Duolingo removed its community and lost the culture with it. The Light Bulb puts both back and makes them profitable.
ROLE
UX/UI & Brand Designer
Problem
Duolingo now explains wrong answers, but the explanations are AI-generated, unreviewed by experts, and stripped of cultural context. Knowing you're wrong isn't enough. You need to know what it means in real life.
APPROACH
5 user interviews, Reddit community research, competitive analysis across Babbel, Busuu and Memrise, 4 personas, RICE scoring across 15 features, 5-user wireframe tests.
SOLUTION
A post-answer explanation feature with teacher-sourced cultural context, tiered access, and an ad model that funds human expert review.
Tools
Figma, FigJam, Otter.ai, Slides, Reddit
STATUS
Concept · Design Lab (Bootcamp)
The Problem
I started by interviewing 5 people who use language learning apps and reading Reddit threads to understand real frustrations. AI replacing human interaction, rising costs, removed features. All came up. But one thing kept appearing above the rest.
Duolingo added AI explanations for wrong answers. But those explanations have no cultural context and no expert review. You learn that your sentence was grammatically wrong, not whether it would confuse a native speaker, sound rude, or only work in one region. And the forums where people used to figure that out together? Gone since 2023.
2021-2023
PROFIT FIRST
Only AI
Forums deleted
No community
Duolingo added AI explanations for wrong answers. But those explanations have no cultural context and no expert review. You learn that your sentence was grammatically wrong, not whether it would confuse a native speaker, sound rude, or only work in one region. And the forums where people used to figure that out together? Gone since 2023.
"Language is the road map of a culture."
Rita Mae Brown
users with no explanation when they get something wrong
community features left after the 2023 forum removal
personas named the context gap as their top frustration
Research
Interviews and Reddit research produced consistent themes. I built 4 personas to represent the range of users, then ran empathy mapping to understand where frustration actually lived.
No explanation for wrong answers * Forums are gone AI feels hollow Context missing Gamification fatigue 5-10 min sessions
"I need to know if what I'm saying would sound weird to an actual Spanish speaker. Not just if the grammar is right."
Process
Setback 01
Designing something that already existed
I got excited and started designing an "explain my answer" feature. Then I tested Duolingo Max, their $30/month tier. They already had it. I'd spent weeks on something that existed behind a paywall. This was my first real mistake.
Pivot
From feature to fairness
I stopped trying to invent and started asking a different question: should explanations be free? I interviewed more people. The answer was clear. That led to the Light Bulb, 5 free explanations that refill over time, earnable through the app, with tiers for paid users.
Setback 02
Duolingo made it free for everyone
A month into this, Duolingo announced AI explanations would be free for all users. I nearly shelved the project. Then I realised the problem wasn't solved, it was just moved. AI explanations with no human review aren't sustainable for quality. That became the new angle.
Pivot
From feature to fairness
Duolingo removed the forums because moderation was too expensive. Now they're giving free AI explanations with no clear funding model. My ad system solves that. Ads fund human experts who review and improve the AI, keeping quality high without another removal in two years.
The Build
The Lightbulb appears after you answer a question, right or wrong. It's collapsed by default. Tap to expand. You get 2-3 sentences of cultural context or usage explanation, written by language educators and reviewed by humans. The lesson never stops for it.
Placement
Post-answer, before Continue. Pre-lesson tips were skipped by 3 of 4 personas, the right moment is right after the question lands.
Visual states
Active: pink icon, shows count. Empty: gray icon, countdown timer showing when refill happens. No confusion about what state you're in.
Access tiers
Free: 5 bulbs, refill over time or earn with Gems. Plus: 10 bulbs. Max: unlimited. Always a way in, including one free ad-unlock.
The model
Optional ads fund human expert review of AI content. Free users get quality explanations. Experts get paid. AI gets better. Sustainable.
The economics behind it: Duolingo has 120 million free users. If just 20% watch one ad per day, that's $8-12M per year. Human expert review costs roughly $750K annually. 90% profit margin, and the AI gets better over time, dropping that cost further.
Free users (est.)
Ad revenue at 20% participation
Human expert review cost
Projected margin
Testing
I tested mid-fidelity wireframes with 5 users via remote moderated sessions recorded with Otter.ai. Tasks covered: finding and using the lightbulb, running out and navigating the empty state, and understanding the tier system.
FIX IMMEDIATELY
Icon not recognisable. 2 of 5 users didn't know what the lightbulb meant without a label. Added a text label and a first-time tooltip.
Empty state felt like a wall. When bulbs ran out, users felt trapped. Redesigned to show all 4 options clearly: wait, spend Gems, upgrade, or watch an ad.
HIGH PRIORITY
Content too long. Explanations over 3 sentences felt like homework. Hard cap at 3.
Tier confusion. Plus vs Max wasn't clear. Added a comparison on the empty state popup.
Result
The Lightbulb addresses three things at once: the cultural context gap Duolingo's AI explanations still leave open, the trust problem of unreviewed AI content, and the sustainability problem of giving things away with no model to maintain quality.
It's different from what Duolingo has now. Yes, they added AI explanations, but grammar corrections without cultural depth, and without expert review, isn't the same as understanding a language. The Lightbulb adds that layer: human-reviewed context funded by a model that keeps it sustainable long-term.
Learnings
Research the existing product before designing
I designed something Duolingo Max already had. Two weeks of work, gone. Check what exists first, always.
A limit without an exit is just a wall
The original Light Bulb system frustrated users more than no feature at all. Constraints need escape routes.
A setback can reframe the whole problem
When Duolingo made AI explanations free, I thought the project was over. Sitting with it revealed a better angle: sustainability, not just access.
"Cultural" was the wrong word
Users didn't respond to "cultural context" as a label. "Usage tips" and "context" landed every time. Words in design are design decisions.
The math has to work
Duolingo removed the forums because the economics didn't hold. Any feature that asks for human effort needs a funding model, or it disappears.
Every pivot in this project came from something breaking. The feature that exists now is better than the one I set out to build, because I had to throw away the first two.
Get in touch
Have a project in mind? I’d love to hear about it. Fill in the form and I’ll get back to you.