PalimioBefore, During and After the Owl Went Silent: Duolingo's TikTok in Numbers
Duolingo's TikTok was the most-studied brand account of the decade. In the first half of 2025 it was pulling 6M-view medians and 11% engagement rates on a 6M-follower account, numbers almost no other brand has ever posted. Then in May, something snapped. Leadership changes, creative-team turnover, the fake death of the mascot, an AI-first repositioning that didn't land. The feed kept publishing but the numbers fell out from under it.
April 23, 2026
It's April 2026 now and the account is rebuilding.
We ran the full 2025-2026 catalogue (450 posts) through Palimio, tagged each one across six creative dimensions, and split the data into three eras: Peak (Jan-Apr 2025), Break (May-Dec 2025), and Rebuild (2026 so far). Here's what the data actually says.
Act 1: the Peak (Jan-Apr 2025)
131 posts, 2.6M median views, 11.2% engagement rate. The numbers on this run are absurd. Top post: the fake death of Duo the Owl, 67.6M views and 14.4% engagement. Second: Duo's "night grind routine" 53.3M views at 15.6% engagement. These aren't outliers, they're the pace. Ten posts cleared 20M views in this window alone.
The formula was tight and specific. Palimio's Content DNA breakdown for this era ranks formats of Mascot Product Demo first (4.9M median), Humorous Meme Overlay second (3.2M) and Mascot Dance Routine third (2.8M). Every high-performing format had the mascot at the centre of it. Topics tell the same story. Brand Humour posts (53 of them) pulled a 3.1M median. Product Promotion pulled 7.5M median across 5 posts. The account wasn't a language-learning feed. It was a mascot-led brand comedy show that happened to sell a language app.
Emotional framing was equally narrow: 102 of 131 posts tagged Humorous, 99 tagged Entertaining. Anything else was noise.
Act 2: the Break (May-Dec 2025)
197 posts. Median crashes to 604k, a 77% drop from peak. Engagement rate falls from 11.2% to 6.8%.
The most interesting thing in this data isn't the drop. It's what happened inside it.
The account kept publishing heavily. Volume actually went UP (197 posts in 8 months vs 131 in 4). POV Sketch post count doubled (20 to 43). Mascot Dance Routine count nearly doubled (28 to 51). Pop Culture Parody tripled (14 topics-tagged to 40). Brand Humour stayed the dominant topic at 63 posts. They didn't pivot the playbook. They ran it harder and it stopped working.
And then the algorithm did something strange.
The two biggest-view posts in the entire dataset happen in this era, both in September 2025: a 111.5M-view office dance sequence, and a 102.1M-view "6-7 meme" mascot sketch. Both pushed to nine-figure reach. Both had engagement rates under 1% (0.9% and 0.4% respectively, roughly 1/15th the account's baseline).
What Palimio's correlation view shows for this era is that the link between views and engagement loosened noticeably. Log-log correlation between likes and views in the Peak era was r=0.96 (every view earned). In the Break era it dropped to r=0.86. Saves-to-views went from 0.95 to 0.81. The feed was picking up massive algorithmic pushes that didn't convert into the follow-on engagement Duolingo's posts used to produce. Reach without resonance.
That's probably the real story of mid-2025. Not a creative collapse (the posts tagged the same on paper). A resonance collapse. The formula kept getting distribution, and then a random share-your-followers's-follower-count-watched pattern pushed a few clips past 100M, but the audience that used to engage with Duolingo was already somewhere else.
Act 3: the Rebuild (2026)
122 posts so far. Median views drops further to 95.9k. But the engagement rate comes back to 9.97% (the highest since February 2025), and by March 2026 hits 11.2% on 43 posts. The audience got smaller. The audience got stickier.
The creative recipe is visibly different now.
Content DNA's top 2026 format is Long Carousel (205k median, 11 posts). Second is Interactive Game (158k, 3 posts). Third is App UI Lesson Screenshot (129k, 3 posts). Mascot Dance Routine, which was the workhorse of both Peak and Break, now sits at 95k median across 23 posts, well below the catch-up formats.
The topic shift is sharper. Language Learning is now both the most-posted topic (41 posts) AND the best-performing (126k median). Brand Humour, which drove the entire Peak era, has been cut in half by post count (53 > 32) and sits at 74k median, barely above the account floor. Fan Engagement is up to 140k median across 16 posts.
The account is visibly pivoting back toward the product. The mascot is still present but no longer the whole personality. There's a language-learning feed underneath the comedy now, not the other way round.
And the Rebuild era's top two posts are where the new voice shows. The Duo / Alastor Hazbin Hotel "I love you" crossover in Korean (6M views, 24.1% engagement rate, the highest of any post in the dataset). The wrestling-style "friend who lost their streak" sketch (3.2M, 24.3% engagement). These aren't brand-humour-for-its-own-sake posts. They're language-flavoured jokes aimed at specific subcultures (fandom + streak-keepers). Smaller hits, but the tightest audience-content fit they've had in over a year.
The counterintuitive finding
The obvious reading of this data is that Duolingo's account died. The numbers don't support that.
The account's median engagement rate in 2026 is already higher than it was during the Break era, and within a point of the Peak. Correlations between every engagement signal and views have snapped back to Peak-era tightness (r=0.94 for likes, 0.93 for saves). The average 2026 post makes fewer people see it but actually makes them care.
What collapsed wasn't the audience. What collapsed was the reach subsidy that a dominant-brand-on-TikTok era gave them. The 6M-median was an era of the platform, not an era of the content. The 95k-median with 10% engagement is the account's actual steady-state.
What the next 90 days will tell us
Three things Palimio will be watching over the next quarter:
Does the Language Learning topic keep expanding? It's currently 41 of 122 posts (33%) and has the best median of any topic. Brand Humour was 40% of the Peak era and is 26% of Rebuild. If Language Learning crosses 50% of the calendar, the pivot is complete.
Does the engagement rate hold above 10% as volume grows? March 2026 hit 11.2% on 43 posts, Peak-era territory. Sustaining that through a 60+ post month would close the loop on the rebuild.
Do the fandom-crossover formats scale? The Hazbin Hotel crossover is the highest-engagement post in the dataset. One post doesn't make a pattern. Five do.
What nobody should miss is the shape of this. Duolingo didn't lose its audience. It lost a version of TikTok that was never coming back. The account that emerges from this year will be smaller in reach and deeper in fit than the one that peaked in February 2025. On the data, that's probably a healthier account to run. A 10% engagement rate on a 100k-reach post is worth more to a business than a 1% rate on 100M.
The owl went silent for a reason. When it spoke back up, it said something slightly different.
What this post is
Every breakdown above came from a single command in Palimio: plug in a TikTok/ Instagram handle, auto-tag every post across format, topic, tone, hook, scene, production elements, run real statistics on the patterns, split by date-range. No tagging spreadsheet, no six-figure enterprise contract.
If you run a brand, agency, or creator business, Palimio will transform the way you use social media data. We'd love to show you what it surfaces.