The Duolingo Case Study: How to Launch AI Without Breaking Your Brand

A thought leadership piece for founders, operators, and anyone building at the frontier

In March 2023, Duolingo launched Duolingo Max — a premium subscription tier built around two AI-powered features. Explain My Answer lets learners ask why a specific response was correct or incorrect and receive a contextual explanation in natural language. Roleplay puts learners into simulated real-world conversations — ordering coffee in Paris, checking into a hotel in Tokyo — and provides feedback on how the exchange went.

The launch was, by most measurable accounts, successful. Subscription revenue grew 50% in 2024 to $607 million. Paid subscribers reached 10.3 million by Q1 2025, up 40% year-over-year. The Max features have since expanded to new markets and platforms, and Duolingo has continued building on the same AI foundation — adding Video Call with a live AI character and a suite of features that would have required a human tutor a decade ago.

Then, in April 2025, Duolingo’s CEO Luis von Ahn sent an all-company memo declaring the company “AI-first.” The memo was straightforward about what this meant operationally: teams would need to prove AI couldn’t do a job before requesting a headcount increase. Contractors doing AI-manageable work would be gradually replaced.

Within days, thousands of users unfollowed Duolingo on TikTok. The backlash was significant enough that von Ahn acknowledged it had dampened user growth, particularly among younger audiences in the US and Canada. The company stopped posting its characteristically edgy social content in an attempt to recover sentiment.

Same company. Same underlying technology. Same CEO. Two very different outcomes, eighteen months apart.

The gap between those two moments is one of the most instructive case studies in AI product strategy available right now — and the lesson it contains is not primarily about technology.

The starting point for understanding why Duolingo Max worked is understanding what von Ahn has said publicly about how Duolingo thinks about its product.

In an interview with Stanford Graduate School of Business, the Duolingo Handbook, and various public appearances, von Ahn has articulated a consistent philosophy: the product must be useful, intuitive, delightful, and polished — in that order. Usefulness comes first. The long-term health of the user relationship comes before short-term commercial pressure. As he has written: “If it helps in the short-term but hurts Duolingo in the long-term, it’s not right.”

The Max features were designed inside that philosophy. Explain My Answer and Roleplay arrived as enhancements to an existing learning habit, not as replacements for it. A user who got an answer wrong — something they had already experienced thousands of times in the app — now had the option to understand why, in natural language, at the moment it mattered. The AI feature inserted itself into an established moment rather than creating a new one.

This is a design insight that is easy to state and consistently ignored in practice. The AI products that achieve genuine adoption are almost always the ones that go to the user — arriving inside something they already do — rather than the ones that propose a new behaviour and ask users to adopt it. GitHub Copilot works inside the code editor. Grammarly works inside whatever text field the user is already typing in. Notion AI works inside the document. The pattern holds across very different product contexts: the most adopted AI features are the ones with the lowest behavioural friction, and the lowest behavioural friction comes from inserting capability into an existing routine.

Duolingo also had a significant structural advantage that most AI product teams do not have: fifteen years of accumulated user trust. When Max launched, Duolingo was not asking users to trust an AI system they had no prior relationship with. It was asking users who already had a Duolingo habit — who already trusted the product to help them learn — to try something new within that established relationship.

That is a fundamentally different ask from the one most AI products have to make. Trust in a new AI feature is largely borrowed from the trust in the relationship it arrives within. Teams building net-new AI products do not have that advantage, which means they have to build trust incrementally, through demonstrated reliability, before they can ask for the kind of reliance that high-value AI features require.

The April 2025 memo was not, in itself, an unreasonable communication. Von Ahn’s stated philosophy — “betting on mobile in 2013 made all the difference; we are making a similar call now” — reflects a genuine strategic view that has characterised Duolingo from the beginning. The company has consistently bet on emerging technology before it was perfect, trusting that capability would catch up.

What the memo got wrong was not the strategy. It was the framing.

Stephanie Liu, a senior analyst at Forrester who studied the backlash, identified the core failure precisely: “One mistake I think that a lot of companies make, not just Duolingo, is they fail to articulate how AI will benefit the customer or the end user. For Duolingo, it’s a very clear value proposition of ‘we’re going to save money by laying off contractors.’ But there was nothing in there that I saw about how it would improve the user experience or help you learn a new language. They basically cut the entire customer out of the messaging.”

That framing failure matters because Duolingo is not just an app. It has cultivated a genuine community — users who feel like stakeholders in the product, who identify with Duo the owl, who have made the streak a part of their daily life. Communicating AI strategy as an internal efficiency decision, without connecting it to user benefit, told that community something it did not want to hear: that the product they were attached to was now primarily a vehicle for cost reduction.

Von Ahn later acknowledged to The New York Times that the memo wasn’t controversial internally — which is itself instructive. What feels routine inside a technology company, where AI deployment is understood as a strategic necessity, can land very differently with a user base that projects its own anxieties about AI onto the brand they trust. Intent does not equal perception, and the gap between them is exactly where trust gets spent.

Read in isolation, the Max launch looks like a story about good product design. The 2025 memo looks like a story about bad communications. Together, they reveal something more fundamental: the relationship between an AI product and its users is not a product relationship. It is a trust relationship. And trust relationships have rules that are different from the ones that govern product decisions.

When Duolingo introduced AI features as enhancements to the learning experience — when the user’s first encounter with AI was “this helped me understand something I got wrong” — it was making deposits into a trust account that had been building for fifteen years. The user’s experience of the AI was positive, personal, and connected to something they cared about.

When Duolingo announced an AI-first strategy by leading with operational efficiency and contractor replacement — when the user’s encounter with AI strategy was “the company is using this to spend less money” — it was making a withdrawal from the same account. The user’s experience of the announcement was impersonal, organisationally focused, and disconnected from the learning experience they had come to rely on.

This is not a communications failure in the conventional sense. It is a failure to understand that the trust a product builds through years of positive user experience is not a fixed asset that survives any communication intact. It is a dynamic relationship that requires active maintenance — and that can be damaged, quickly, by a change in the register of how the company talks about itself.

The Duolingo case study is worth examining carefully precisely because it shows both the ceiling and the floor of what AI can do for a consumer product. At its best — in Max, in Explain My Answer, in the Video Call feature — AI enabled experiences that would have been impossible or prohibitively expensive without it, delivered in a way that users experienced as genuinely helpful. At its worst, the same underlying technology became the subject of a trust crisis when its deployment was communicated in terms that prioritised operational logic over user benefit.

Most of the AI product conversation focuses on the technical question: is the capability good enough? Duolingo’s experience suggests that is the wrong first question. The right first questions are:

Where does this feature arrive in the user’s existing experience, and does it enhance something they already value or ask them to do something new? The lower the behavioural friction, the higher the adoption. The higher the adoption, the more trust gets built, and the more trust gets built, the more ambitious the next feature can be.

What trust account is this feature drawing on, and how full is it? Teams with existing user relationships are drawing on accumulated trust. Teams with no existing relationship are starting from zero and need to build the account before they can make significant withdrawals.

When we talk publicly about our AI strategy, are we talking in terms of user benefit or operational logic? The framing that works internally — efficiency, cost, speed — is often the framing that damages trust externally. The framing that builds trust externally is the one that connects AI capability to the specific thing the user cares about.

None of these questions are about model performance. They are about the relationship between a product and its users — a relationship that AI features can strengthen or weaken, depending on how they are introduced, positioned, and communicated.

That relationship is the variable that most AI product teams are not managing. It is also, as Duolingo’s experience makes clear, the variable that determines whether impressive technology produces genuine adoption or becomes the subject of a backlash that its creators did not see coming.

Is your product roadmap ready for the ‘AI-first’ transition? Don’t let a communication gap stall your growth. [Book a 15-minute strategy audit] and ensure your AI features build loyalty, not backlash

[embed]AI Founder Narrative; Positioning Advisory | InSightEdge We help AI founders turn complex technical systems into high-impact narratives. Bridge the gap between engineering and…insightedge.technology

A note on this essay: the analysis above draws on publicly reported information, including Duolingo’s investor communications, Luis von Ahn’s published writing and interviews, and documented media coverage of the 2025 backlash. Where I draw conclusions about why specific decisions worked or did not work, these represent my analytical interpretation of the public record rather than confirmed internal strategy.

About the Author:

I have worked across global financial services, early-stage technology ventures, and international AI governance forums. I write on AI strategy, policy, and the communication gap between builders and institutions.

[embed]Beyond The Buzzword | LinkedIn Beyond the Buzzword is a thought leadership newsletter that cuts through AI, data, and analytics hype.www.linkedin.com

Originally published at https://www.linkedin.com.