The Science of Sticky Apps study guide · Lesson 9 of 11
Case Studies in Stickiness
Duolingo: the gamification benchmark · TikTok & Instagram · Snapchat & FarmVille
Unofficial study guide for Hooked: How to Build Habit-Forming Products by Nir Eyal, Ryan Hoover. FlashFeed is not affiliated with the author or publisher.
This is the lesson's study guide from the FlashFeed app, plus its key facts and a self-test.
What you'll be able to do
- Explain Duolingo's demotion threat and variable XP as loss-aversion and variable-ratio mechanics, and how AI scaled its content.
- Cite Duolingo's DAU/MAU ratio versus the industry average, its day-over-day retention, and its market-cap growth.
- Contrast TikTok's interest graph and completion signal with Instagram's social-graph moat.
- Explain Snapchat's ephemeral content and Group Streaks as FOMO and loss-aversion engines.
- Explain FarmVille's social-obligation gift-sending loops and cite its peak monthly active users.
Duolingo: the gamification benchmark
Duolingo didn't invent gamification, but it operationalized it at a scale that is now the reference case. The mechanics below are straightforward; what is instructive is how precisely each one maps to a known psychological lever.
The fire and the hourglass
The streak counter is a number that grows by one each day you open the app. Duolingo then threatens to zero it: the fire icon dims, an hourglass appears, and a push notification warns that your streak dies at midnight. The mechanism is loss aversion — people feel losses roughly twice as intensely as equivalent gains — so the dread of losing a 47-day streak outweighs the mild satisfaction of extending it. The counter does not measure fluency; it measures the strength of a daily habit loop. That is precisely the point.
Leagues and the fear of dropping
Duolingo's weekly leaderboards sort users into named tiers and promote or demote them based on XP. The fear of demotion reliably outweighs the desire to advance — users who sit near the cutoff line earn far more XP than users safely in the middle. Loss aversion, again. Duolingo's DAU/MAU ratio — the fraction of monthly users who open the app daily — reached 36%, roughly double the 20% industry average, which is the metric most analysts credit to these retention mechanics. Day-over-day retention of daily active users sits at 55%. The company's market cap grew from 5 billion at IPO to 22 billion by Q1 2025. When generative AI landed, Duolingo used it to compress course production from 12+ months per course down to weeks, launching 148 new courses in under a year — proof that the content engine now scales as fast as the engagement model.
XP that isn't fixed
Flat rewards train users to expect a predictable payout and, eventually, to stop caring. Variable XP — bonus rounds, surprise multipliers, Perfect Lesson celebrations — keeps the distribution uncertain. Variable-ratio reinforcement produces the most durable behavior: the next lesson might yield double XP, or nothing extra, and you won't know until you finish. That unpredictability is what makes starting the next lesson worth the cost of opening the app.
TikTok & Instagram
These two platforms made opposite bets about what keeps users — one on the social web, one on a machine-built model of individual taste — and both won, which is why the distinction matters.
What you're drawn to
Every feed algorithm makes a structural choice: organize content around who you know (the social graph) or around what you consume (the interest graph). TikTok chose the latter. The For You Page is cold to strangers — a new account with zero followers sees a fully personalized feed within hours, assembled entirely from behavioral signals. The signals the algorithm weights are completion rate, shares, comments, follows, and re-watches. Completion and sharing rank highest. The result is a feed that reflects your revealed preferences rather than your social identity, which is why it consistently surfaces content from creators you have never encountered.
Watched to the last second
Completion rate — the fraction of a video watched to the final second — is TikTok's dominant ranking signal, weighted above likes, comments, and follower count. This creates a structural incentive for creators: a video that hooks but doesn't hold is actively penalized. The underlying recommendation engine uses matrix factorization, a technique that finds latent patterns across millions of user-video interactions to predict which unseen video a given user is most likely to finish. Each swipe updates the model. The result is a variable-ratio delivery schedule: most videos are ordinary; occasionally one lands exactly right, which is enough to keep swiping.
The graph is the moat
Instagram copied short-form video with Reels and matched TikTok's content quality. It did not close the gap, and the reason is structural. Research found that users value connections with family and close friends at a qualitatively different level than content-only connections. The social graph — years of accumulated relationships, shared photos, and mutual follows — cannot be ported. It is the moat that content quality alone cannot breach, and it is why Instagram remained the second-largest short-form platform despite replicating the surface mechanics of its rival.
Snapchat & FarmVille
Two case studies in obligation-based retention. One built its entire product thesis on content that vanishes; the other was a browser game that held 74 million people not through enjoyment but through reciprocal social debt.
Disappearing by design
Ephemeral content — posts that auto-delete after viewing or a set time window — is often described as a privacy feature. It was always primarily a behavioral one. When a Snap disappears, it creates two effects: FOMO (the recipient must open it now, not later) and reduced social anxiety (low-quality posts cost nothing to send because no permanent record exists). Disappearing makes posting feel cheap, which makes it frequent, which feeds the habit loop. Snapchat's streak mechanic then converts that frequency into a loss-aversion engine — you're not protecting quality content, you're protecting a number. Ephemeral content was the mechanism; the daily open was the goal.
Streaks go collective
Individual streaks require only two users. In September 2025 Snapchat launched Group Streaks, which extended the obligation to multi-person chains: most members of the group must Snap for the collective streak to survive. Non-participation now costs you the group's streak, not just a single friendship. This multiplies social pressure by the size of the group — a well-understood principle that FOMO and social obligation scale with the number of relationships implicated. The architecture is the same loss-aversion loop; the blast radius is larger.
The proof of concept
FarmVille (2009, peak 74 million monthly active users) was the earliest mass demonstration that obligation loops, not fun, could sustain engagement. Players needed gifts from real Facebook friends to unlock crops and buildings; those friends needed gifts in return. Progress was structurally coupled to other people's participation. The game wasn't particularly enjoyable — most players described it as a chore — but the social-obligation loop and loss aversion over planted crops kept them returning. FarmVille proved that reciprocal mechanics could hold millions of users without the product being intrinsically rewarding.
Key terms
- Demotion threat
- A league mechanic where falling below a tier's cutoff drops you a rank — the fear of dropping is engineered to motivate more than the desire to advance.
- Interest graph
- A machine-curated map of what you are drawn to rather than who you know. TikTok's For You Page replaced the social graph with an interest graph, letting a follower-less user get viral distribution in 24 hours.
- Ephemeral content
- Posts that auto-delete after viewing or a set window — engineered to create urgency and to lower the social anxiety of sharing imperfect moments.
Key concepts
- This platform copied short-form video and matched its rival's content quality — but still couldn't close the gap.
- The reason is structural: its moat is the social graph, years of accumulated real-world relationships that can't be ported.
- It launched Reels to counter its rival and stayed the second-largest short-form platform on the strength of family and close-friend ties.
Snapchat
- This app built its thesis on ephemeral content — posts that auto-delete after viewing or a set window.
- Disappearing posts create FOMO and lower social anxiety, making posting feel cheap and frequent — feeding a daily-open habit.
- Its streak mechanic turns that frequency into a loss-aversion engine; in September 2025 it launched Group Streaks to extend the obligation to whole chat groups.
Key facts
The fire and the hourglass
Duolingo's streak turns a counter into a daily anxiety. The fire emoji and the hourglass warning create a low-grade dread that resolves only one way: opening the app. The mechanism is loss aversion — you log in to avoid losing what you've built.
Leagues and the fear of dropping
Duolingo's weekly leaderboards promote and demote players between tiers. The design exploits a known asymmetry: the fear of dropping a league tier is more motivating than the desire to climb one. Loss aversion, dressed as competition.
XP that isn't fixed
Lesson completion yields variable XP — bonus rounds, surprise multipliers, and 'Perfect Lesson' celebrations. The reward isn't a flat payout; it's unpredictable, which is exactly what keeps the next lesson worth starting.
AI Collapsed The Timeline
Generative AI let Duolingo launch 148 new courses in under a year, down from more than 12 months of production time per course — a step-change in content breadth that feeds retention.
Double The Industry Average
Duolingo's DAU/MAU ratio reached 36% — roughly double the 20% industry average, and the number analysts point to as evidence its retention mechanics are working.
What you're drawn to
TikTok swapped the human-curated social graph for a machine-curated interest graph. The For You Page reads completion rate, shares, comments, follows, and re-watches — weighting completion and sharing most heavily.
Watched to the last second
TikTok's algorithm weighs completion rate — watch-through to the final second — above almost every other signal. That single weighting creates a race among creators to hold attention to the end, and a per-user model refined by matrix factorization with every swipe.
The graph is the moat
Even as Instagram chased content, the social graph stayed its moat. Research found users value connections with family and close friends far above content-only connections — so the graph is functionally irreplaceable even when a competitor clones the content quality.
Family Ties Hit Different
Research behind the Instagram case study found that users value connections with family and close friends at a qualitatively different level than purely content-based follows.
Sharing Beats Social Proof
TikTok's ranking weighs completion and sharing above comments, follows, and re-watches — a bet on revealed preference, what you actually finish and pass on, over social signals like who you follow.
Disappearing by design
Snapchat built around ephemeral content — but that wasn't a privacy feature, it was behavioral design. Disappearing Snaps create urgency (FOMO) and lower the social anxiety of sharing imperfect content. Vanishing makes posting feel cheap, which makes it frequent.
Streaks go collective
Snapchat's September 2025 Group Streaks extend the obligation to multi-person relationships: collective Snapping from most of a group keeps the streak alive. Now non-participation costs you the group, not just a friend — multiplying the social pressure.
The proof of concept
FarmVille (2009) was the prototype for social-obligation loops. Its gift-sending mechanics required other players to participate — your progress depended on theirs, and theirs on yours. The loop ran on social obligation and loss aversion, not on the game being fun.
Protecting A Number
Snapchat's streak mechanic turns posting frequency into a loss-aversion engine: users keep opening the app to protect a number, not because that day's content matters.
74 Million Farmers
FarmVille peaked at 74 million monthly active users in 2009 — the earliest mass demonstration that an obligation loop could sustain engagement at that scale.
By the numbers
- New Duolingo courses launched in under a year using generative AI (vs. 12+ months each before)
- 148 courses
- Duolingo's DAU/MAU ratio
- 36 %
- Industry-average DAU/MAU ratio (Duolingo sits well above this)
- 20 %
- Duolingo's day-over-day retention of daily active users
- 55 %
- Duolingo's market cap at IPO
- 5000000000 USD
- Duolingo's market cap by Q1 2025
- 22000000000 USD
- FarmVille's peak monthly active users (2009-era)
- 74 million-users
Matched pairs
Each item on the left goes with the one on the right.
- TikTok
- Interest graph + variable-ratio For You Page
- Social graph moat + algorithmic Reels broadcast
- Snapchat
- Ephemeral content + loss-aversion streaks
- FarmVille
- Social-obligation gift-sending loops
Self-test
Answer each question from memory, then check your work against the answer key.
1. Lesson completion yields **variable XP** — bonus rounds, surprise multipliers, and '___' celebrations. The reward isn't a flat payout; it's unpredictable, which is exactly what keeps the next lesson worth starting.
- ACombo Bonus
- BLegendary Level
- CPerfect Lesson
- DLightning Round
2. What did generative AI change about Duolingo's content production?
- AIt removed the need for leaderboards
- BIt let them launch 148 new courses in under a year, instead of 12+ months per course
- CIt replaced the streak with an AI tutor
- DIt auto-translated user posts
3. Duolingo's league demotion threat works because of which psychological lever?
- ALoss aversion — the fear of dropping a tier outweighs the desire to climb
- BAnchoring — the first XP number sets expectations
- CReciprocity — users feel obliged to repay the app
- DSocial proof — users copy what top-ranked players do
4. Duolingo's variable XP (bonus rounds, surprise multipliers) is an example of which reinforcement schedule?
- AContinuous reinforcement
- BFixed-ratio reinforcement
- CFixed-interval reinforcement
- DVariable-ratio reinforcement
5. How did generative AI change Duolingo's course production timeline?
- AIt compressed course-building from 12+ months each down to weeks
- BIt slowed launches to ensure higher quality
- CIt eliminated the need for human course designers entirely
- DIt doubled the length of each course
6. What is Duolingo's DAU/MAU ratio, the metric analysts credit to its retention mechanics?
- A55%
- B20%
- C74%
- D36%
7. Duolingo's 36% DAU/MAU ratio is roughly double what industry average?
- A10%
- B20%
- C55%
- D30%
8. What is Duolingo's day-over-day retention of its daily active users?
- A20%
- B22%
- C55%
- D36%
9. How did Duolingo's market cap change from its IPO to Q1 2025?
- AFrom $5 billion to $12 billion
- BFrom $2 billion to $10 billion
- CFrom $5 billion to $22 billion
- DFrom $10 billion to $22 billion
10. Which figure is NOT one of Duolingo's reported metrics?
- AAn 80% DAU/MAU ratio
- BA 55% day-over-day retention rate
- CA $22 billion market cap by Q1 2025
- DA 36% DAU/MAU ratio
11. Which signal does TikTok's algorithm weight above almost all others?
- ANumber of followers a creator has
- BCompletion rate — watch-through to the final second
- CHow recently the video was posted
- DWhether you follow the creator
12. TikTok's For You Page is built around which organizing principle?
- AThe location graph — where you are
- BThe social graph — who you know
- CThe follower graph — who follows you
- DThe interest graph — what you consume
13. Research cited in the Instagram case found users value which connections at a qualitatively different level?
- AConnections with creators discovered algorithmically
- BConnections with brands and advertisers
- CConnections with the highest-follower accounts
- DConnections with family and close friends
14. Why did Instagram fail to close the gap with TikTok despite matching content quality with Reels?
- AIt did not weight completion rate
- BIts recommendation algorithm was technically inferior
- CThe social graph — years of accumulated relationships — cannot be ported
- DIt had far fewer monthly users to learn from
15. Which two signals does TikTok's algorithm weight most heavily?
- ACompletion and sharing
- BLikes and follower count
- CComments and posting recency
- DFollows and re-watches
16. According to the case study, when a Snap disappears it produces two behavioral effects. Which pair?
- ACompetitiveness and a fear of losing streaks
- BFOMO for the recipient and reduced social anxiety for the sender
- CNostalgia and a sense of exclusivity
- DTrust in the app's privacy and reduced screen-time guilt
17. How does the case study characterize Snapchat's ephemeral content?
- AA moderation tool to limit harmful posts
- BA monetization feature for ads
- CPurely a privacy feature with no behavioral aim
- DPrimarily behavioral design, not a privacy feature
18. What did Snapchat's September 2025 Group Streaks change about the streak mechanic?
- AThey capped streaks at a fixed number of days
- BThey removed streaks to reduce user anxiety
- CThey extended the obligation to multi-person chains, multiplying social pressure
- DThey made streaks purely individual again
19. Snapchat's streak mechanic converts posting frequency into an engine driven by which lever?
- ALoss aversion — protecting a number, not the content
- BScarcity of storage space
- CNovelty — always showing fresh content
- DSocial proof — mimicking popular users
20. Which principle explains why Group Streaks increase pressure more than individual streaks?
- AStreaks reset faster with more participants
- BFOMO and social obligation scale with the number of relationships implicated
- CGroup members can see each other's private messages
- DLarger groups earn more in-app currency
21. What was FarmVille's peak monthly active users?
- A36 million
- B22 million
- C148 million
- D74 million
22. What mechanic structurally coupled a FarmVille player's progress to other people's participation?
- AEphemeral crops that vanished after viewing
- BA shared leaderboard that reset weekly
- CNeeding gifts from real Facebook friends to unlock crops and buildings
- DA subscription that friends had to co-sign
23. What did FarmVille prove about what sustains engagement?
- AVariable rewards are the sole driver of retention
- BReciprocal obligation loops can hold millions without the product being intrinsically rewarding
- COnly genuinely fun games can retain large audiences
- DContent quality always beats social mechanics
24. In what year did FarmVille launch and reach its mass-engagement peak?
- A2009
- B2012
- C2005
- D2015
25. Which statement about FarmVille is NOT supported by the case study?
- AIt peaked at 74 million monthly active users
- BPlayers returned mainly because the game was intrinsically fun
- CLoss aversion over planted crops helped retain players
- DProgress depended on friends sending gifts
Go deeper
- ReadHooked by Nir Eyal, Ryan Hoover · on Amazon ↗
- StudyThe Science of Sticky Apps in the FlashFeed app
These are affiliate links: if you buy a book through one, FlashFeed may earn a small commission at no extra cost to you. The commission helps pay for FlashFeed's servers.
As an Amazon Associate we earn from qualifying purchases.
Answer key
- 1. C Perfect Lesson
- 2. B It let them launch 148 new courses in under a year, instead of 12+ months per courseGenerative AI collapsed course-building time from 12+ months each to 148 new courses in under a year — a step-change in content breadth, which feeds retention by giving users more to do.
- 3. A Loss aversion — the fear of dropping a tier outweighs the desire to climbThe demotion threat exploits loss aversion: users near the cutoff earn far more XP than those safely mid-tier because the dread of dropping outweighs the pull of advancing.
- 4. D Variable-ratio reinforcementUnpredictable payouts are variable-ratio reinforcement, which produces the most durable behavior — you won't know if the next lesson yields double XP or nothing until you finish it.
- 5. A It compressed course-building from 12+ months each down to weeksAI collapsed production from 12+ months per course down to weeks, enabling 148 new courses in under a year — the content engine now scales as fast as the engagement model.
- 6. D 36%Duolingo's DAU/MAU ratio reached 36%, roughly double the 20% industry average. 55% is its day-over-day retention; 74 million was FarmVille's peak MAU.
- 7. B 20%The industry-average DAU/MAU ratio is about 20%, so Duolingo's 36% is roughly double it. 55% is Duolingo's own day-over-day retention figure.
- 8. C 55%Day-over-day retention of daily active users sits at 55%. 36% is the DAU/MAU ratio and 20% is the industry-average DAU/MAU.
- 9. C From $5 billion to $22 billionDuolingo's market cap grew from $5 billion at IPO to $22 billion by Q1 2025.
- 10. A An 80% DAU/MAU ratioDuolingo's DAU/MAU ratio is 36%, not 80%. The 55% retention and $22 billion Q1 2025 market cap are both real reported figures.
- 11. B Completion rate — watch-through to the final secondCompletion rate is weighted above almost all other signals, pushing creators to hold attention to the last second. Following is largely irrelevant on the interest graph.
- 12. D The interest graph — what you consumeTikTok organizes content around the interest graph (what you consume), not the social graph. A follower-less new account gets a fully personalized feed within hours from behavioral signals alone.
- 13. D Connections with family and close friendsResearch found users value family and close-friend connections at a qualitatively different level than content-only ones — the social graph Instagram's moat rests on, which TikTok's interest graph can't replicate.
- 14. C The social graph — years of accumulated relationships — cannot be portedInstagram's moat is the social graph: users value family and close-friend connections at a qualitatively different level than content-only ones, and that graph can't be copied even when content quality is matched.
- 15. A Completion and sharingAmong completion, shares, comments, follows, and re-watches, completion and sharing rank highest — reflecting revealed preference rather than social identity.
- 16. B FOMO for the recipient and reduced social anxiety for the senderDisappearing content creates FOMO — the recipient must open it now, not later — and reduces social anxiety, because imperfect posts cost nothing when no permanent record exists. That makes posting feel cheap, which makes it frequent.
- 17. D Primarily behavioral design, not a privacy featureEphemeral content is often described as privacy, but it was always primarily behavioral: disappearing posts create FOMO and lower the anxiety of sharing imperfect content, making posting cheap and frequent.
- 18. C They extended the obligation to multi-person chains, multiplying social pressureGroup Streaks require most members to Snap for the collective streak to survive, so non-participation costs the group's streak — the same loss-aversion loop with a larger blast radius.
- 19. A Loss aversion — protecting a number, not the contentThe streak converts frequency into a loss-aversion engine — you're protecting a number, not quality content. The daily open is the real goal.
- 20. B FOMO and social obligation scale with the number of relationships implicatedFOMO and social obligation scale with the number of relationships implicated, so a group streak multiplies pressure by the group's size versus a two-person individual streak.
- 21. D 74 millionFarmVille (2009) peaked at 74 million monthly active users, the earliest mass demonstration that obligation loops could sustain engagement.
- 22. C Needing gifts from real Facebook friends to unlock crops and buildingsPlayers needed gifts from real Facebook friends to unlock crops and buildings, and those friends needed gifts back — progress was coupled to reciprocal participation.
- 23. B Reciprocal obligation loops can hold millions without the product being intrinsically rewardingMost players called FarmVille a chore, yet social-obligation loops and loss aversion over planted crops kept them returning — proving reciprocal mechanics can hold users without intrinsic fun.
- 24. A 2009FarmVille launched in 2009 and reached a peak of 74 million monthly active users, the prototype for social-obligation loops.
- 25. B Players returned mainly because the game was intrinsically funThe case study says most players described FarmVille as a chore; retention came from social obligation and loss aversion, not intrinsic fun. The other three statements are all supported.