PO Box 26214, Baltimore, MD 21210
855-721-MDDC
kberrier@mddcpress.com
If you’ve ever opened TikTok “just for a minute” and then looked up an hour later wondering what happened, you’re not alone. You felt it—the pull, the rhythm, the uncanny sense that the app knew exactly what you wanted next. From a technical standpoint, it’s impressive. From a human standpoint, it’s a little unsettling. And from a marketing standpoint? It’s fascinating.
For years, marketers have tried to reverse-engineer TikTok’s recommendation system by chasing trends, decoding hashtags, and copying whatever went viral yesterday. But recently, something far more valuable happened—not from a brand or a platform, but from a newsroom.
Jeremy Merrill, a data reporter at The Washington Post, led one of the most ambitious algorithm investigations we’ve seen to date. Instead of guessing how TikTok works, he and his team measured it—by analyzing the real feeds of more than 1,100 volunteers and more than 15 million videos watched over a six-month period
peeking-behind-tiktoks-curtain-…
This wasn’t a marketing experiment. It wasn’t built to sell ads or grow followers. It was journalism. And that’s exactly why it matters so much to marketers.
Why This Study Was Different
Most algorithm analysis relies on surface-level signals: what gets the most views, what trends globally, or what creators say is “working.” But TikTok doesn’t work at the surface level. It works at the personal level.
Every TikTok user’s “For You” page is uniquely theirs. Your feed is not my feed. And that’s what makes TikTok so powerful—and so difficult to understand.
Instead of scraping popular creator pages, Merrill’s team asked users to download their own TikTok data files—buried deep in the app’s settings—and voluntarily share them. Those files contained every video watched, when it was watched, and for how long. On their own, the files are nearly useless. Together, they became one of the largest real-world datasets ever used to study a social recommendation engine.
The result was a detailed, interactive map of TikTok’s content ecosystem—showing which topics cluster together, which repel each other, and how the algorithm quietly learns not just what you like, but what you don’t.
TikTok Isn’t Just Learning Your Interests—It’s Learning Your Opposites
One of the most eye-opening findings was how strongly TikTok defines content boundaries.
If you watch a lot of gaming content, you’re unlikely to see “summer vibes.” If you engage with car videos, you’ll see less celebrity culture. If you fall deep into Taylor Swift content, you’re likely on what the researchers jokingly called “Taylor Swift Island”—a content cluster so distinct it barely overlaps with anything else
peeking-behind-tiktoks-curtain-…
From a marketing lens, this is critical.
TikTok doesn’t just group people by demographics. It groups them by behavioral proximity. Two users of the same age, income, and geography may live on completely different sides of the content map. And TikTok will treat them accordingly.
This is why broad targeting assumptions fail on the platform—and why relevance beats reach every time.
The Stickiness Problem: Some Content Is Harder to Shake
Another major finding centered on what the researchers called “stickiness.” In simple terms: how hard is it to undo the algorithm’s assumptions once you engage with a topic?
Mental health content stood out.
When users watched a mental health-related video—topics like anxiety, depression, ADHD, or self-care—it took more than two skipped videos to return their feed to its previous balance. Other categories, like sports or entertainment, reset far more quickly
peeking-behind-tiktoks-curtain-…
Why does this matter?
Because TikTok isn’t just optimizing for clicks. It’s optimizing for long-term engagement. And some topics—especially deeply human ones—signal something more complex than “interest” or “disinterest.” Skipping a mental health video may not mean “I don’t care.” It may mean “not right now.”
The algorithm appears to have learned that nuance.
For marketers, this is both a caution and an opportunity. Certain categories carry more algorithmic weight. They linger longer. They shape feeds more persistently.
Brands operating in wellness, lifestyle, beauty, or personal development should understand that engagement in these spaces has longer-term consequences—for better or worse.
Viral Isn’t the Point—and It Never Was
Here’s another stat worth sitting with: of the 15+ million videos analyzed, nearly two-thirds were seen by only one participant in the study
peeking-behind-tiktoks-curtain-…
.
That’s not a failure of TikTok. That’s the system working as designed.
TikTok is not just a viral machine. It’s a distribution engine for niche relevance. Most videos are not meant to reach millions. They’re meant to reach someone. The right someone.
For brands, this reframes success. The goal isn’t to be everywhere. It’s to be exactly where you belong.
What Marketers Should Actually Take Away
This investigation wasn’t built for advertisers—but it delivers some hard truths we can’t ignore:
- Your audience is more fragmented than you think. Cultural relevance is contextual, not universal.
- Behavior matters more than identity. TikTok learns through action, not intention.
- Content adjacency is power. What you’re associated with algorithmically matters as much as what you say.
- Human systems are now shaped by machine logic. And no one fully understands that logic—not even the platforms themselves.
We’ve handed enormous influence over our media diets to systems optimized by computers, not people. That doesn’t make them evil—but it does make them opaque. And opacity is dangerous if we stop questioning it.
The Bigger Picture
At its core, this work is a reminder of why professional journalism matters. This level of insight doesn’t come from hot takes or influencer threads. It comes from time, rigor, trust, and the willingness to ask uncomfortable questions.
For marketers, the lesson isn’t “game the algorithm.” It’s understand the environment you’re operating in. Respect it. And remember that behind every data point is a human being—scrolling, reacting, learning, and forming a worldview one video at a time.
The algorithm isn’t just showing us content.
It’s shaping how we see the world.
And that’s something worth paying attention to.