Duolingo

One lesson a day now keeps a streak alive, lifting 7+ day streaks 40%. Duolingo used to require users to hit their full daily goal to keep a streak, then decoupled the two so a single lesson preserves it. Learners on a 7+ day streak rose by over 40%.

The old rule punished ambitious learners who set high goals; decoupling let them aim high without risking the habit. The 7+ day share served as a concrete early-habit metric to compare variants.

abtest.design

Duolingo's freemium playbook fits a huge funnel, not most apps. Freemium works for Duolingo because it teaches the basics of every language to a vast base of hobby learners and monetizes a slice. Phil Carter says hard paywalls convert 5x better and suit most apps; one client's freemium shift cut subscriber conversion by more than 50% and was pulled within weeks.

BoldVoice ignores basic learners and charges $150 to $200 a year to professionals whose alternative is a coach costing hundreds of dollars an hour; Praktika likewise targets serious career English. Anada Lakra warns against combining pieces of different apps' playbooks into a "mishmash strategy."

Sub Club

Repeated trending formats keep giving a famous brand fresh impressions. An article claimed Duolingo's numbers fell when it stopped its unhinged content. Each trending-format post is another chance to convert or bring a viewer back.

Adam Turaev notes the owl is wilder on social media than inside the app.

Sub Club

Its notifications are reverse-engineered and copied by other apps. A wrestling app team copied Duolingo's best notifications, sending at least one a day tied to past actions like beating a run record, plus "don't give up" messages to users who did not pay after onboarding.

They do not yet track which notifications drive clicks.

Subversive▶ 20:25

Duolingo's freemium playbook fits a huge funnel, not most apps. Freemium works for Duolingo because it teaches the basics of every language to a vast base of hobby learners and monetizes a slice. Phil Carter says hard paywalls convert 5x better and suit most apps; one client's freemium shift cut subscriber conversion by more than 50% and was pulled within weeks.

BoldVoice ignores basic learners and charges $150 to $200 a year to professionals whose alternative is a coach costing hundreds of dollars an hour; Praktika likewise targets serious career English. Anada Lakra warns against combining pieces of different apps' playbooks into a "mishmash strategy."

Lenny's Podcast▶ 34:34

"Try for $0.00" beat "Try for free" on the trial button. Button copy went from "Subscribe" to "Start my free trial" to "Try for free" to "Try for $0.00", which lifted both trial starts and trial-to-paid conversion. The graphs looked so green the team suspected a measurement error.

No size of uplift is given. It is one of the rarest CTA phrases in RevenueCat's paywall data, where "Continue" and "Subscribe" dominate. Eva of Solid Starts cites the big $0.00 button as showing how subtle the line on paywall tactics can be.

Lenny's Podcast

BirdBrain tunes how many of 10 answers a learner gets right. The system adjusts difficulty to keep learners motivated, which a host compared to Purpose balancing tough love with support.

App Masters

Others measure retention and category size against Duolingo. Opal reports early retention on par with or better than Duolingo, which is cited as proof a category can hold winners making over $1B a year. Andrew Maguire puts its value at roughly $4B to $5B as one of consumer's few giant outcomes.

It is also named a compounding winner unlikely to shrink, an example of consumer expansion through gems and app colors, and, for ClassDojo's founders, the only other company living or dying by whether users benefit.

Sub Club

Others measure retention and category size against Duolingo. Opal reports early retention on par with or better than Duolingo, which is cited as proof a category can hold winners making over $1B a year. Andrew Maguire puts its value at roughly $4B to $5B as one of consumer's few giant outcomes.

It is also named a compounding winner unlikely to shrink, an example of consumer expansion through gems and app colors, and, for ClassDojo's founders, the only other company living or dying by whether users benefit.

Subversive

New subjects live in the main app, not separate apps. Math first launched as its own app on the assumption that math and language learners barely overlap. Duolingo folded it into the main app after finding users want several subjects and that growing a separate app is extremely expensive.

Math, music and chess now sit in the course switcher. Life360 and Simply cite this as a sign second apps rarely succeed. Quote: "why are we playing hard mode when we can play easy mode?"

Superwall

Avoid the premium trap: paywalling free features works briefly, then stalls. Kansu says moving free features behind a paywall turns every metric green for about 6 months to a year, then stops working, and invites a rival to offer the same thing free within about two years.

Duolingo grew early because Rosetta Stone and similar rivals were paid and expensive. Its chief product officer coined the term "premium trap" for locking features to hit quarterly revenue. The claim rests on Kansu's judgment rather than published numbers.

Sub Club

New subjects live in the main app, not separate apps. Math first launched as its own app on the assumption that math and language learners barely overlap. Duolingo folded it into the main app after finding users want several subjects and that growing a separate app is extremely expensive.

Math, music and chess now sit in the course switcher. Life360 and Simply cite this as a sign second apps rarely succeed. Quote: "why are we playing hard mode when we can play easy mode?"

Sub Club

Others measure retention and category size against Duolingo. Opal reports early retention on par with or better than Duolingo, which is cited as proof a category can hold winners making over $1B a year. Andrew Maguire puts its value at roughly $4B to $5B as one of consumer's few giant outcomes.

It is also named a compounding winner unlikely to shrink, an example of consumer expansion through gems and app colors, and, for ClassDojo's founders, the only other company living or dying by whether users benefit.

Sub Club▶ 1:11:24

Duolingo's freemium playbook fits a huge funnel, not most apps. Freemium works for Duolingo because it teaches the basics of every language to a vast base of hobby learners and monetizes a slice. Phil Carter says hard paywalls convert 5x better and suit most apps; one client's freemium shift cut subscriber conversion by more than 50% and was pulled within weeks.

BoldVoice ignores basic learners and charges $150 to $200 a year to professionals whose alternative is a coach costing hundreds of dollars an hour; Praktika likewise targets serious career English. Anada Lakra warns against combining pieces of different apps' playbooks into a "mishmash strategy."

App Masters

Put AI usage in a higher tier to cover its compute costs. Duolingo had near-zero marginal cost per subscriber until Max placed AI features in a higher tier. Phil Carter recommends a basic plan with a taste of AI and a pro or max plan with full usage, as Duolingo Max, Perplexity, ChatGPT and others do.

Rik Haandrikman extends this to metered credits plus consumables, to "drop the floor and raise the ceiling." Max adds voice role plays with Lily.

App Masters▶ 11:50

AI is an accelerant for content and features, not a bolt-on chatbot. Duolingo uses AI mainly to produce content at higher quality and volume and to build features like conversation practice. Kansu says being able to teach a language is no threat unless it is also fun.

Explain My Answer was too costly at first due to LLM calls, then went free as costs fell and caching improved. Earlier personalization took hundreds of engineering hours.

App Masters

Copying free-first breaks when AI token costs scale. Koby Conrad copied Duolingo's free-first approach, since it waited six or seven years to charge, but Sunflower's token costs reached hundreds of thousands of dollars and investors wanted revenue, so he added a monthly paywall.

Quote: "turns out investors want to see money."

Subversive▶ 42:51

Put AI usage in a higher tier to cover its compute costs. Duolingo had near-zero marginal cost per subscriber until Max placed AI features in a higher tier. Phil Carter recommends a basic plan with a taste of AI and a pro or max plan with full usage, as Duolingo Max, Perplexity, ChatGPT and others do.

Rik Haandrikman extends this to metered credits plus consumables, to "drop the floor and raise the ceiling." Max adds voice role plays with Lily.

Sub Club▶ 39:55

The real competition is other screen time, so make learning fun. Churned users mostly go to Instagram, YouTube and other phone fun, not other education apps, so Duolingo invests heavily in fun to offset learning's extra effort.

Haandrikman adds that anyone can copy Duolingo's look, but its depth and community are hard to rebuild.

Sub Club▶ 6:43

"Try for $0.00" beat "Try for free" on the trial button. Button copy went from "Subscribe" to "Start my free trial" to "Try for free" to "Try for $0.00", which lifted both trial starts and trial-to-paid conversion. The graphs looked so green the team suspected a measurement error.

No size of uplift is given. It is one of the rarest CTA phrases in RevenueCat's paywall data, where "Continue" and "Subscribe" dominate. Eva of Solid Starts cites the big $0.00 button as showing how subtle the line on paywall tactics can be.

Subversive▶ 25:19

Letting users pick their reminder day added friction yet raised conversion. Duolingo normally reminds users about 2 days before a trial ends, then tested letting them choose the reminder day. The extra step drew attention to the reminder, raised confidence to start the trial, and lifted conversion.

Users felt safe waiting until day five or six to judge the value, and it may reduce day-zero cancellations. No conversion numbers were shared.

Subversive▶ 21:58

Others measure retention and category size against Duolingo. Opal reports early retention on par with or better than Duolingo, which is cited as proof a category can hold winners making over $1B a year. Andrew Maguire puts its value at roughly $4B to $5B as one of consumer's few giant outcomes.

It is also named a compounding winner unlikely to shrink, an example of consumer expansion through gems and app colors, and, for ClassDojo's founders, the only other company living or dying by whether users benefit.

Subversive

Machine learning chooses between paid ads and ads for Duolingo's own subscription. After each lesson a model picks a programmatic ad or an internal ad for Super, Max or the Duolingo English Test, optimizing LTV with likelihood to purchase as the objective, and backs off after someone sees the same Super ad three times.

It replaced fixed ratios of ads per tier, which Anmol Tiwari called "quite an inefficient way to operate," using signals like past taps on a plan and interest in video call. A bandit picks which Super creative shows to whom, and Tiwari says personalization has barely scratched the surface.

Sub Club▶ 15:44

Give users clarity and confidence to buy instead of pushing harder. Duolingo's trial paywalls show a day-by-day timeline, such as a 7-day trial with a day-five notification, and state refund amounts when users switch plans. Pages that spell out the trial timeline won A/B tests.

This eases subscription aversion and fear of scams, especially outside the US. Tiwari warns that without trust you get poor conversion or complaints about unexpected charges. Quote: "Our goal is not to persuade users to try this. Our goal is to give them clarity and confidence to purchase."

Sub Club▶ 21:41

Letting users pick their reminder day added friction yet raised conversion. Duolingo normally reminds users about 2 days before a trial ends, then tested letting them choose the reminder day. The extra step drew attention to the reminder, raised confidence to start the trial, and lifted conversion.

Users felt safe waiting until day five or six to judge the value, and it may reduce day-zero cancellations. No conversion numbers were shared.

Sub Club▶ 20:46

Free trials help even inside an already-free product. Kansu resisted trials for years because the app was free, but trials removed payment hesitation for interested, skeptical users, even when the first paid plan (no ads plus offline) needed no real trial experience.

Tiwari says premium benefits like unlimited energy or no ads only make sense once felt, so trials are the main lever for paid features. Skylight, by contrast, saw no lift from its own trial test.

Sub Club

Cutting the trial from 14 to 7 days doubled experimentation velocity. Duolingo ran 14-day trials until the end of 2024. Moving to 7 days halved the wait to read conversion experiments and, combined with allowing more trials per user, brought more conversions over time.

Losing tests get killed sooner, and the gain compounds over a quarter or year. Because the shorter trial and repeat trials were changed together, each one's share is unclear.

Sub Club▶ 25:37

Upsells framed as onboarding questions beat standard upsells for new users. Instead of dropping a standard Super upsell into onboarding, Duolingo asked whether users want to learn with or without ads, or faster versus at their own pace, and positioned the offer as that choice. Opt-in rose significantly, one of last year's biggest wins.

Brand-new users are far more likely to convert, so their offers should fit where they are.

Sub Club▶ 51:36

Framing a trial as keeping progress makes users start it. When hearts or energy run out, Duolingo positions the choice as losing progress unless the user continues, and users were much more likely to start a trial.

The screen still shows progress and offers paying gems as the alternative to starting the trial so the lesson can be finished.

Sub Club▶ 1:00:56

Track day-zero cancellations as a read on trial intent. The share of users who cancel right after starting a trial signals intent, and once users cancel it is very hard to get them to un-cancel, so problems must be caught earlier.

The host adds that paid trials can give ad networks a cleaner signal than free trials that people switch off immediately.

Sub Club▶ 30:30

Adapt trials to regional norms, like nominal paid trials in China. Almost no subscription service in China offers a zero-dollar trial; users expect a nominal fee around $1 or 99 cents, partly because auto-renewal is unfamiliar.

A paid trial also makes clear the product costs money, and similar models appear in parts of Europe. Duolingo is still working out how to serve these markets.

Sub Club▶ 11:11

Offer multiple trials because users are at different life stages. A user who did not grasp the benefits or was too busy the first time gets another chance, so over time they decide three or four times whether to buy. Combined with the 7-day trial, this produced net more conversions.

Apple allows an intro offer once per product, so workarounds are atypical and not easy. The cost of repeat trials is not discussed.

Sub Club▶ 54:33

A no-strings 3-day free taste reaches hesitant users and paying subscribers. Duolingo gives a 3-day trial-like experience of Super or Max with no auto-conversion; users can convert during or at the end. Because Apple blocks a second free trial, it also upsells Super subscribers to Max, returning them to their original tier afterward.

It has been effective for both free-to-paid and paid-to-paid conversion. A taste can be a single feature, such as one isolated video call, which pushes users toward Max. No-card tastes are hard to explain because users expect an automatic charge.

Sub Club▶ 58:00

Frame free tastes as an earned, time-limited reward. Copy says "you just earned this," states the three-day length, and uses timers for urgency. Duolingo moved the reward chest from the learning path to after a lesson because earning it there felt better.

Users must know a Max taste won't change what they are charged, or Super subscribers may cancel.

Sub Club▶ 49:15

End a free taste plainly and make returning easy. A home drawer states that the benefits have ended and offers two options: continue, or explore and resubscribe.

Tiwari puts clarity that it ended ahead of loss-aversion copy.

Sub Club▶ 1:02:00

Gamification retains users by making them feel they are learning. Connor Zwick of Speak points to 700-day streaks from people who say they cannot speak the language, and Dris notes hundreds of millions of users though he knows no one who learned a language on it.

Dris's lesson: apps that feel like they work win. Speak instead bets that personal relevance and a relationship with the AI motivate better than streaks and dopamine loops.

Sub Club▶ 16:42

Ads run only after a completed lesson, with dicey categories blocked. Duolingo shows ads after a lesson, never before or during, and retention stayed mostly neutral across ad tests. It blocks gambling, sketchy games and anything unsafe for young audiences or brands, even at a revenue cost.

The team dogfoods placements, since an app-open ad can look fine in metrics but feel awful. It started on AdMob, then added Facebook, Flurry and Unity, and asked networks to drop formats like ads with three exit buttons. Mojo's interviewer still cites the full-screen, unskippable post-lesson ad as aggressive monetization.

Sub Club

Free trials help even inside an already-free product. Kansu resisted trials for years because the app was free, but trials removed payment hesitation for interested, skeptical users, even when the first paid plan (no ads plus offline) needed no real trial experience.

Tiwari says premium benefits like unlimited energy or no ads only make sense once felt, so trials are the main lever for paid features. Skylight, by contrast, saw no lift from its own trial test.

Sub Club

Gamification retains users by making them feel they are learning. Connor Zwick of Speak points to 700-day streaks from people who say they cannot speak the language, and Dris notes hundreds of millions of users though he knows no one who learned a language on it.

Dris's lesson: apps that feel like they work win. Speak instead bets that personal relevance and a relationship with the AI motivate better than streaks and dopamine loops.

App Masters▶ 14:32

Speaking practice lagged because speech models failed on accented learners. Early on, speech recognition had roughly an 80% error rate for strongly accented learners, so Duolingo and Rosetta Stone offered little speaking. Speak used that gap as its initial wedge.

Zwick sees Duolingo as complementary and its high stock price as validation of the market. Duolingo has since added speaking practice in Max.

App Masters▶ 16:57

Repeated trending formats keep giving a famous brand fresh impressions. An article claimed Duolingo's numbers fell when it stopped its unhinged content. Each trending-format post is another chance to convert or bring a viewer back.

Adam Turaev notes the owl is wilder on social media than inside the app.

Superwall▶ 16:55

Others measure retention and category size against Duolingo. Opal reports early retention on par with or better than Duolingo, which is cited as proof a category can hold winners making over $1B a year. Andrew Maguire puts its value at roughly $4B to $5B as one of consumer's few giant outcomes.

It is also named a compounding winner unlikely to shrink, an example of consumer expansion through gems and app colors, and, for ClassDojo's founders, the only other company living or dying by whether users benefit.

Subversive▶ 7:00

Streak repair, sold for about $1 to $3, was the first purchase. Users cared enough about their streak that a good share would pay a dollar or two to repair a broken one, which proved Duolingo could monetize extras while keeping education free.

Luis von Ahn had publicly promised never to add ads, in-app purchases or subscriptions; Duolingo built all three, in that order, monetizing only the "bells and whistles." Margins plans the same model: free core features, with streak freezes, streak rescue and cosmetics paid.

Sub Club▶ 6:49

Free users are the growth engine, so free value stays strong. About 10% of users subscribe and subscriptions bring in over 80% of revenue, but word of mouth from free users drives most growth whether or not they ever pay. Duolingo moved Practice Hub and Explain My Answer from paid to free and accepted the revenue loss.

Explain My Answer started in the top paid tier because LLM costs were high. When costs fell, a test showed revenue dropped if it went free, and Duolingo moved it to free anyway. Quote from CPO Cem Kansu: "free users are our marketing engine."

Superwall▶ 25:50

Avoid the premium trap: paywalling free features works briefly, then stalls. Kansu says moving free features behind a paywall turns every metric green for about 6 months to a year, then stops working, and invites a rival to offer the same thing free within about two years.

Duolingo grew early because Rosetta Stone and similar rivals were paid and expensive. Its chief product officer coined the term "premium trap" for locking features to hit quarterly revenue. The claim rests on Kansu's judgment rather than published numbers.

Superwall▶ 24:25

The first business model, crowdsourced translation, failed on price and quality. Duolingo launched as a free website that sold learners' sentence translations to outlets like CNN and BuzzFeed. It earned a little, but learners translated poorly and rivals could always undercut on price.

Fixing quality pulled resources from teaching toward running a translation business. By 2016, when Kansu joined, the model was dropped and the company was searching for the next one.

Superwall

Streak repair, sold for about $1 to $3, was the first purchase. Users cared enough about their streak that a good share would pay a dollar or two to repair a broken one, which proved Duolingo could monetize extras while keeping education free.

Luis von Ahn had publicly promised never to add ads, in-app purchases or subscriptions; Duolingo built all three, in that order, monetizing only the "bells and whistles." Margins plans the same model: free core features, with streak freezes, streak rescue and cosmetics paid.

Superwall

Monetizing required investor pressure and a mission-driven culture shift. Around 2016 a Capital G board member told Luis the next round should be the last before a monetization plan, and Kansu's sole job became figuring out how to make money.

Many of the roughly 60 employees saw the company as almost a nonprofit. The message became that revenue funds the mission. Duolingo already had tens of millions of users when it started monetizing.

Superwall

A janky one-month ad-free purchase proved demand before building a subscription. After reviews asked to pay to remove ads, Duolingo tested a one-time ad-free month at $3, $5 and $10. It sold far better than expected, leading to Duolingo Plus in 2017 with no ads, offline use and a "support the mission" line.

Early buyers were roughly half mission supporters and half people wanting no ads. Quote: "Even though we created the most bizarre package to sell, people want to buy this."

Superwall▶ 21:30

Unlimited hearts tied the subscription to real usage friction. Folding the hearts pacing mechanic into the subscription with unlimited hearts, plus custom practice, was a turning point that lifted both conversion and subscription retention.

Many users want new content rather than practicing old content, so removing the limit had real value.

Superwall▶ 13:55

Ads run only after a completed lesson, with dicey categories blocked. Duolingo shows ads after a lesson, never before or during, and retention stayed mostly neutral across ad tests. It blocks gambling, sketchy games and anything unsafe for young audiences or brands, even at a revenue cost.

The team dogfoods placements, since an app-open ad can look fine in metrics but feel awful. It started on AdMob, then added Facebook, Flurry and Unity, and asked networks to drop formats like ads with three exit buttons. Mojo's interviewer still cites the full-screen, unskippable post-lesson ad as aggressive monetization.

Superwall▶ 4:00

Count complaints per million DAUs to put backlash in perspective. In 2016, while deciding whether to launch ads, Kansu found complaints about the missing Finnish course were 3x the complaints about ads.

Pair complaint counts with retention, app closes and review trends from A/B tests; retention was neutral to slightly down in most ad tests. Without this, every negative comment feels like the world is falling apart.

Superwall▶ 7:47

Machine learning chooses between paid ads and ads for Duolingo's own subscription. After each lesson a model picks a programmatic ad or an internal ad for Super, Max or the Duolingo English Test, optimizing LTV with likelihood to purchase as the objective, and backs off after someone sees the same Super ad three times.

It replaced fixed ratios of ads per tier, which Anmol Tiwari called "quite an inefficient way to operate," using signals like past taps on a plan and interest in video call. A bandit picks which Super creative shows to whom, and Tiwari says personalization has barely scratched the surface.

Superwall▶ 12:35

Direct, character-led brand ads are replacing programmatic volume. In the US, one of eight characters introduces a direct advertiser's ad with a short custom script, such as Lily joking about laundry before a detergent ad. The 2026 goal is more direct volume and much less programmatic.

Scaling is slow because each advertiser needs the custom format explained.

Superwall

Give users clarity and confidence to buy instead of pushing harder. Duolingo's trial paywalls show a day-by-day timeline, such as a 7-day trial with a day-five notification, and state refund amounts when users switch plans. Pages that spell out the trial timeline won A/B tests.

This eases subscription aversion and fear of scams, especially outside the US. Tiwari warns that without trust you get poor conversion or complaints about unexpected charges. Quote: "Our goal is not to persuade users to try this. Our goal is to give them clarity and confidence to purchase."

Superwall▶ 23:35

"Try for $0.00" beat "Try for free" on the trial button. Button copy went from "Subscribe" to "Start my free trial" to "Try for free" to "Try for $0.00", which lifted both trial starts and trial-to-paid conversion. The graphs looked so green the team suspected a measurement error.

No size of uplift is given. It is one of the rarest CTA phrases in RevenueCat's paywall data, where "Continue" and "Subscribe" dominate. Eva of Solid Starts cites the big $0.00 button as showing how subtle the line on paywall tactics can be.

Superwall▶ 1:56

Free trials help even inside an already-free product. Kansu resisted trials for years because the app was free, but trials removed payment hesitation for interested, skeptical users, even when the first paid plan (no ads plus offline) needed no real trial experience.

Tiwari says premium benefits like unlimited energy or no ads only make sense once felt, so trials are the main lever for paid features. Skylight, by contrast, saw no lift from its own trial test.

Superwall▶ 9:09

Users asked for human tutors but were too scared to use them. Duolingo built an on-demand human tutor tier after years of requests, but no one used it despite heavy iteration. The later AI video call with Lily worked because it removed the fear of speaking to a person.

Kansu says video call with Lily has helped his own Spanish more than anything in nine years. Quote: "Do not take user feedback at face value."

Superwall▶ 17:30

AI is an accelerant for content and features, not a bolt-on chatbot. Duolingo uses AI mainly to produce content at higher quality and volume and to build features like conversation practice. Kansu says being able to teach a language is no threat unless it is also fun.

Explain My Answer was too costly at first due to LLM calls, then went free as costs fell and caching improved. Earlier personalization took hundreds of engineering hours.

Superwall

Big revenue tradeoffs go to senior leaders; teams decide small ones. Duolingo runs about 400 concurrent A/B tests and often weighs learning gains against revenue or DAU, with no single metric to settle it. Large strategic hits go to senior people, including the CEO, for sign-off.

Superwall▶ 46:00

Use 3 to 6 month holdouts for effects that shift over time. Long holdouts suit things like leaderboards that take 10 weeks to climb or push notifications whose novelty fades; a new notification can turn negative after 6 weeks.

Duolingo once ran a one-to-two-year ads holdout but now avoids year-long ones because teams forget them and bugs pollute the group.

Superwall▶ 34:25

New subjects live in the main app, not separate apps. Math first launched as its own app on the assumption that math and language learners barely overlap. Duolingo folded it into the main app after finding users want several subjects and that growing a separate app is extremely expensive.

Math, music and chess now sit in the course switcher. Life360 and Simply cite this as a sign second apps rarely succeed. Quote: "why are we playing hard mode when we can play easy mode?"

Superwall▶ 44:25

Pick new areas by mobile fit, existing demand and internal champions. Math, music and chess can be taught in fun, bite-sized mobile lessons, already have big markets, and each had tenured builders eager to build it. Each took 6 to 12 months from idea to launch.

Coding was skipped as too hard on mobile and running as too offline.

Superwall▶ 51:55

Build for a general audience rather than personas. People learn languages for jobs, moving countries, heritage and more, so picking one persona cuts off the long tail. Duolingo asks new PMs to stop building persona maps.

Superwall▶ 21:12

Guide users with design, not text, because they don't read. On mobile, attention is split and users focus on roughly the core 30% of a screen.

Quote: "if you need to write a tutorial on what to do, you've already lost it."

Superwall▶ 18:53

The real competition is other screen time, so make learning fun. Churned users mostly go to Instagram, YouTube and other phone fun, not other education apps, so Duolingo invests heavily in fun to offset learning's extra effort.

Haandrikman adds that anyone can copy Duolingo's look, but its depth and community are hard to rebuild.

Superwall