Start with a number.
There’s a video on Bilibili called 【東方】Bad Apple!!PV【影絵】. It was published 17 years ago. In the past year alone, it picked up another 2.03 million views. In the comments, someone wrote, “This is a faith video that always has someone watching.” Someone else wrote, “Once a day.”
17 years. How many platforms have risen and vanished in that time? How many accounts went viral and then went silent? This video is still growing.
My first reaction: that doesn’t make sense. My second reaction: that’s exactly the point—AI can produce ten thousand articles a day, but it can’t produce a video that lives for 17 years.
According to Bilibili’s 2026 Q2 earnings report, total revenue hit 7.94 billion yuan, up 8% year over year. Net profit was 700 million yuan—eight consecutive profitable quarters. Daily active users reached 117 million, with average daily time spent at 113 minutes, up 8 minutes year over year.
At the platform’s 17th anniversary, Chen Rui said something I think is the most valuable line in the entire report: “Internet content supply is already in surplus, but good content remains scarce.”
That’s not a platitude today. It’s a judgment call.
1. AI Turned “Output” Into a Commodity—Scarcity Moved to Retention
Let me start with a mistake I made.
Last year I tried using AI to mass-produce topic drafts—20 a day, published, and the read counts were fine. But three months later I looked back at the backend data: the long-tail traffic on those drafts was basically zero. They died the moment they were published. No search traffic, no saves, no comments.
That’s not AI’s fault. I misjudged where scarcity lives.
The old competitive logic in content was “whoever produces more, publishes faster, wins.” AI punched straight through that. Daily video submissions are up 28% year over year (per Bilibili’s 2026 Q2 report). The supply side is completely saturated.
When output stops being a moat, retention becomes the moat.
What’s retention? It’s whether anyone still searches for, watches, or discusses a piece of content a year after it’s published.
One data point from the earnings report says a lot: watch time for videos over 5 minutes grew 18% year over year. Users aren’t short on time. They just won’t give it to something they’ll forget the moment it ends.
Video podcast daily watch time broke 100 million minutes, with cumulative viewers exceeding 236 million. Lu Yu’s show Slow Talk—one episode with Jiang Sida ran nearly 4 hours and pulled close to 8.2 million views.
4 hours. In an era where short videos fight for your attention in 3 seconds, people are willing to spend 4 hours listening to two people talk.
What does that tell you? Users are voting with their time—and they’re voting for content worth spending time on.
2. “Outliving Time” Isn’t Mystical—Break It Into Three Observable Metrics
“Good content outlives time” sounds vague. Let me try to break it into something verifiable.
Whether a piece of content survives three years usually comes down to three metrics:
| Metric | What it looks like | Counter-example |
|---|---|---|
| Searchability | Users actively search for it, not just scroll past | Pure trend memes—nobody searches once the moment passes |
| Rewatchability | The same people come back for a second viewing | One-time information, consumed and discarded |
| Discussability | The comment section grows new conversations | No point of view, nothing to respond to |
Check these against Bilibili’s cases:
Shi Pin Dao‘s “Lost in Tokyo” was published 3 years ago, and people are still writing hundreds of words of reflection in the comments—that’s discussability.
The CCTV version of Journey to the West went online in 2020. For six years, people have been checking in at different times—nostalgic old viewers and first-timers alike—that’s rewatchability.
Drawing Sword remix videos have been around for nearly a decade, and views keep climbing—that’s searchability. Every time a new meme appears, they get dug back up.
Of the three, discussability is the hardest to build and the most valuable. It requires the content itself to have a point of view, room to breathe, and something that makes people want to respond.
Bilibili’s monthly interactions hit nearly 17.4 billion, up 9% year over year. Among them, comments over 100 characters jumped 67% year over year. Long comments are growing—meaning users aren’t just consuming content. They’re participating in it.
3. Long Tail Isn’t Luck—It’s Decided by Community Structure
A lot of people assume the long tail is “getting lucky and having the algorithm dig you back up.”
Not entirely.
The underlying mechanism of the long tail is: does the content have a structure that keeps generating new interactions?
Bilibili’s danmaku, comment sections, and “watching now” counter essentially keep injecting new signals into old content. A video published years ago can get pushed back to the front page because of one new danmaku or one new comment.
That’s not mysticism. That’s mechanism.
Now look at the creator side: creators with 1,000+ followers grew 30% year over year; those with 10,000+ grew 21%. Creators who’ve kept publishing for 5 consecutive years reached 1.67 million. And 8.13 million users have maintained 10-year follow-and-interact relationships with creators.
10 years. 8.13 million people.
Behind that number is a very simple relationship: users aren’t here for one video. They’re here for a person.
Creator Zhou Yang HOPICO has been publishing since 2018—8 years, 2.639 million followers, steady 800K views per episode. His episode with Jam Hsiao broke 7.5 million views, and someone joked in the comments, “Jam Hsiao’s doing well for himself, getting to talk to Zhou Yang.”
That kind of interaction isn’t built by one viral hit. It’s built episode by episode over 8 years.
The essence of the long tail is the compounding of trust.
4. What Topics Are Naturally Decay-Resistant?
Not all content outlives time. Some topics are doomed to a short life from the day they’re born.
Here’s a rough standard I use—you can run your own topics through it:
Decay-resistant topics:
– Solve a long-standing problem (how to do it, why it happens, what to do about it)
– Document an irreplaceable real experience
– Propose a viewpoint that sparks ongoing debate
Naturally short-lived topics:
– Pure trend-chasing—once the trend passes, the content resets to zero
– Pure information relay—users search once and they’re done
– “Objective summaries” with no stance—anyone can watch, nobody remembers
A concrete example. Both are about AI: “AI tool reviews” go stale in three months because tools iterate too fast. But “how content creators should position themselves in the AI era”—people will still be asking that three years from now.
The former is information. The latter is judgment. Information expires. Judgment settles.
Chen Rui said “quality content is the most effective way for a content platform to sustain growth.” I think that applies just as much to individual creators—your judgment is the only asset AI can’t replace.
5. Three Things You Can Do Today
Enough talk. Here are three things you can act on today:
First, dig up what you published three years ago and check if anyone’s still watching.
If yes, analyze why it’s still alive. If no, figure out which of the three metrics it’s missing. This one move is worth more than reading a hundred operations guides.
Second, write down the “shelf life” of your next topic.
Before publishing, ask yourself: will anyone search for this a year from now? If the answer is “no,” treat it as traffic content—don’t expect it to deliver long-term value.
Third, seriously reply to one long comment.
Comments over 100 characters are up 67% year over year—meaning more users are willing to say something real. Reply to one, and you might gain a 10-year follower. Those 8.13 million 10-year relationships were built one comment at a time.
FAQ
Q: Can AI-generated content outlive time?
A: Tools don’t determine content lifespan—judgment does. AI can help you write faster, but “what to write, why write it, and from what stance” are three questions you still have to answer yourself. Answer those well, and AI-written content can live just as long.
Q: Is deep content always better than short content?
A: No. Short content has its own value—it handles reach. But if what you want is an asset that “outlives time,” deep content has a higher probability. The two don’t conflict. The key is not treating traffic content as an asset.
Q: How do small creators build a long tail without community mechanisms?
A: Through search. Write your content so it can be found—put searchable keywords in the title, and concrete questions and answers in the body. Search is the fairest long-tail entry point. It doesn’t care about your follower count.
Final Thoughts
The stronger AI gets, the more valuable slow content becomes.
That’s not consolation. It’s something Bilibili has verified over 17 years, with a 17-year-old video, and with 8.13 million people’s 10-year attention.
Output gets flattened by AI. Judgment doesn’t. Trends recede. Good content doesn’t.
Do you have a piece of content that’s still growing long after you published it? What type is it? Drop it in the comments—I’ll pick a few and break them down together.