CHI Company Investigation · Technology
Duolingo
Hearts punished mistakes. Energy limits use.
Language-learning app · Free courses · Super Duolingo · Duolingo Max
Standard company investigation · Concerning · Research current through 9 October 2026
Duolingo still gives away substantial value, but it has also learned how to monetize the friction between a learner and the next lesson. Some of what the subscription sells is relief from friction the company itself controls.
Under Hearts, a free learner lost capacity only by answering incorrectly, so an accurate learner could keep going. Under Energy, introduced in 2025, every exercise uses capacity whether the answer is right or wrong. Relief comes from waiting, a rewarded ad, gems or a subscription. Duolingo told investors Energy was designed to boost engagement and conversion
, and separately wrote that adding friction had been its main way of raising bookings per user. The point is not that Duolingo charges money. In 2026 it began to pull back. Energy remains.
CFS = CVI − CHI (79 − 50 = +29). Duolingo delivers considerably more value than hostility. The hostility that remains sits in one place: free daily practice became a metered quantity, and relief from the meter is sold. Methodology →
What Duolingo gives The free curriculum
Every course is free, and about nine in ten monthly users never pay.
Duolingo’s 2025 annual report says anyone can complete as many of our courses as they choose, all without paying anything
, and that about 9% of monthly users were paid subscribers.S6 The catalogue more than doubled in 2025, and in 2026 several paid features moved to the free tier.S13S10S11
What changed The limit on free use
The limit stopped being a consequence of error and became a consequence of use.
Hearts charged a free learner for each wrong answer. Energy charges for each exercise, right or wrong, and the company describes it as a feature designed to boost engagement and conversion
.S1S4 Waiting, ads, gems or a subscription restore it.S9
The subscription is not the problem. The meter is. CHI separates paying for something new from paying to remove a limit the company sets. Duolingo does both.
01What changed
Hearts → Energy: the constraint moved from mistakes to usage.
Both systems limit how much a free learner can do on the mobile app before something has to give. The difference is what triggers the limit. The strips below apply each rule to the same ten exercises, two of them answered incorrectly.
Hearts
Mistake-based constraint- ✓
- ✓
- ✕
- ✓
- ✓
- ✓
- ✓
- ✕
- ✓
- ✓
Exercises that used capacity2 of 10
- Learners start each day with five Hearts
- Incorrect answers consume Hearts; correct answers do not
- An accurate learner could keep going without a volume limit
- Practice lessons could earn Hearts back
- The paid plan offered unlimited Hearts
Duolingo S-1, June 2021: learners lose a heart for each exercise completed incorrectly
and can earn them back by completing practice Lessons
.S1
Energy
Usage-based constraint- ✓
- ✓
- ✕
- ✓
- ✓
- ✓
- ✓
- ✕
- ✓
- ✓
Exercises that used capacity, before any bonus10 of 10
- Every exercise uses a unit of Energy
- Correct answers still consume Energy; runs of correct answers earn some back
- Daily free use becomes finite; Energy recharges over about a day
- Waiting, a rewarded ad or gems restore capacity
- The paid plans remove the limit: Energy is a system
for free users
Q2 2025 shareholder letter: Each exercise uses a unit of energy, regardless of whether the answer is right or wrong, but learners get back energy when they get multiple answers correct in a row.
S4S9
The limiting condition moved from failure to use.
The strips illustrate the published rules, not measured data. Duolingo has not published how large the Energy allowance is, how much a run of correct answers returns, or how many lessons a full bar covers, so this page does not state those figures. How often a lesson is cut short is also unestablished, and depends on how much a learner practises.Who this affects most. The change is redistributive. Duolingo says beginners were 2X more likely to run out of hearts mid-lesson
, so learners who make many mistakes can be better off under Energy.S9 The limit now binds hardest on accurate learners who practise a lot without paying, who under Hearts were not volume-limited at all. A learner doing a short daily lesson may rarely meet it. Not all users are affected equally.
02The company’s own words
Duolingo described the mechanism itself, in filed documents.
CHI normally has to infer purpose from conduct. Here it does not. The quotations below are from shareholder letters filed with the SEC and from the 2021 prospectus. Each was checked against the filed text. No earnings-call transcript is quoted on this page.
“Up until now, the main way we’ve increased bookings per user is by adding small amounts of friction to encourage more people to subscribe, for example by increasing ad load or subscription upsells.”
The same letter says this friction is part of the reason our DAU growth has slowed
, and that some of the slowdown is a function of our increased focus on monetization in recent years
. It does not name Energy.
“…a new core feature designed to boost engagement and conversion: Energy.”
The letter adds that on iOS Energy increased DAUs, median time spent learning well, and subscriber conversion
.
“Our freemium business model allows free learners on Duolingo to develop a regular habit of using our product, ultimately leading to high quality paid conversion.”
The same filing says Duolingo runs thousands of A/B tests to optimize each feature for maximum engagement
.
One feature, two descriptions
Duolingo explained Energy to learners and to investors in the same summer. Both accounts are on the record and they are not contradictory. They emphasise different things.
Hearts were not the most effective way to support learning; beginners were 2X more likely to run out of hearts mid-lesson
.S9
Unlike Hearts, which penalize mistakes, Energy is usage-based and rewards success.
S4
Why the admission matters. CHI does not infer intent from the fact that a company benefits. Here Duolingo’s own written account links friction to subscription, and links Energy to conversion. That turns a plausible inference into stated purpose, and it is the main reason confidence in the hostility score is medium rather than low.
03How the model works
Free learning, a trained habit, a limit, and a paid way past it.
Each step is described in Duolingo’s own filings. The sequence is the economic structure CHI assessed. It is not a claim that every step was designed with the last one in mind.
Habit and engagement
Streaks, leagues and gamification built to bring learners back daily. Over 10 million users held streaks of a year or more by early 2025.S1S3
The relief ladder: four ways past the Energy limit, paid in different currencies
These are not equivalent. Waiting costs nothing but time. An ad costs attention. Gems can be earned in the app or bought. A subscription costs money and removes the limit entirely.
Wait
Energy fully recharges in about a day—just like Hearts.
Free, but practice stops until it does.S9
Rewarded ad
You can always watch a rewarded ad to earn back some Energy.
The scarce resource creates ad inventory. The amount restored is not published.S9
04Selling value vs selling relief
Paying for something new is not the same as paying to remove a limit.
This distinction is central to how CHI scores Duolingo. CHI does not penalize a feature because it is paid. The concern is the second column, where the thing being removed is a constraint the company sets and can change.
Max and the AI features
Capabilities that did not exist before, with real costs to serve.
- Roleplay and Explain My AnswerLaunched with Duolingo Max in March 2023 as new AI features.S14
- Video CallAn AI conversation partner that Duolingo calls Max’s
core feature
.S5 - A real serving costDuolingo reported lower margins from
increased generative AI costs due to the increased adoption of Duolingo Max
.S3 - Moved down as costs fellExplain My Answer became free in 2026; Video Call is moving into Super.S10S8
Super’s headline benefits
Removal of things the free tier contains by design.
Ads and a usage limit are an ordinary freemium bargain. What CHI scores is the change: the limit was extended from mistakes to successful use, and the company says friction of this kind was how it raised bookings per user.
05Why it scores 50, not higher
The case against a higher CHI score.
This evidence is the reason Duolingo sits at 50, mid-range, and not in the 60s or 70s with the most hostile companies CHI has assessed. It is given full weight.
All courses remain free
The annual report says anyone can complete any course without paying. CHI found no course paywall.S6S12
About 91% of monthly users pay nothing
About 9% of monthly active users were paid subscribers at the end of 2025.S6
High customer value
A complete curriculum at no charge, used daily by a very large audience. The catalogue more than doubled with 148 new courses in April 2025.S13
No benefit stripping was established
No Super benefit was found removed and resold in Max. Max sold new AI features, and some of them later moved down a tier.
Streak protection did not get more expensive
No evidence was found that streak protection was increasingly monetized. In June 2026 a free event let 15.4 million learners revive lost streaks.S8
Friction is being reduced
Paid features moved to free, Video Call is moving into Super, and Duolingo is accepting more than $50M of foregone bookings to do it.S5
AI did not demonstrably degrade courses
CHI tested whether AI-generated content reduced course quality. No systematic evidence was found, so nothing is scored for it.
No cancellation obstruction was found
The research found no established evidence that cancelling a subscription is made deliberately difficult.
062026 reversal
Duolingo pulled back.
In February 2026 Duolingo said it would prioritize user growth over monetization
and estimated the cost at more than $50M of foregone bookings, about five points of growth. It said it had repurposed part of the monetization team
to work on user growth instead.S5 The strongest extractive conduct belongs to the period that letter calls up until now
, and most clearly to 2025.
Friction optimization
- Friction added to encourage subscription,
for example by increasing ad load or subscription upsells
, by Duolingo’s own account.S5 - Energy announced as the replacement for Hearts on the free mobile app, rolled out from iOS and
designed to boost engagement and conversion
.S9S4 - Duolingo later named the initial rollout of Energy, a price increase and advertising as reasons its mid-2025 bookings were hard to beat.S8
Partial retreat
- Explain My Answer becomes free, formerly a Max feature.S10
- Practice sessions free
for the first time
to all learners.S11 - Strategy shift: user growth over monetization; Video Call to move into Super; free AI speaking practice.S5
- Video Call available to most new Super subscribers; longer free trials; work focused on revenue
without adding friction to free users
.S8
Reduced friction
- Explain My Answer → free
- Practice sessions → free
- Video Call → moving from Max into Super
- Free trials → longer
- Ad load and upsells → named as the friction behind slower growth; user growth now put ahead of monetization
Still present
- Energy remains on the free mobile app
- Free daily use is still finite; no source reviewed shows the allowance loosened
- Paid relief: ads, gems and subscription still restore it
- The February 2026 letter does not mention Energy among its changes
The retreat changed the score. It did not erase the mechanism.
Under CHI’s methodology a correction is credited in full for today’s customer, and the conduct before it stays on the record. That is why both the 2025 friction and the 2026 retreat appear on this page.The 56 is not a published historical score. It is the scoring analysis’s estimate of what the same evidence would have produced before the 2026 changes, shown only to make the effect of the retreat visible.
07Score architecture
Why the scores sit where they do.
Duolingo delivers substantial value and contains a deliberate, documented friction mechanism at the same time. CHI measures the friction, CVI measures the value, and CFS reconciles them. Under CHI’s anti-double-counting rule, the Energy change is scored once for its monetization (Revenue Extraction) and once only for the separate loss of free capacity (Customer Restriction).
The gap between value delivered and hostility measured is +29: value currently exceeds hostility, with a CHI that sits in the Concerning range.
CHI breakdown
50/100By Duolingo’s own account, friction (ad load, upsells) was the main lever on bookings per user. Energy turned previously unlimited accurate practice into a metered allowance, with paid restoration through ads, gems and subscription.
All courses are free, about nine in ten users pay nothing, Max sells new capability, and the 2026 retreat is credited.
A habit system tested for maximum engagement
that the company links to paid conversion, with a scarcity mechanic placed on the habit it trains.
No addiction or switching-cost finding is made, and streak protection did not become more expensive.
Accurate, high-volume free learners lost daily practice capacity they previously had. The 2022 move to a single fixed learning path also reduced learner control over sequencing.
No content paywall; practice sessions and Explain My Answer returned to the free tier in 2026; no cancellation obstruction was established.
Ordinary data practices for an ad-supported app. This dimension was not a focus of the research, is scored conservatively, and carries no headline finding.
The learner-facing account of Energy does not mention conversion or a cap; the investor account does. Always free
sits beside a usage limit. Energy appears in shareholder letters but not in the annual or quarterly reports, and the 2025 annual report qualified one statement from “none” of the content behind a paywall to “the majority” not behind one.
The written admission itself and the 2026 corrective action count as mitigation, as the methodology requires.
CVI breakdown
79/100A complete language curriculum at no charge, with no course paywall.
On the free mobile app it now arrives in a capped daily quantity.
The catalogue more than doubled in 2025; new AI features; paid features moved to free and to Super in 2026; more free speaking practice.
Learner control over the path was reduced in 2022.
A mature cross-platform product delivered free at very large scale.
Dependable delivery of the core product. Held back by forced redesigns and by learning-outcome evidence that is mostly produced or funded by Duolingo itself.
Max, Video Call and AI-assisted course production.
Pushed up by
- Written admission that friction drove bookings per user
- Free limit extended from mistakes to successful use
- Energy described to investors as designed for conversion
- Monetized recovery: ads, gems, subscription
- Learner-facing account silent on conversion
Held down by
- All courses free; no paywall
- About 91% of users pay nothing
- Max sold new value
- 2026 retreat, costed by the company
- Candid written disclosure
Driven by
- Free, complete curriculum
- Catalogue more than doubled
- New AI capabilities
- Paid features moved to free in 2026
- Free speaking practice
Held back by
- Capped free daily volume
- Reduced learner control
- Mostly company-produced efficacy evidence
Does Duolingo deliver enough value to offset the extraction observed? Yes, at present. The value reaches everyone. The extraction is concentrated on free learners who practise more than the daily allowance permits.
A light free user and a subscriber both experience a relationship better than +29. An accurate, high-volume free learner, the person the habit system worked best on, experiences a worse one. The score would fall quickly if the allowance were tightened.
08Pattern classification
One mechanism, seen two ways.
Both CHI signal patterns apply, and they describe the same conduct. They are not two independent offences and are not scored twice.
Duolingo raised its yield from an engaged, habit-formed free base by adding friction, most clearly by extending the free usage limit from mistakes to successful use, and selling relief from it.
Existing-Customer Yield Optimization
Primary pattern · the strategyDuolingo’s own account says added friction, ad load and upsells increased bookings per user. The pressure fell on users it had already acquired, rather than on new paid functionality. Max is excluded: it sold new features.
Unmonetized Access / Entitlement Monetization
Secondary pattern · its main instrumentEnergy extended the usage constraint from mistakes to successful use, so accurate practice that was unlimited became a constrained allowance whose removal can be bought. It applies to Energy on the free mobile app, not to course content.
Temporal state: implemented, partially retreating. The friction strategy and Energy are both in place after implementation. The 2026 changes withdrew part of the strategy and left Energy as it was.
09Final verdict
Duolingo
Concerning
Duolingo still provides substantial free educational value. Every course is free, the catalogue has more than doubled, and roughly nine in ten of its users never pay. That is why its CVI remains high.
But the economic relationship changed when successful free use itself became a metered resource. Under Hearts, a learner who made few mistakes could practise for as long as they liked. Under Energy, each exercise uses capacity, right or wrong, and capacity is restored by waiting, watching an ad, spending gems or subscribing. Duolingo described Energy to investors as designed to boost conversion, and wrote separately that adding friction had been its main way of raising bookings per user.
In 2026 the company began removing some of that friction, at a stated cost of more than $50 million, and that materially improves the current score. Energy remains.
10Limits of this assessment
What this page does not claim, and what remains open.
Not findings
- That Duolingo is addictive. Gamification is documented; addiction is not claimed.
- That streaks are a proven switching cost. Plausible from company data; no independent study measures it.
- That Energy exists solely for monetization. Duolingo reports engagement and beginner benefits too.
- That Energy routinely cuts lessons short. Whether and how often that happens is not established.
- That AI degraded course quality. Tested; not established.
- That Super benefits were stripped and resold in Max. Not found.
- That courses are paywalled, or that users are trapped in subscriptions. Neither is supported.
Still open
- The size of Energy’s effect. Duolingo has disclosed direction, not magnitude, for conversion, daily users and time spent.
- Energy’s parameters. Allowance size, gem price, ad reward and recharge rate are not published by Duolingo and are not stated here.
- The dates of the friction period. Duolingo says only
up until now
andrecent years
. - Ad load over time. No source gives ad frequency by year.
- Energy outside the mobile app, for school accounts and for family-plan members.
- Future pricing of Energy. Duolingo has discussed plan experiments on earnings calls; nothing in its filed documents describes a plan that prices Energy quantity, so none is reported here.
- Q3 2026 results, due in early November, may change the picture.
11Methodology and sources
How this page was built, and how far it can be relied on.
Evidence base
One research dossier compiled on 9 October 2026 from eight evidence streams, followed by an independent verification pass and a separate scoring analysis under CHI Methodology v2.0. SEC filings and shareholder letters were preferred throughout.
Quotation standard
Every quotation on this page was checked against the filed SEC text or the live Duolingo page. Earnings-call statements were excluded because only third-party transcripts were available. User reports were used only to locate mechanisms, never as evidence.
Limits
Duolingo’s help centre, web archives and community forums could not be read, so Energy’s exact parameters are not established. No first-hand product testing was done. Duolingo did not participate in or review this research.
How confidence is expressed. Facts from Duolingo’s own documents are stated plainly. Company-reported results are labelled as such. Figures marked CHI arithmetic are calculations on reported numbers. Score confidence is medium for CHI, CVI and CFS.
Duolingo, Inc., Form S-1 registration statement, 28 June 2021: Hearts mechanics, unlimited Hearts as a subscription benefit, the freemium habit-to-conversion model, A/B testing for engagement. sec.govFirst-party · SEC filing
Q3 2022 shareholder letter, 10 November 2022: gems as a virtual currency earned in lessons or bought; gems as the main source of in-app purchases. sec.govFirst-party · SEC filing
Q4 2024 shareholder letter, 27 February 2025: over 10 million users with streaks of a year or longer; gross margin reduced by generative AI costs from Max adoption. sec.govFirst-party · SEC filing
Q2 2025 shareholder letter, 6 August 2025: Energy introduced as a feature designed to boost engagement and conversion
; the per-exercise rule; reported effects on DAUs, time spent and conversion. sec.govFirst-party · SEC filing
Q4 2025 shareholder letter, 26 February 2026: the friction admission; more than $50M of foregone bookings; user growth prioritised over monetization; Video Call to move into Super; free AI speaking practice. sec.govFirst-party · SEC filing
Annual report on Form 10-K for fiscal 2025, 27 February 2026: courses free to complete; an ad at the end of each free lesson; about 9% of MAUs paid at year-end. sec.govFirst-party · SEC filing
Q1 2026 shareholder letter, 4 May 2026: Video Call access expanding to Super; new speaking practice for free and paid users. sec.govFirst-party · SEC filing
Q2 2026 shareholder letter, 5 August 2026: Video Call for most new Super subscribers; longer free trials; revenue work without adding friction to free users
; Streak Revival; the Energy rollout cited in the bookings comparison. sec.govFirst-party · SEC filing
Duolingo Blog, “Why Duolingo switched to Energy”, published 3 July 2025, last modified 3 November 2025; checked against the live page on 9 October 2026. blog.duolingo.comFirst-party
Duolingo Blog, Explain My Answer is now free, 1 January 2026: free for learners in courses teaching seven languages. blog.duolingo.comFirst-party
Duolingo Blog, guide to the Practice Hub, 18 February 2026: practice sessions available to all learners for free. blog.duolingo.comFirst-party
Duolingo Blog, “Is Duolingo free?”, modified 23 March 2026: all courses free. blog.duolingo.comFirst-party
Duolingo investor relations release, 148 new language courses, 30 April 2025. investors.duolingo.comFirst-party
TechCrunch, Duolingo Max launch, 14 March 2023: the new tier with Explain My Answer and Roleplay. techcrunch.comJournalism
Reading this beside the rest of the index
Every company here is assessed against the same question. On Duolingo, the answer depends on how much you practise.
Read this page alongside the methodology, the other technology companies CHI has assessed, and the CHI Calculator.