Gina Gotthilf on Duolingo Growth, Experiments and Public Relations

Gina Gotthilf tells Harry Stebbings that her early career did not follow a neat plan. After studying philosophy, she struggled to find work in New York, encountered difficult jobs and visa constraints, and returned to Brazil intending to work on organic farms. She shares the less polished parts of the story alongside her later roles.

Before leaving New York, she met people she had previously worked with, including Mark Coatney, who had joined Tumblr. Months later Tumblr sought help in Brazil. She initially declined, accepted a one-month project, then became responsible for the market after the company struggled to find another candidate.

At Tumblr she learned that even impressive startups often have to solve basic problems as they go. She recalls borrowing money to pay contractors when transfers to Brazil were delayed and says she went unpaid for six months while living with her parents. This is her recollection of that period.

At Duolingo she first worked on expansion in Brazil and other markets, using press and partnerships. When the founders asked her to lead growth, she says she researched what the role meant and started with one backend engineer. The episode title’s user figures describe the period in which she led growth, not growth attributable to one person alone.

The small team tested notification wording even though the founders believed notifications had already been optimized. Gotthilf says the work was inexpensive and reached many users, so modest improvements could have substantial value. She prioritizes expected return against the time required, while noting that previous optimization does not rule out more learning.

A statistically meaningful negative experiment can also reveal a sensitive part of the user journey, she argues: the team can investigate alternative changes rather than simply abandoning the area. An inconclusive or negligible result may teach less about that specific lever. Her framing does not mean every negative result should be repeated indefinitely.

The team wanted to introduce game-like badges, inspired by studying popular games for retention ideas. Gotthilf initially resisted because a full implementation seemed expensive. They tried a minimal test: a celebratory badge for signing up, presented early in the user experience.

The test showed no useful effect, but a year later the team reconsidered whether the minimum version had captured the appeal of a badge. Signing up is not an achievement users necessarily value, and one badge does not create a collection. Stebbings emphasizes the missing connection between the action and a meaningful reward.

Gotthilf says the team began actually trying its own experiments instead of only reading the results. A more complete badge system later took substantially more work and, in her account, became one of the team’s strongest experiments across growth and other metrics. She provides no numerical effect size in the interview.

She explains that small samples can produce misleading results and says statistical significance helps evaluate whether an observed difference is distinguishable from chance. Smaller effects generally require more data than large ones. Early-stage startups may struggle to collect enough observations, and significance alone does not guarantee a result will repeat.

Gotthilf sees retention as central to product-market fit: acquiring users matters less if they do not return. At Duolingo, she says the north-star metric was daily active users, with supporting metrics to understand the drivers. She also judged PR by whether it reached people who became active users.

PR worked for Tumblr and early Duolingo, in her experience, because the underlying products already retained users. She suggests starting with a real story about the problem and the people it affects, then noticing which details prompt listeners to ask questions. Her claim about its contribution to growth is retrospective.

She recommends distilling the story into a concise initial message that gives a journalist a reason to care. Relevant background, an event or a credible partner can help establish context. She describes arranging talks for Duolingo’s founder and contacting journalists when he visited a market; her view is that a specific, relevant hook is more useful than a generic introduction.

Gotthilf recounts partnerships with governments to introduce Duolingo in schools. She says implementation and classroom usage were often weak, yet the partnerships and a government landing page helped attract press. The existence of a partnership should not be mistaken for evidence of widespread adoption or educational outcomes.

Rejection is common in media outreach, she says. She has contacted multiple journalists at one outlet but acknowledges this can backfire and violate PR convention; if one responds, she advises disclosing the other approaches. She recommends practicing on less-prioritized outlets first and refining the pitch based on responses.

Her own initial messages were short and designed to invite a reply; she would provide more detail once a journalist expressed interest. She acknowledges that withholding detail can irritate journalists. This is a tactic she says worked for her, not a guarantee that recipients will prefer it.

An exclusive can offer one publication a stronger story and potentially lead others to follow, but other outlets may no longer see it as news. A coordinated embargo can support broader simultaneous coverage if journalists agree to it. Gotthilf says the right choice depends on audience, relationships and the desired reach.

She suggests allowing journalists time to report, beginning relationship-building earlier and aiming to share substantive information roughly one to two weeks before a planned announcement. Timing varies with the story and the reporter. A sender cannot assume that an unsolicited embargo binds a recipient; Stebbings later notes he receives such messages without agreeing.

For a broad consumer product, a paywall can limit reach; for a specialist business, the credibility and audience of a particular publication may matter more than raw breadth. Stebbings says founders should define the objective and audience first: fundraising coverage is not necessarily the best way to reach customers.

Gotthilf stresses that independent journalists control their own reporting. A clear release with links and adequate context can make facts easier to check, but it cannot guarantee a preferred article. She discusses how relationships and power can affect requests for changes, while acknowledging journalists may fairly disagree.

She recalls managing global Duolingo launches across many countries and encountering broken embargoes. Her response was to contact the publisher and try to resolve the situation, but some outcomes could not be controlled. The episode illustrates the limits of a coordinated launch, not a promise that every embargo will hold.

Measuring PR is difficult and should follow the objective. For Duolingo she looked for new daily active users; for fundraising, a founder might track investor references or inquiries. Some brand effects are less tangible, she says, but early companies still need to relate effort to business goals.

After publication, Gotthilf recommends using social channels and community distribution. Give people a reason to care, connect the story to a mission and make sharing easy with a link and suitable copy. Stebbings agrees that reducing friction helps, while the speaker’s examples reflect her own campaigns.

At Latitud, she says the team made personalized graphics and suggested copy for founders admitted to its community. Founders sharing their selection made their own achievement visible and introduced Latitud to new audiences. She openly describes the self-presentational incentive behind that tactic; the growth effect is her account.

For outreach at scale, she would personalize the beginning and end of a message while retaining a prepared core and links. Stebbings dislikes receiving cold press releases labeled with an embargo he never accepted. Their point is to respect the recipient’s context instead of assuming a mass message creates an agreement.

Both are critical of many PR firms, though Gotthilf recognizes that agencies may provide relationships and expertise founders lack. She favors doing the work internally when possible; when outside help is necessary, she suggests having someone learn from the agency. That is her preference, not an assessment of every firm.

The most common error, in her view, is assuming what matters to the founder will automatically matter to a reporter. A pitch should account for the publication, journalist and audience. She favors a steady cadence of relevant stories, rather than relying only on occasional funding announcements.

She defines PR broadly as reaching people with audiences, including creators, micro-influencers or a teacher with a relevant community. The scale and usefulness of a channel depend on whom a startup needs to reach. The discussion broadens from traditional news outlets to earned distribution more generally.

In quick questions, Gotthilf says durable growth thinking starts with understanding people and identifying high-impact levers. She is skeptical of relying on paid ads indefinitely because costs and competition can rise and the company becomes dependent on a channel. She also notes that paid acquisition can help early testing and sample collection.

A tactic can become less distinctive as competitors adopt it, she says, citing subway advertisements and notifications. The underlying work is to keep experimenting rather than assume a specific channel is permanently special. Her prediction that AI will change these tools is speculative.

She warns founders against hiring a growth team before product-market fit or treating an underpowered test as “directionally statistically significant.” She also worries that hiring an overly senior leader too early can leave nobody willing to run hands-on experiments. She favors a balance of experience and willingness to try things.

For postmortems, she suggests meeting while memories are fresh, documenting what worked and did not, and surfacing views without blame. Anonymous collaborative input can make discussion easier. The purpose is to understand causes and make concrete improvements rather than hide mistakes.

Gotthilf makes a 2023 case for Latin American technology investment: large unmet needs, founders with experience at scaled companies, greater connectivity and increasing capital. She is co-founder of Latitud, which supports and invests in the region, so she has a direct interest in that outlook. Her broad generalizations about regional conditions are opinions rather than uniform facts.

She points to services for small businesses, banking, health and education as potential opportunities and hopes local founders can solve problems they understand firsthand. Her account of available talent, money and connectivity is an optimistic market assessment, not a guarantee of returns or universal access.

For a company that interests her, she mentions Artifact, the news app from Instagram co-founders Mike Krieger and Kevin Systrom, and discloses her friendship with Krieger. She hopes better news experiences can help people encounter different viewpoints. This was her 2023 view of the product; the conversation ends with thanks.