How Duolingo grows

After reading reviews where users asked to pay to remove ads, tested a one-time purchase of 30 ad-free days at $3, $5 and $10 instead of a subscription.

Purchase pages explain the mechanics: 'this is a seven day free trial; on day five we'll send you a notification to see if you still want your subscription.'

Iterated purchase button copy in A/B tests: 'Subscribe' (baseline) -> 'Start my free trial' (big uplift) -> 'Try for free' (better) -> 'Try for $0.00' (best; believed live). Part of ~400 concurrent A/B tests, many on the purchase flow.

Instead of asking users at onboarding when they want to be reminded (morning, evening, a set time), schedule the next day's practice reminder for 23.5 hours after the user's first practice of the current day, so it tracks when they actually practiced the day before. Tested against many other scheduling strategies; user-set times lose because people are rarely free at the same time forever.

Daily practice reminder sent every day; stops after 7 days of inactivity. ~250 copy templates are eligible, and a multi-armed bandit learns which to send to each user. Templates carry eligibility criteria (e.g. streak length) instead of feeding many user features into the model. PMs keep writing new templates; each batch is A/B tested: control = bandit with 250 templates, variant = bandit with 260 (10 new). Measured locally (click-through of the new ones) and globally (top-line metrics). Bandit tuning changes can be A/B tested the same way. CTR is also drilled down by UI language; weak languages (e.g. Romanian) go to the localization team.

After an iOS update started truncating notification text after ~4-5 words, tested very short copy (titles under 30 characters, e.g. 'Got 3 minutes? Study Spanish now.'), especially for new or less engaged users on short streaks. For highly engaged users (e.g. 50-day streak), more complex copy referencing mechanics works too ('Hey, you're out of streak freezes. Don't lose your 50-day streak.'). Targeting is done via eligibility criteria on each template.

Paywall call-to-action reads 'Try for $0.00' rather than 'Try for free'. App Masters also tried it (no result given).

Duolingo changed its paywall call-to-action text to 'Try for $0.00' (released around January), replacing 'Try for free'.

Default: a notification is sent about 2 days before the free trial expires. Test: at trial signup, let the user choose which day they get that reminder. The extra step draws attention to the fact that a reminder is coming and they can cancel later, so users wait until day 5 or 6 to judge whether they are getting value instead of cancelling immediately.

Instead of dropping the standard Super Duolingo upsell into the onboarding flow, make the offer contextual to onboarding: ask 'How do you want to achieve your goals? With ads or without ads? Learn faster or at your own pace?' so the subscription is part of the onboarding answers. New users are far more likely to convert than existing ones.

Tested reducing the free trial from two weeks to seven days in several combinations with how many trials a user is eligible for. Beyond conversion, the shorter trial means experiment readouts come in 7 days instead of 14, so losing tests are killed sooner and more tests run per quarter.

'Free taste': a three-day, no-strings trial-like experience of Super or Max, with no auto-convert. User can convert during or after. Also used for paid-to-paid: Apple won't allow another free trial to an existing subscriber, but a Super subscriber can get a Max free taste and then return to Super. Aimed at users hesitant to start a real trial.

Copy always states it's a three-day free period/preview and says 'Hey, you just earned this.' The reward chest used to sit on the learning path; moved it to after completing a lesson because earning it there felt more rewarding. Uses timers for urgency, and the subscriptions page says you can upgrade anytime and makes clear there won't be a surprise charge (so Super subscribers don't cancel).

Duolingo's first monetization experiment (2016): users could buy a repair of a broken streak for about $1-3.

Introduced ads in 2016 with a single placement: after a lesson is completed (not app open, not mid-lesson), still the only placement today. Team dogfooded the app daily to judge the feel, and A/B tested each ad change against retention, app closes and reviews.

Before deciding to launch ads in 2016, quantified complaints and negative reviews about ads per million DAU and compared them to complaints about other topics (e.g. a missing language course), alongside neutral retention in A/B tests.

At each lesson end an ML model picks between a programmatic ad and an internal ad (Super, Max, Duolingo English Test), maximizing LTV with purchase likelihood as the objective. If a user ignores the Super ad about three times it backs off. Internal ads are character-driven videos that explain the subscription features.

Ran a 1-2 year ads holdout, but no longer recommends year-long holdouts (they get forgotten and polluted by bugs). Now runs 3-6 month holdouts or long experiments on things like leaderboards (10 weeks to climb from lowest to top league) and push notifications, where novelty lifts fade.

Duolingo Plus (2017): no ads plus fully offline use; because the purchase page looked thin with two features, added a third line 'Support Duolingo's mission', echoing what users said in reviews.

Hearts (now Energy) limit new lessons after a number of mistakes; free users regenerate by practicing old content, watching ads, or (then) an IAP. Made unlimited Hearts a subscription feature. Also added a Practice Hub tab to the subscription (later made free).

A/B tested giving Explain My Answer (an LLM-based Max feature explaining your specific mistake) free vs keeping it in Max. Revenue impact was small, and LLM costs had fallen with better caching, so it moved to free. Practice Hub was also moved from Super to free. Rationale: free users are the word-of-mouth engine; big trade-offs are signed off by senior leaders including the CEO.

Each day a user skips the app they get a practice reminder push (Duolingo's most powerful notification), up to a series of 7. At 7 days of inactivity the practice reminders are turned off because the extra value from continuing is much lower. The key window is the first 3-4 days: for a daily-use app, users not back by then are very hard to win back.

For users lapsed from about 1 to a few weeks, send a weekly progress-report email comparing this week's activity to last week's (XP earned, time spent). Email gives more room than a push to users who may have uninstalled. Test many copy tones for the lapsed case: hopeful ('Have you taken a break? Come back and jump in'), a challenge ('seems like a slow week. How much are you going to learn this week?'), or slightly passive-aggressive. Showing simple engagement stats and letting the user infer the value was cheaper to build than content-specific recaps (words learned) and got most of the benefit.

At Duolingo's scale, retention-team tests reach statistical significance on top-line metrics (DAU, D1 retention) the morning after launch, but experiments still run at least 3-4 weeks on average to rule out novelty effects. Teams with small user bases (e.g. Duolingo for Schools, a teacher dashboard) calculate required test length up front.

For bigger features, after launch keep ~1-2% of users in control for about 3-4 months to measure long-term impact, especially for features (like social features) whose effect on DAU builds slowly via network effects. After fixing a big bug, may run a holdout that gives the bug back to ~1% of users to measure its impact. Each launched feature also gets a feature dashboard tracking engagement over time to spot novelty decay.

Multiple notifications per day each get a distinct purpose and CTA. Practice reminder: clear CTA ('Hey, it's time for your daily practice ... study Spanish for 5 minutes'). For users on a streak who haven't practiced by ~10 PM, a 'streak saver' with a more alarming tone ('Last reminder', 'Hey, it's getting late'). Framed as a narrative so it feels like the mascot is having a conversation with the user.

Notifications are written in the voice of the mascot (Duo), mostly by PMs with content designers reviewing. Tone varies across the up-to-7-day schedule: encouraging ('Amazing job. You're on a 10-day streak. Come back and do Spanish.') and passive-aggressive ('You can practice Spanish if you want.'). The bandit learns which ones work over the schedule.

Tested emoji placement in push notifications. Emoji in the title, as the leading character, performed better than emoji in the body or at the end of the text. Hypothesis: a title emoji grabs attention; body emoji make the message look noisy.

Goal: get users to tap the notification itself, since it deep-links into a lesson and raises lesson completion. Tested copy templates explicitly instructing users to tap, e.g. 'Tap here to start a quick lesson.'

Losing a streak is a major churn point. Added a streak freeze item: equipped in advance, it keeps the streak if a day is missed. Bought with the in-app currency (gems) or earned as a reward (e.g. completing a daily quest). Iterated on gem pricing and reward drop rates, and raised the cap so users can equip two at a time. Guardrail: protect the 'sanctity of the streak' so it stays meaningful.

iOS home screen widget showing current streak length and the mascot, who gets angrier/more worried through the day if the user hasn't practiced (Tamagotchi-like). Feels less spammy than a push because the user opted in, yet it is a constant reminder. Educational content in the widget was on the roadmap.

When a user finishes a session or lesson, show an in-app card promoting the home screen widget ('Hey, we have a widget. You should install it.'). Tested as an experiment. Separately optimize what the widget looks like.

Moved from a 14-day free trial to a 7-day free trial so paid-conversion results of experiments come back sooner.

Building on the Blinkist-style trial reminder paywall ('we'll remind you before the trial ends'), Duolingo tested letting users choose which day they get the reminder, drawing attention to the reminder promise.

Goal is clarity and confidence, not persuasion: explain in a timeline what happens on which day of the trial and when the user is charged; when switching plans, state clearly how much money they get back.

When a learner runs out of hearts/energy mid-lesson, position the choice as: if you don't continue you lose your progress; pay gems or start the free trial now to keep your progress and finish the lesson. Still shows a sense of progress so it isn't punitive.

Users are eligible for multiple free trials, spread over time, because they are in different life stages or didn't get to experience the benefits the first time. Each user faces the purchase decision three or four times. (Apple allows the intro offer once per SKU, so this needs workarounds.)

Instead of trialing the whole tier, offer a free taste of one feature in isolation: just one video call (part of the Max tier), then see if users convert.

After a free taste ends, a home drawer comes up stating 'Your X benefits have ended' with two options: continue (free), or explore/resubscribe. Emphasis on being clear it has ended and that rejoining is easy.

Free tier shows ads: a mix of house ads promoting subscriptions and third-party ads. First ML use: optimize the mix of own-product promos vs external ads. Then predict which plan (Super vs Max) a likely subscriber will pick, replacing fixed per-tier ad ratios, using signals like past taps on a plan or engaging with a video-call lesson.

Many creatives promote Super as an ad at the end of a lesson; a bandit optimizes which creative works best for which user.

Duolingo's original model (~2013): the website was free; learners' translation exercises were crowdsourced into full article translations sold to outlets like CNN and BuzzFeed. Quality work pulled resources from teaching toward running a translation business.

Started with AdMob, then added Facebook, Flurry and later Unity to broaden demand. Blocked every category not brand- or young-audience-safe (gambling, sketchy games) and worked with networks on custom blocks (e.g. ads with three X buttons).

US-only direct ad deals: one of Duolingo's eight characters (e.g. Lily) introduces the brand's ad with a short custom-written joke script, then the ad plays. Formats are Duolingo-built. Goal: cut programmatic volume substantially in 2026.

Added free trials to the subscription even though the core product was free and paid features (no ads, offline) did not need to be experienced. The trial removes payment hesitation.

Launched math as a separate app, planning one app per subject, then moved math, music and chess into the main app's course switcher with light promos. Picked subjects that are fun to teach on mobile, have existing demand, fit bite-sized skills and have a senior internal champion (skipped coding and running). Each took 6-12 months of prototyping and dogfooding.

Built a higher-tier subscription where users pressed a button to connect instantly with a human Spanish tutor, no booking. Users had requested it for years. Iterated to ease users in, then shut it down. Replaced by an AI video call with the character Lily.

Machine-learning-driven system chooses the notification copy (push and email) for each user and optimizes on its own; PMs just add copy variants to the pool (soon AI may write them). Never treat a notification as 'done': the practice reminder has been under continuous experimentation for 6 years. Judge variants on open and click-through rates, which reach significance fast at scale. Lapsed users deserve the most experimentation because you know least about what they need.

End-of-year recap in the style of Spotify Wrapped showing each user's learning stats. Even users who only used the app for a few days months earlier get a delightful, funny recap the moment they reopen. Outreach to long-lapsed users is tied to a moment and personal value rather than 'here's a product update'. Timed so returning users flow into New Year's resolution season.

Using the in-app social graph (55-60% of users have a friend), prompt active users to reach out to friends who have been gone for months ('get them to come back'). The message comes from a person, not the brand. Returning users are then eased into social features such as friend streaks (learning alongside that friend). Even winback messages from strangers met on the leaderboard get higher click-through than Duolingo's own messages.

Every change, down to back-end changes with no UX impact, is released as an A/B test to catch regressions elsewhere in the app. In-house experimentation platform built by a data infrastructure/experimentation team: engineers set up tests once a feature is built, and reports on new data are generated every morning; PMs check the experiments dashboard first thing each day. One retention team runs ~20 experiments at a time. Each test yields 5-6 learnings that are turned straight into new tests. Different experiments may run on the same surface on iOS and Android, then the winner is aligned across platforms.

Rule: no multiple tests on the same real estate at once; when unavoidable, analyze the intersection of the two experiments for interaction effects. When a product review council (CEO and others) approves a test, the PM emails the product org what is being tested and what was discussed; at launch decision, another email shares data and learnings, so other teams can spot impacts on their features or metrics. Teams also cross-analyze overlapping experiments.

All 600+ employees use the app daily. A Slack channel (#feedback-product) lets anyone dump feedback on any part of the app; PMs watch it for insights and test ideas. Junior PMs are also empowered to own roadmap areas and ideate bottom-up.

Streak repair restores a streak that was actually lost after all streak freezes were used. Historically sold as an in-app purchase and included as a perk of the Super subscription. Team works to keep streak features aligned and not too complicated.

Basic plan gets non-AI features and a limited taste of AI; full/unlimited AI (e.g. Duolingo Max's voice role-play with Lily) requires a higher Pro/Max tier, to cover compute costs. Also seen at Perplexity, ChatGPT, Claude, ElevenLabs, Tolan.

Build reactivation campaigns around moments that match users' mental model of the product: back to school (people get into a learning mindset even if not in school) and the start of the year (learning a language is a common New Year's resolution). The end-of-year review brings users back just before resolution season. Guest values timing plus the right message over the specific creative.

Launched in 2011/2012 and focused only on the free experience, with no Duolingo Plus subscription until 2017. Now expanding into math and music to reach early adopters in new categories.

As a large company facing rising costs for each new user, Duolingo adds more ads to the free experience. This raises revenue from non-subscribers and pushes annoyed free users to pay for Duolingo Plus or Max.

Launch a family plan: acquire one family member (e.g. the mom), who then adds the dad and kids under a higher-priced family subscription, expanding revenue from one acquired customer.

Tinder sells baskets of super likes that are not part of the subscription but enhance use; Duolingo sells one-time items such as streak recovery. Guest's caveat: keep the subscription as the main focus and only add one-offs if they don't distract from it.

Duolingo showed the ATT prompt after the user had completed a few lessons, so tracking focuses on users who stick past the first sessions (the audience worth modelling).