Duolingo

Most education products are judged on outcomes: grades, certificates, fluency. Duolingo is judged by its investors, and by itself, on daily active users. That one choice of north-star metric explains more of the interface than any style guide could.

The growth has been steep and steady. Each first quarter shows a New Year’s-resolution bump, and the line never gives it back.

In 2023 Duolingo’s former CPO published the company’s growth model.2 Every user sits in one state on any given day, and the business is the sum of the rates at which people move between them. When the team modelled which rate mattered most, one stood out: Current User Retention Rate (CURR), the rate at which already-engaged users stay engaged.

This is the key to the whole product. A feature here is not judged on whether it teaches better. It is judged on whether it moves someone from “at risk” back to “current”.

Most subscription products open with a form. Duolingo’s first ten screens contain no text fields at all. Every input is a single tap, and every question gives back something: a chosen goal, a personalised promise, a lesson.

The order is deliberate. The questions before the lesson are less about personalisation than about commitment. Picking a reason (“for travel”) and a daily goal (“10 min/day”) are small public promises, and people work hard to stay consistent with promises they have made.

The lesson itself is excellent interaction design: one decision per screen, instant feedback, a forgiving input model and a progress bar that never stalls. It is a good example of keeping cognitive load low.

What follows the lesson is a different system. Completing a lesson triggers a run of full-screen confirmations: XP, accuracy, streak, quest progress, league position and, on the free tier, an advert or an upsell. Each one is short. Together they are where the habit gets built.

On paper, 18% overhead is modest. But it is spent at the moment that matters most for a habit: straight after the effort, while the learner’s attention is open. The ceremony doesn’t teach anything. Its job is to make sure there is a next lesson.

Every retention mechanic in the app was mapped against the behavioural principle it uses, and each pairing was rated for strength. The pattern is clear: loss aversion is the strongest column, and the streak and its related features dominate the top of the chart.

Variable reward, the slot-machine mechanic usually blamed for compulsive products, is used sparingly here. Duolingo is not built mainly on the hope of winning. It is built on the fear of losing.

The streak is a product inside the product

The streak has its own currency (freezes), its own insurance (repair), its own social layer (friend streaks), its own notification schedule and its own celebration screens. It is the most heavily designed object in the app, more than any lesson type.

This works because the streak turns a long, vague goal (“become conversational”) into a short, concrete one (“don’t break the chain tonight”). The risk is that, after a few weeks, the streak becomes the goal and the language becomes the means.

Duolingo’s own filings show how far the streak reaches. At the end of 2025, 43 million daily users had a streak of a week or more, and 15 million had kept one going for over a year.1

Scored against Nielsen’s ten usability heuristics,3 the app splits clearly. It scores at the top for system status and consistency, which is the visible polish everyone praises. It scores at the bottom for user control and flexibility.

Both results come from the same decision. The 2022 move from a branching skill tree to a single linear path made progress easier to read and choices harder to make.

Severity is rated on learning harm, not on lost revenue. Several of these are strong business decisions. They are listed because a learner pays for them.

Each mechanic is plotted on two axes: how much it brings people back, and how much it helps them learn. The top-right quadrant is where a learning product should want to be. The highlighted bottom-right quadrant is where Duolingo puts most of its effort.

The top-left is just as telling. The features that teach the most, such as targeted review, explanations and stories, get the least retention support. Some of them, like answer explanations, are behind the paywall.

Removing streaks or leagues would be naive. They work, and a learner who doesn’t come back learns nothing. The goal is to keep the habit engine and connect it to learning, so that the easiest way to protect your streak is to actually learn something.

Most product teams spend their best design effort on the first session and almost none on the hundredth. Duolingo does the opposite. Its onboarding is good, but its real craft is in the moment someone decides whether to open the app tonight.

Any product that needs repeat use, such as fitness, finance or health, can borrow the structure: give value before asking for anything, turn long goals into short daily ones, make progress visible and give people something to protect.

The lesson to avoid copying is just as important. A metric left unchecked will shape the product around itself. Pick two metrics that keep each other honest.