The only hard proof that social media worked came from the month they stopped

You have read about this account before. Every agency blog has a version, and most of them say the same four things: they were funny, they were fast, they let the intern post, and you should be brave like them.

Almost none of them mention the part that matters. In November 2025, Duolingo told its shareholders, in a document filed with the American securities regulator, that its user growth had slowed partly because it had stopped posting unhinged content. The share price fell roughly a quarter in a single session.

That sentence is the most valuable thing in this entire field. Nobody in social media marketing can prove their work causes growth: there is no holdout group, no control market, no clean experiment. Duolingo accidentally ran the experiment in reverse: they turned it off, growth dropped, and they said so under legal obligation to be accurate.

So this is not a story about being funny. It is a story about what an account is worth, what it costs the people who make it, and which parts of the method survive contact with a business that is not Duolingo.

The numbers, and where they come from

Most published follower figures for this account are copied between blogs without a date. The series below was reconstructed from archived captures of the public TikTok profile, so every line has a day attached to it.

• December 2021: 2.0M followers, 90 videos online
• January 2024: 8.5M
• October 2024: 13.3M
• February 2025: 15.1M
• March 2025: 16.7M
• 25 April 2025: 17.1M, the peak
• 17 May 2025: 16.7M, and zero videos online
• January 2026: 17.0M

Two things in that series are almost never reported. The account has been flat since April 2025: 17.1 million then, 17.0 million nine months later. And its rate of accumulating likes fell from around 51 million in the month of February 2025 to roughly 3 million a month by the end of that year. The growth story has an end date.

The team was always small

One person in September 2021, plus a manager at half time. Four by mid-2023. Six by November 2023. In April 2025, at seventeen million followers, the social team was four people inside a marketing department of fifty-one.

This is the number worth carrying away, because it is the one that transfers. Three to six people ran the most visible brand account in the world for a product with more than fifty million daily users. Whatever the advantage was, it was not headcount.

There was also no agency. Their social lead was blunt about why, and the reason is a number rather than a philosophy: “Turnaround time is a week. I’m not waiting two months.”

The three mechanisms that actually did the work

One week, end to end

Their standard time from idea to published was a week. When they killed their mascot in February 2025 (the campaign that produced their best month ever), the whole thing went from concept to live in six days.

They also refused to plan far ahead: never more than about two weeks out. That sounds reckless and is the opposite. A four-week content calendar guarantees that everything you publish was decided before anything happened.

No approval chain, but a written boundary

Nobody signed off individual posts. Their creative director put it plainly: “The biggest enemy of creativity is those lengthy, rigorous approval processes.”

The part that gets dropped from the retelling is the constraint that made it survivable. From their own social lead: “Risks are more likely to be approved if there’s a direct connection to our brand. Not just stirring the pot to stir the pot.” And a mascot specialist, formerly of Sesame Street, advised on where the character’s limits were, by advising rather than blocking.

Speed without a boundary is not this method. It is how brands end up apologising.

The comment section as the brief

Four different Duolingo executives have said a version of the same sentence in public over two years. “The comment section is our social brief, and the community really drives what we do.” It is not a slogan they repeat; it is a description of where the ideas came from.

The mechanism underneath it is more specific and more useful: they replied under other people’s videos, not only their own. Their lead described their comments as functioning like a push notification, but only when the comment was relevant to that video. “We never commented something as blatant as ‘do your lesson’.”

And they never used a call to action. Their own shareholder letter says it: “we don’t sell directly in our campaigns, you’ve never seen a ‘try Duolingo’ button on any of our social posts”.

What it cost, which nobody puts in the case study

The person who built this account took medical leave in 2024. In her own words afterwards: “I would get three hours of sleep at night. I would be incessantly trying to figure out, how do I be creative?” She left the company in August 2025.

This belongs in the story. The output everyone admires came from a very small number of people operating at a pace that is presented as a virtue and was, for at least one of them, not sustainable. If you are reading this as a one-person marketing team, that is the part to take seriously, more seriously than the tone of voice.

Two failures, in public

February 2025. The mascot’s death was a genuine success by every available measure, and it was also criticised for reusing a joke invented in the comments by a Black creator, without credit. Their acknowledgement post reached 818,000 views. The original had ten million. If your brief comes from the comment section, this is the failure mode of the method, and it is worth building a rule against before it happens to you.

April–May 2025. A memo about becoming “AI-first” leaked, and the account lost more than 400,000 followers in three weeks: the drop starts before the posts were deleted, which places the cause on the memo rather than the deletion. They deleted everything on 17 May, went quiet, and came back with a video whose caption read: “DUOLINGO WAS NEVER FUNNY. WE WERE.”

Their CEO walked it back in four separate public statements over ten months, ending with “At the end, we backtracked”. The lesson is not about AI. It is that an audience built on a personality will hold the company to that personality, and a memo about replacing people is a story about personality whether you meant it that way or not.

The admission, and what it proves

In its quarterly letter of 5 November 2025, the company wrote that daily active user growth “was slightly slower than Q2 in part because we posted less ‘unhinged’ content on our English-speaking social media accounts”, and that having resumed, they had seen a significant increase again.

The market took roughly a quarter off the share price that day. Daily active user growth across those quarters went from +40%, to +36%, to +30%, to +21%.

Set against that, the positive evidence is thin. Asked for proof that the account brought learners, the company’s answer was a self-reported onboarding survey: people who chose “TikTok” from a dropdown. No holdout, no geo-test, no modelling. A spike in fame mechanically increases how often people tick the famous box.

Which leaves the honest summary: the strongest evidence that this work produced growth came from the period when they stopped doing it.

What you cannot copy

The menacing owl was not invented by the marketing team. It was a community meme from 2017 to 2019, with posts in the hundreds of thousands of notes, and the company started using it officially in March 2019, eighteen months before the TikTok account was relaunched. They did not create the character. They noticed one that already existed and stopped resisting it.

The timing was also unrepeatable. September 2021 was early on TikTok for brands, and organic reach was cheap in a way it is not now. Their CEO said so himself in February 2026: “Algorithms have changed enough that it’s harder to do that consistently.” Their own social lead refused to sell the recipe: “You can’t plan virality. But you can plan mentality.”

What you can

Four things, none of which require a mascot or a budget.

Decide your turnaround time and write it down. A week is aggressive; two is fine. What kills accounts is not slowness, it is that nobody ever decided, so everything takes as long as the slowest person is busy.

Remove the approval chain, and replace it with a written boundary. One person decides. The list of subjects nobody jokes about is agreed in advance, in writing, once. That trade is what makes speed survivable.

Read your comments as a brief rather than a chore. Repeated questions are next month’s posts, already written by your audience. This costs nothing and is the single most transferable habit in the case.

Look for the personality that already exists. Duolingo’s advantage was recognising a character the internet had made for them. Most brands have something similar and smaller: a recurring complaint, a nickname, a thing customers always say. It is usually in your comments, which is where this list started.