to train AI models. Recognised text, never audio, is sent through LingoSide's server to translation and AI providers for a specific operation.
|
Limits your exposureOwn AI development uses voice recordings only in anonymised form.Past its review date. Read the current policy at the source.Scope and sourceVoice recordings are anonymised for speech analysis and internal machine-learning development. The policy lists AI training and quality assurance as a purpose with no use of identifiable voice recordings for it.
|
||||
| Vendor trainingMay the AI vendors it uses train on it? | Limits your exposureEvery speech, translation or AI provider is configured so lesson data does not train or improve its models.Past its review date. Read the current policy at the source.Scope and sourceThe FAQ states that each speech, translation or AI service is configured so that lesson data is not used to train or improve that provider's models, and that a provider which cannot confirm this is not used. The policy names OpenAI and Google Gemini for text and an EU-region speech-to-text provider for audio.
|
Limits your exposureText and audio may go to OpenAI and Google; contracts bar vendor use for their own purposes.Past its review date. Read the current policy at the source.Scope and sourceThe named AI vendors are OpenAI and Google, with Google, Apple and Amazon Web Services for speech recognition. The policy states that its contracts prevent vendors from using personal information for their own purposes.
|
Not establishedChegg permits training provider models; Busuu restricts providers’ own purposes. The scope for individual vendors is unresolved.Past its review date. Read the current policy at the source.Scope and sourceChegg’s User Content/Audio and Visual Information clause permits training or fine-tuning its own and third-party providers’ models. The Busuu supplement takes precedence and prohibits service providers from using personal data for their own purposes, allowing specified processing under its instructions. These clauses do not establish whether Azure AI Speech, OpenAI or another named vendor may train on particular Busuu data. A restriction on own purposes is not an explicit ban on all commissioned training.
|
Limits your exposureVendors act as processors; contracts must exclude training on the data. No vendor is named.Past its review date. Read the current policy at the source.Scope and sourceAI tools come from unnamed vendors acting as processors. The policy states that contracts must ensure the data is not used for the vendors' or third parties' training purposes.
|
Not establishedNo AI vendor is named; vendor training is not addressed.Past its review date. Read the current policy at the source.Scope and sourceAWS and Google are named as hosting, analytics, marketing and advertising providers, and anonymised data may go to researchers. No AI vendor is named and vendor training is not addressed.
|
Not establishedOpenAI and Microsoft receive audio to provide and improve the Services; vendor training is not addressed.Past its review date. Read the current policy at the source.Scope and sourceOpenAI and Microsoft are named as recipients of audio information for providing and improving the Services. The notice is silent on whether those vendors may train their own models.
|
| Human reviewDo people listen to or read it? | PartlyNo one at LingoSide listens to a lesson or reviews transcripts to train models; OpenAI may review text its abuse classifiers flag.Past its review date. Read the current policy at the source.Scope and sourceThe privacy policy states that LingoSide creates no audio recording, so there is nothing to listen back to, that transcripts are not reviewed to train or improve models, and that access by the team stays limited to running and supporting the service. It also states that OpenAI, which produces translations and suggestions from recognised text, may review text its abuse classifiers flag and keeps that text for up to 30 days.
|
Not establishedNot stated for audio or AI conversations.Past its review date. Read the current policy at the source.Scope and sourceThe policy does not address human review of audio or AI conversations. Its only review language concerns public profile content, which may be scanned or sent to a moderation vendor.
|
Not establishedNot stated.Past its review date. Read the current policy at the source.Scope and sourceThe policy does not mention human or manual review of voice recordings, transcripts or AI conversations.
|
Not establishedNot stated; session-recording data is accessible only to authorised personnel.Past its review date. Read the current policy at the source.Scope and sourceHuman review of audio or conversations is not mentioned. The policy says only that session-recording data is accessible solely to authorised personnel.
|
Not establishedNot stated.Past its review date. Read the current policy at the source.Scope and sourceThe notice does not mention people listening to, annotating or reviewing user audio or conversations.
|
|
| AdvertisingTargeted ads, sale or sharing, retargeting, cross-device tracking | Limits your exposureNo advertising, no sale or sharing for advertising, no cross-site tracking, no advertising identifiers.Past its review date. Read the current policy at the source.Scope and sourceThe privacy policy states that LingoSide shows no advertising, does not sell or share data for advertising, and uses no cross-site tracking or advertising identifiers. The site counts page views without cookies and uses a daily server hash of IP, User-Agent and hostname for approximate visitor and session counts, without a stored browser identifier or linkage across days. Behavioural analytics require consent; Do Not Track and Global Privacy Control also suppress page views.
|
|||||
| RetentionFor how long is it kept? | Limits your exposureLesson archive until you delete a lesson or the account; account data while the account exists.Past its review date. Read the current policy at the source.Scope and sourceAccount data is kept while the account exists, the lesson archive until you delete a lesson or the account, waitlist and contact entries until handled, and server logs according to the hosting provider's rotation, typically weeks. No numeric term is stated for the lesson archive. The policy adds that OpenAI keeps the text it receives for up to 30 days to detect abuse and stores nothing else from those requests.
|
PartlyPurpose-based under the Chegg policy; no numeric term, with deletion or deidentification once the purpose ends.Past its review date. Read the current policy at the source.Scope and sourceVoice recordings are described as retained for the service and for improvement, and the Busuu policy gives no period of its own. The Chegg policy it delegates to keeps personal data no longer than reasonably necessary for the purpose it was collected for, allows longer retention where law, disputes or legal rights require it, and deletes or deidentifies the data once that need ends, isolating it from active processing where deletion is not yet possible. No numeric term and no audio-specific term is given.
|
PartlyPurpose-based with statutory periods; raw audio only until transcription.Past its review date. Read the current policy at the source.Scope and sourceThe default is purpose-based, as long as necessary, plus statutory periods for accounting records, commercial correspondence and research data. Raw audio has its own rule: deletion after transcription. No numeric term is given for transcripts.
|
PartlyNo longer than reasonably necessary; no numeric term for audio.Past its review date. Read the current policy at the source.Scope and sourceData is kept no longer than reasonably necessary, then deleted, de-identified or isolated in backups. No numeric term is given for audio; the Korea section cites statutory commerce-record periods only.
|
||
| Export and deletionCan you take it out and delete it, including at processors? | Limits your exposureAccess, correction, export and deletion on request; account deletion purges telemetry and asks the telemetry providers to delete.Past its review date. Read the current policy at the source.Scope and sourceUnder the GDPR you can ask LingoSide to access, correct, delete or export your data. Deleting the account removes account data and the lesson archive from LingoSide's server and purges unsent telemetry; LingoSide then asks the telemetry providers to delete what they hold.
|
PartlyAccess, export, correction and deletion via the Data Vault; propagation to AI vendors not stated.Past its review date. Read the current policy at the source.Scope and sourceUsers may access, export in a transferable format, correct and delete personal data, and delete the account, through the Data Vault or by email. The policy notes that these rights are not absolute. Whether a deletion reaches AI vendors is not stated.
|
PartlyCopy, change or deletion requests by email to privacy@busuu.com; propagation to processors is not specified.Past its review date. Read the current policy at the source.Scope and sourceBusuu’s Data Privacy Choices section directs copy, correction and deletion requests to privacy@busuu.com. It does not specify how deletion reaches processors.
|
Limits your exposureDeletion and a machine-readable copy on request; deletion instruction to service providers stated in the US section.Past its review date. Read the current policy at the source.Scope and sourceUsers may request deletion and receive the data they provided in a machine-readable format. In the CCPA section Babbel states that it will instruct service providers to delete as well.
|
Limits your exposureDownload a copy via a link; delete in settings or on request; recipients alerted to erase where possible.Past its review date. Read the current policy at the source.Scope and sourceUsers can download a copy of their data through a link and delete their account in the web or app settings or by request. Recipients of shared data are alerted to erase it where possible.
|
PartlyAccess, portable copy and deletion by email or phone; propagation to vendors not addressed.Past its review date. Read the current policy at the source.Scope and sourceUS and EU or UK rights to access, receive a portable copy and request deletion are exercised by email or phone; account details can be edited after login. Whether a deletion reaches vendors is not addressed.
|
| Where it is processedData residency and international transfers | Limits your exposureSpeech-to-text in the provider's EU region; providers outside the EEA under GDPR safeguards such as standard contractual clauses.Past its review date. Read the current policy at the source.Scope and sourceThe speech-to-text provider is used in its EU region, and PostHog and Langfuse in the EU. Where a provider processes data outside the European Economic Area, the policy relies on GDPR safeguards such as standard contractual clauses. Hosting locations are not named.
|
PartlyProcessed in the United States under the EU-U.S. Data Privacy Framework.Past its review date. Read the current policy at the source.Scope and sourceDuolingo is US-based and transfers non-US users' data there. The stated mechanism is certification under the EU-U.S. Data Privacy Framework, its UK extension and the Swiss-U.S. framework. Where AI vendors process the data is not stated separately.
|
PartlyLondon controller; transfers outside the EEA, UK and Switzerland under adequacy decisions, standard contractual clauses or the EU-U.S. Data Privacy Framework.Past its review date. Read the current policy at the source.Scope and sourceThe Busuu policy gives a London controller address and refers EEA and UK readers to the Chegg policy's European disclosures. That policy states that personal data may be transferred to, stored in or accessed from a country other than the one where it was collected, including the United States, and that transfers out of the EEA, UK and Switzerland rely on adequacy decisions, standard contractual clauses or the EU-U.S. Data Privacy Framework. No processing region is named for the AI speaking features specifically.
|
Limits your exposureProcessed in principle on European servers; transfers under adequacy decisions or standard contractual clauses.Past its review date. Read the current policy at the source.Scope and sourceData is processed in principle on European servers. Transfers outside the EEA, including to the US subsidiary, rely on adequacy decisions, EU standard contractual clauses or consent.
|
PartlyProviders in the United States, described as Privacy Shield certified; no other mechanism stated.Past its review date. Read the current policy at the source.Scope and sourceData may be shared with providers in the United States, named as AWS and Google and described as Privacy Shield certified. The controller is Memrise Limited. No server location, standard contractual clauses, Data Privacy Framework or adequacy decision is mentioned.
|
PartlyMay be stored outside your jurisdiction; EU and UK transfers to the US rely on standard contractual clauses.Past its review date. Read the current policy at the source.Scope and sourceData may be stored outside the user's jurisdiction, where Speak and its providers operate. EU and UK transfers to the US rely on standard contractual clauses, with the EU-U.S. Data Privacy Framework and adequacy decisions also cited.
|
| Roles and childrenTeacher–learner roles and children's data | Limits your exposureOne teacher, one learner; where a learner is a child, the teacher, school or guardian arranges the account.Past its review date. Read the current policy at the source.Scope and sourceTwo sidecars share one lesson: the learner's private scaffolding and the teacher's evidence and decisions. Every participant must be informed and give permission before a lesson. Where a learner is a child, the teacher, school or guardian arranges the account and informs the learner.
|
Limits your exposureDuolingo for Schools has teacher and learner roles; child accounts get reduced data use.Past its review date. Read the current policy at the source.Scope and sourceTeachers in Duolingo for Schools see a learner's name, email and progress and can act on the account. Child users register with a username only, receive no personalised ads, and their speech is not shared for product improvement. Whether Video Call is available to child users is not stated.
|
Limits your exposureNo teacher–learner roles for AI features; under-13s excluded, EEA under-16s need parental consent.Past its review date. Read the current policy at the source.Scope and sourceLive-lesson audio and video go to the teacher, and sponsoring employers or schools receive usage reports. No teacher or learner role is defined for the AI speaking features. The service is not directed at under-13s, and EEA users need to be 16 or have parental consent.
|
PartlyNo teacher–learner role distinction for data handling; no sale or sharing of under-16s' data without opt-in.Past its review date. Read the current policy at the source.Scope and sourceBabbel Live is covered only as session booking and delivery; no separate teacher or learner data role is defined. The minors clause appears in the US section only, and no general minimum age is stated.
|
Limits your exposureNo teacher–learner roles; under-13 data not knowingly collected; users must be at least 13.Past its review date. Read the current policy at the source.Scope and sourceNo teacher, learner, school or classroom role is mentioned in the policy or terms. Data from under-13s is not knowingly collected and is deleted on notice; the terms require users to be at least 13.
|
Limits your exposureNo teacher–learner roles; not intended for under-13s; no sale of known under-16s' data without opt-in.Past its review date. Read the current policy at the source.Scope and sourceNo teacher, tutor, school or classroom role is mentioned. The service is not intended for under-13s, data of known under-16s is not sold without opt-in, and Korea and Japan thresholds are stated separately.
|
LingoSide's column
Only what we can show.
LingoSide's column is held to a stricter rule than the others: it may carry only statements that LingoSide's own privacy policy or FAQ already publishes, read with the same codebook as everyone else. A question those documents do not answer stays empty on purpose.
What LingoSide states
- No audio recording is created or saved; audio is processed live to produce a transcript.
- Transcripts, translations and suggestions stay in your lesson archive until you delete them, and archiving can be switched off in a setting.
- Lesson data trains no AI model, at LingoSide or at the providers it uses.
- No advertising, no sale or sharing for advertising, no cross-site tracking.
Why a cell can still be empty
An empty cell is neither a bad answer nor a good one. It means LingoSide's published documents do not answer that question for the whole path from the microphone to every external service. Statements are added only with proof behind them, never by stretching a narrower fact over the whole path.
Ask about a specific questionSources
Every document this page reads from.
Document dates are the ones printed on the documents; a document without a printed date is marked as such. Checked dates are ours.
LingoSide
- LingoSide Privacy policy — How LingoSide handles your dataDocument date: 2026-09-15Checked:
- LingoSide FAQ — How is lesson privacy handled?Document date: no date printedChecked:
- LingoSide Technology — memory, checkpoints and archivingDocument date: no date printedChecked:
Duolingo
- Duolingo Privacy PolicyDocument date: 2026-05-26Checked:
- Duolingo Store — Your Privacy ChoicesDocument date: no date printedChecked:
Busuu
- Supplemental Privacy Policy for BusuuDocument date: 2026-09-09Checked:
- Busuu Conversations — product page with the in-app voice noticeDocument date: no date printedChecked:
- Chegg Privacy Policy — parent policy the Busuu supplement delegates toDocument date: 2026-09-09Checked:
Babbel
- Babbel Privacy PolicyDocument date: February 2026Checked:
- Babbel — Do Not Sell or Share My Personal InformationDocument date: no date printedChecked:
Memrise
- Memrise Privacy PolicyDocument date: no date printedChecked:
- Memrise Terms of UseDocument date: no date printedChecked:
Speak
- Speak Privacy NoticeDocument date: 06/05/2026 as printedChecked:
A design principle
Privacy creates the question. Technology earns the answer.
Privacy is easy to promise. The harder task is building technology that remains genuinely useful without asking people to surrender control.
That tension shaped LingoSide: not privacy added after the product was built, but privacy treated as a design constraint from the beginning.
The details stay under the hood. What matters here is the result: technology should not merely ask for trust. It should earn it.