You open Duolingo at 11:47pm, score 80% on a Spanish lesson, fail the pronunciation exercise three times, and close the app after six minutes. To you, this is a Tuesday night. To Duolingo, it’s a data point in a profile that spans every session you’ve ever had.
Language learning apps sit in an unusual position in the privacy landscape. They’re positioned as self-improvement tools — aspirational, educational, even admirable. Nobody thinks of them the way they think of a social media app. But the data they collect is far more detailed than most people realize, and what they do with it varies considerably across services.
What Language Apps Actually Track
The most obvious data is your learning performance: which words you know, which grammar rules you struggle with, your overall proficiency level by skill area. But this is the surface layer.
The more revealing data is behavioral:
Session timing. Language apps know exactly when you practice — what days, what times, for how long. Combined over months, this maps your daily routine in granular detail: when you commute, when you have downtime at work, when you can’t sleep.
Engagement patterns. They know when you skip lessons, when you rage-quit mid-exercise, when you push through a difficult unit, and when you stop returning for weeks. These patterns reveal attention span, stress responses, and consistency in a way that raw performance scores don’t.
Streak behavior. Streak mechanics are deliberately addictive. Apps that use them also log streak freezes, streak breaks, and purchases made to protect streaks. This data is directly useful for optimizing push notification timing and in-app purchase prompts.
Pronunciation attempts. Apps with speaking exercises capture audio samples of your voice. How this audio is stored, processed, and retained — and whether it’s used to train speech models — depends on each company’s terms of service.
Subscription and payment history. When you’ve subscribed, what tier, when you cancelled, when you returned. This is used for lifecycle targeting, but it also creates a payment history linked to your account.
Duolingo: The Largest Dataset in Language Learning
With roughly 100 million monthly active users, Duolingo has built the largest behavioral dataset in consumer language education. The app is free to use, which means the product is funded by advertising, subscriptions, and increasingly by what Duolingo can do with its data at scale.
Duolingo’s privacy policy discloses that it shares data with third-party advertising partners, analytics services, and “business partners.” Advertising targeting can use behavioral signals from within the app — including your learning patterns, engagement levels, and device data — to serve ads both inside Duolingo and elsewhere through ad networks.
In early 2025, Duolingo made headlines for replacing a significant portion of its human contractor workforce with AI systems. The company was explicit that this shift was driven by AI capabilities — the same AI that benefits from the scale of data its users generate daily. The relationship between user data, model training, and product development isn’t always clearly disclosed in terms most users would parse.
Duolingo’s terms do allow the company to use your content and data to improve its services. Whether “improve its services” covers training AI models is a question whose answer matters significantly but isn’t answered clearly in the typical privacy notice.
Babbel and Rosetta Stone: A Different Model
Babbel operates on a subscription-only model, which changes its relationship with user data. Without an advertising-based revenue stream, there’s less incentive to monetize behavioral data through ad targeting. Babbel’s privacy policy is generally more restrictive about third-party data sharing, though it still uses analytics services and third-party infrastructure.
Babbel is also a German company, which means it operates under GDPR as a baseline — generally a more protective framework than US standards. European privacy law requires a lawful basis for each category of processing, explicit disclosure of what data is shared with whom, and enforceable rights to access and delete your data.
Rosetta Stone, now owned by IXL Learning, is primarily positioned at enterprise and institutional customers — schools, corporations, government agencies. Personal subscription users exist, but the product’s data practices are shaped partly by the requirements of institutional contracts, which often include stricter data handling agreements than consumer products.
The Pronunciation Data Problem
Many language learning apps include speaking exercises — you say a word or phrase, the app scores your pronunciation. This requires capturing audio.
The questions that aren’t always clearly answered:
- Is your audio stored, or processed locally and immediately discarded?
- If stored, for how long and under what security measures?
- Is audio data used to improve speech recognition models, and if so, is this covered by the consent you gave when signing up?
- Can you request deletion of your stored audio samples?
Voice data is biometric data. It’s uniquely identifying in ways that most other data isn’t. A voiceprint built from thousands of pronunciation exercises is a detailed biometric record — and it’s worth understanding what specifically happens to it.
What Your Learning Choices Reveal
There’s a layer of inference that goes beyond the behavioral data itself. What languages you choose to learn, and why, can be deeply personal.
Someone learning Arabic may be connecting with cultural heritage, preparing for travel, following a romantic partner, or pursuing a career in a sensitive field. Someone learning Mandarin may have business, family, or political motivations that they wouldn’t choose to disclose. Someone learning a language associated with a specific country or community may be doing so for reasons they’d prefer remain private.
Language learning apps don’t inherently need to know why you’re learning — but the data they collect allows for inferences. Combined with other signals (location, device data, linked accounts), the picture that emerges can be richer than users anticipate.
Privacy Controls Available to You
Most language apps offer some controls, though how useful they are depends on the app:
Data export. GDPR-based rights (available to EU users and increasingly to users in other jurisdictions) include the right to request a copy of your data. This tells you what’s stored. Duolingo supports this; so does Babbel.
Account deletion. You can delete your account in most apps. What deletion actually does to your historical data depends on the company’s retention practices — some data may be retained in anonymized or aggregated form even after deletion.
Ad personalization opt-out. If an app uses advertising, there’s typically an opt-out for personalized ads. This doesn’t stop data collection; it stops the most targeted use of that data for advertising.
Microphone permission. You can revoke microphone access at the OS level (iOS: Settings → Privacy → Microphone; Android: Settings → Apps → [app name] → Permissions). This disables speaking exercises but also limits what audio the app can capture.
The Broader Pattern
Language learning apps aren’t uniquely predatory. But they’re a good example of a category where the privacy implications are systematically underestimated because the product feels benign and educational.
The data model is familiar: a free product supported by data and advertising, with premium subscriptions available for users who prefer a cleaner relationship. The scale of behavioral data collected is significant. The clarity of what happens to that data varies by company.
If you’re using a language learning app casually, this may be an acceptable tradeoff. If you’re learning a language connected to something sensitive in your life — heritage, a relationship, a profession, a cultural identity — it’s worth understanding what you’re sharing before you start.
Notes and Voice Memos in the Language You’re Learning
One thing many learners do is write notes or record voice memos to practice outside the app. If those go into a personal notes app or voice memo app that syncs to a cloud service, the same questions apply — who can see them, and what are the terms?
For personal notes, vocabulary lists, pronunciation recordings, or journal entries you’re writing in the language you’re learning, keeping them in a private storage environment gives you control that a general-purpose cloud app doesn’t. daftei stores files — including documents, voice memos, and photos — with TLS 1.3 encryption in transit and AES-256 at rest, without advertising or data sales. The free tier starts at 5GB.
Your language practice is personal. The data it generates doesn’t have to fund someone else’s ad network.