Why Did That Video Get 10 Million Views? Douyin's Algorithm, Fully Dissected

"I posted and nobody saw it."

The wall everyone hits first on Douyin. On a platform with 900 million daily active users, watching your video disappear into silence.

This is usually not a talent problem. In most cases, it's the result of operating without understanding how the system works.

Douyin's algorithm underwent significant evolution in 2025. This article explains the latest version of that system in accessible terms.

2025: Douyin Made Its Algorithm Public for the First Time

In March 2025, Douyin's president Han Shangyu announced at China's Internet Media Forum that Douyin would publicly disclose its algorithm principles and platform operating system for the first time. A website called "Douyin Safety and Trust Center" was launched the same day, with documentation of the recommendation algorithm's basic logic made available to the public.

This sent shockwaves through the industry. The algorithm had long been called a "black box." Now, for the first time, its overall structure was officially describable.

At the core of what was revealed: a transition from the older approach of "match content tags to user tags" toward a new system built on deep learning prediction of user behavior probabilities.

In simple terms, Douyin is no longer simply saying "this video is about cooking, so show it to people who like food." It is now calculating, in real time: "What is the probability that this specific user will watch this video to the end? Will they comment? Will they share it?" The recommendation engine acts on those probabilities.

The Traffic Pool: An 8-Stage Promotion Game

Douyin's traffic distribution uses the same tiered "traffic pool" architecture as Xiaohongshu, but with more stages and greater granularity. As of 2025, a newly posted video first enters the initial cold start pool, where it is shown to just 200 to 500 users. If the response is strong, it advances to Pool 2 with around 3,000 views, and from there to Pool 3, which covers 12,000 to 15,000 views. Continued strong performance pushes the video into Pool 4 at 100,000 to 150,000 views, then Pool 5 at 400,000 to 800,000, and further into Pool 6 at 2 to 3 million views. The most exceptional content reaches Pool 7, where view counts range from 7 to 11 million, and beyond that the top pool, reserved for videos that break 30 million views.

2025's Most Important Rule Change: From "24 Hours" to "7-Day Maturation"

The old Douyin conventional wisdom said: "Everything is decided in the first 24 hours. If it doesn't explode immediately, it's over."

The 2025 algorithm reform changed this significantly.

The promotion window was extended to a maximum of 7–15 days.

A real example from a beauty account: a skincare review video was stuck at 200 views for three days. Then on Day 4, it suddenly crossed 50,000 — and continued to climb from there.

This is the result of what's been called the "7-day slow-heat promotion" (慢熱推流) mechanism. The algorithm continues testing a video's potential over a week, meaning that a slow start is no longer necessarily a death sentence. Continuing to refine your approach even when early numbers are poor has become more viable than it once was.

The First 3 Seconds Account for 40% of the Score

The most critical metric governing traffic pool advancement is completion rate (完播率) — the percentage of users who watch the video to the end.

According to 2025 Douyin algorithm data, the retention rate in the first 3 seconds of a video accounts for 40% of its overall score.

This makes intuitive sense. Douyin's UI is full-screen auto-play. If a user decides within the first second or two that this video doesn't interest them, they swipe up and it's gone. That departure is recorded. The algorithm reads it as: "This content isn't welcomed by users" and reduces distribution.

This is why, on Douyin, the most important question in content creation is: "What are we showing in the first three seconds?"

Three types of effective openings:

① Start with the conclusion"If you watch this, you'll solve [specific problem]." Lead with why finishing is worth it.

② Start with surprise or counterintuitionAn image or statement that defies expectations. Make them think "Wait — why?"

③ Start with an empathetic question"If you've ever struggled with [X], you need to see this." Make your target viewer feel immediately seen.

Interaction Weight: Shares > Comments > Likes

Alongside completion rate, interaction rate (互動率) — how much users respond — is equally important.

In Douyin's 2025 algorithm, the relative weight of user actions looks like this:

Share (转发) > Comment (評論) > Like (点赞) > Save (收藏)

Shares carry the highest weight because "wanting to send this to someone else" is the most direct signal that the content has genuine distribution value.

The implication: a video that makes people share is more algorithmically rewarded than a video that makes people like. Emotional resonance, humor, surprise, practical utility — anything that triggers "I need to send this to someone" is the surest path to viral spread.

S/A/B/C: The Hidden Account Rating System

One more system worth knowing: the four-tier creator classification Douyin introduced in 2025.

Based on historical performance, every account is rated S, A, B, or C.

S-tier: Traffic allocation doubled. Astructural advantage from the first moment of posting.

A-tier: Normal distribution plus somedegree of preferential treatment.

B-tier: Standard traffic pool startingpoint.

C-tier: Stricter review process;commercial feature access restricted.

What makes this significant: the rating is based on content quality consistency, depth of user response, and historical account performance — not follower count.

This means a small account that consistently produces quality content can work its way toward S-tier. And a large account that starts producing mediocre content will see its rating decline. This mechanism is what keeps Douyin's competition relatively fair.

Content the 2025 Algorithm Rewards

Based on current data, Douyin is allocating enhanced traffic to these types of content:

① Cultural heritage, traditional crafts, rural/agricultural content (三农)Aligned with Chinese government policy, content with cultural preservation value receives elevated treatment. Japanese traditional craftsmanship and artisanal manufacturing content may benefit from similar tailwinds.

② Knowledge and in-depth analysisIn 2024, a 450-minute "Dream of the Red Chamber" lecture video accumulated 300 million views. Long-form, substantive content is being rewarded — overturning the old rule of "shorter is always better."

③ Practical how-to content"How to choose the right X," "I tried Y and here's what happened" — actionable, useful information consistently achieves high completion and save rates.

Content that is losing traffic:

Videos with a compelling hook but thin substance

Content where AI generation or editing is obvious

Mass-produced templated content with repetitive structure

The Gap Between Understanding the System and Consistently Using It

Having read this far, you may feel: "Alright — completion rate, first three seconds. I've got it."

The direction is correct. But effective operation means building on this understanding with detailed analysis after every post — identifying where viewers dropped off, feeding that insight into the next video, refining the approach.

The algorithm itself is adjusted regularly. What "worked" six months ago may not work today.

Between "knowing the rules" and "producing consistent results" lies a practice of daily observation and iteration.