Does Duolingo actually work? What independent research says

Somewhere around day 400 of a streak, a lot of people have the same uncomfortable thought: I have done this every single day for over a year, so why can I still not follow a conversation? This post is an attempt to answer that honestly, using the published research, including Duolingo’s own. The short answer is that the apps are optimised to be picked up daily, which is not the same thing as being optimised to teach, and that the device they live on works against you in ways the lesson content cannot fix.

Do the Duolingo effectiveness studies hold up?

The famous claim that 34 hours of Duolingo equals a university semester comes from a 2012 study commissioned by Duolingo itself . Read past the headline and it gets shaky. Of 156 volunteers, only 88 final scores were counted, so roughly four in ten dropped out. The gain figure was a mean pulled up by a few outliers; the median learner gained less than half as much. And the yardstick was a multiple-choice placement test. Nobody spoke, nobody wrote. Stephen Krashen picked the study apart in an independent critique two years later and concluded the university comparison was meaningless as measured.

The more recent studies are better run and still carry the same two asterisks. A 2021 paper found Duolingo completers matched fourth-semester university students on reading and listening, and a follow-up whitepaper extended the claim to five semesters . But four of five authors on the first, and six of seven on the second, were Duolingo employees. Both studies tested only people who had already finished the course, a small, unusually determined slice of users who were paid to be assessed. And neither tested speaking or writing at all.

Buried in that second report is the most interesting number in this whole literature: among the Spanish completers, time spent in the app correlated negatively with proficiency. More hours went with lower scores, not higher. Whatever was driving learning for the people the app worked for, it was not the minutes of tapping.

To be fair, the independent evidence is not damning, just modest. A university team that had nine learners study Turkish on Duolingo for a semester found real, measurable gains, alongside declining motivation and complaints that the exercises never connected to actual language use. A systematic review of a decade of Duolingo research found most studies focused on engagement with the app rather than on whether anyone learned the language. That sentence, more than any single result, describes the field.

Is the streak helping you learn?

None of this is a scandal. It is a business model. Duolingo has written openly about how the streak was engineered, drawing on habit research and loss aversion, the fear of losing something you have earned. The outcomes it reports for the streak are retention outcomes: users with a seven-day streak became far more likely to keep going, and design tweaks were measured in daily active users. Learning does not appear in that post because learning is not what the streak measures.

The psychology here is old and solid. A meta-analysis of 128 experiments found that expected rewards reliably undermine intrinsic motivation: once the points become the reason, the activity stops carrying its own. In the classic 1973 experiment, children rewarded for drawing later drew half as much for pleasure as children never rewarded. A semester-long classroom study of leaderboards and badges found students in the gamified section ended up less motivated and scored worse on the final exam. Gamification is genuinely good at making you show up. The evidence that it deepens learning is thin, and there is real evidence it corrodes the internal motivation that long-term language learning runs on.

Is learning on your phone part of the problem?

Suppose the lessons were perfect. They would still arrive on the most interruption-dense object you own. Laboratory work shows that merely receiving a notification, without even checking it, disrupts sustained attention about as much as actively using the phone. Attention research adds that each switch leaves a residue: part of your mind stays on the previous thing, degrading performance on the next. One well-known study found that the mere presence of a phone on the desk reduced available working memory, though later replications have been mixed, so hold that one loosely. The pattern across the literature is consistent: five minutes of study on a phone is rarely five minutes of study.

What works better than an app?

The research on reading is the mirror image of the research on apps: less marketing, better results. A month-long case study of extensive reading in French found measurable growth in spelling, meaning or grammar for 65 percent of the words tracked. Paul Nation’s corpus work estimates that about 25 novels’ worth of reading supplies enough repeat encounters to learn the 9,000 most frequent word families. A 2025 meta-analysis across dozens of studies found extensive reading improved comprehension, vocabulary, fluency, writing and even speaking . And the extensive reading tradition states its first principle in terms no app would ship: the reading should be easy, self-chosen and its own reward.

Even the one thing apps genuinely do well, spaced repetition, works best on a schedule no streak permits. The definitive study of spacing found optimal review gaps of days to weeks, stretching as knowledge consolidates. A streak compresses that into daily contact forever, which is cramming with extra steps.

If you keep the app

There is no need to be absolutist. An app is a fine on-ramp for script, sound and the first thousand words, and five minutes in a queue beats nothing. But the evidence suggests three rules. Treat the streak as a tool and let it die without guilt, because the moment protecting it becomes the goal, the reward has replaced the language. Turn the notifications off and do the practice in one sitting, because interrupted minutes barely count. And move your real study time to comprehensible input you actually enjoy, in quantity, mostly reading, because that is where the independent evidence has pointed for forty years.