AtomSlice

220,000 AI Short Dramas, Only 0.47% Passed 100M Views

One AI short drama, three view counts: 4B, 6.6B, 12B. The creators never bought traffic or counted revenue. The repeatable part is not the tool.

· Guide ·Chinese
220,000 AI Short Dramas, Only 0.47% Passed 100M Views

One show, three view counts.

In early November, Shangguan News reported more than 4 billion total views across the web. Banyuetan, in the same period, also said 4 billion, with 10.98 million collections on the Hongguo platform. A November 9 report from Jiemian News put cumulative views above 6.6 billion and collections at 11 million. In a November 9 interview published by The Paper, the number had become "over 12 billion."

All three were describing Chi Xie Ru Song, an AI-made time-travel drama. It went live on August 21 and, over a little more than two months, ran ten seasons and close to 1,000 episodes, each under two minutes long.

The three numbers differ by a factor of three. That is not a reporting error. "Views" simply has no standard definition in the short-drama business: whether duplicates are removed, whether clipped segments count, whether one person watching ten episodes counts as one view or ten — every platform sets its own rules.

But something else here is more interesting than the numbers.

After the show took off, a story circulated online: 1 million yuan of compute produced tens of millions in monthly returns; one team member had entered the field carrying hundreds of thousands in debt and cleared it within months.

A Jiemian News reporter put that story to the chief planner, Zhang Hansheng. His answer: the team only knows how to upload. They do not buy traffic and do not understand traffic buying. The compute bill really was 1 million yuan, but they had not yet had time to calculate returns, and the specific income was not something he would disclose.

In an interview with China Business Journal, carried by Tencent, he denied it more directly: financially they were certainly much better off than before, but they still had not had time to work out the returns.

One easily missed line in the same batch of reporting came from the Hongguo short-drama side: platform revenue share is not calculated from view counts; it is tied more closely to watch time.

That line deserves to sit with for a moment. It means the "X billion views" you see trending and what the show actually earned are separated by a conversion formula, and that formula is not in the audience's hands.

I do not think this is a scam. Quite the opposite: the people watching from the outside did the arithmetic wrong themselves.

What actually made me want to write this is how the show was made.

There were six core creators, and not one had formal film or television training. The chief planner is 54. The team included a former ride-hailing driver, a wholesale delivery self-employed worker, and a mother who needed to pick up her children. They were spread across Guangdong, Anhui, Hubei, Jiangxi, Fujian and Hebei; most have still never met in person, and the whole thing ran on WeChat groups and shared cloud drives.

They did not appear out of nowhere. They started making AI film and video together last year, first taking commercial orders from other people, and spent more than a year of real work filtering each other out — Zhang Hansheng says that among the people who took the same online courses with them, a large number dropped out along the way.

Then came choosing the source material.

When he picked the original novel behind Chi Xie Ru Song, its traffic on the Fanqie Novel platform was only in the tens of thousands. It was not famous. There were two reasons it got chosen: first, although it is a time-travel story, the protagonist has no "system", which sets it apart from the wall-to-wall overpowered time-travel plots; second, the author writes well, with careful attention to Song-dynasty institutions, official titles, and the way people of the period actually spoke.

Notice those two: one is "against the formula", the other is "credible detail". Neither is "this genre is hot right now".

Now look at how they defined acceptable quality.

Making video with AI means "drawing cards": generate the same shot repeatedly and keep one. The common industry floor is "good enough to use" — reshoot only when something visibly breaks. This team's standard was: a merely flat, 60-out-of-100 shot counts as a dead card even if nothing is broken. Some highlight shots were drawn dozens of times.

Li Chaohua put it plainly. At peak capacity the team ran five groups and produced 20 assembly-line dramas a month, burning tokens every day, and quality did not keep up. After the hit, they made a decision that ran against the trend: cut monthly output to single digits and spend all the saved time reading the source novel and thinking about the content.

The result: token consumption per drama went up, while total consumption went down.

There is a line from that Tencent interview I copied down: AI lowers the barrier to getting something made; it does not lower the cost of making it well.

So the truth of this affair is probably a little colder than "a grassroots team made good".

A November 10 article on 36Kr gave one set of numbers: in the first half of 2026, more than 221,900 new AI short dramas went online across the web; 1,055 of them passed 100 million views, a hit rate of about 0.47 percent.

Less than five in a thousand.

In the same month, Zhang Hansheng said in an interview that in the second half of this year many AI short-drama companies across the country were shutting down in bulk, and many companies he knew were laying people off.

On one side, more than 200,000 new dramas pouring in; on the other, a little over 1,000 passing 100 million views and large numbers of companies going under. This is what "the barrier came down" actually looks like — when the barrier comes down, more people come in than there are opportunities.

I am not going to follow that with "AI is a scam". The tools genuinely work. They brought production costs down from several hundred yuan per minute, released IP that had been sitting in web-novel platforms with no way to monetize it, and let six people who have never met make 1,000 episodes. None of that is a story; it already happened.

But if you are about to quit your job to make AI short dramas, or you have just signed up for a course, I would suggest answering three questions first.

First, can you say why this particular source? Not "this genre is hot right now", but specifically where this book differs from others and which passage made you unable to sit still. Zhang Hansheng's answer was "no system" and "careful writing", not "time travel is hot".

Second, which item in your cost sheet cannot be compressed? Their incompressible items were compute (over 1 million yuan in total, about 100,000 per drama) and the time spent reading the original novel. If the only item on your cost sheet is "buy a subscription", you have not started yet.

Third, do you treat uploading as the finish line or the starting line? When season one went live, collections were six or seven hundred thousand, and the team's own verdict was "lukewarm". It fermented season by season through word of mouth. If your expectation is an instant hit on launch, you are betting on luck — and when Zhang Hansheng reviewed it himself, he put luck first, with choices and effort after it.

Writing to this point, I think the most learnable thing about this show is not how it became a hit, but the two years before it: taking commercial orders, being filtered out, giving up, redoing work, making 20 bad dramas a month, and then voluntarily cutting output.

None of that is sexy, and none of it makes a good short video.

So I will leave you with a choice: if AI drove the barrier to making content down to zero, would you make volume first and bet on one hit, or make one piece you are willing to put your name on? Tell me in the comments which side you are on.


Sources and caveats: the 4B, 6.6B and 12B view counts come from Shangguan News and Banyuetan, Jiemian News, and The Paper respectively. Their definitions differ, so all three are presented side by side rather than collapsing them into one conclusion. The August 21 launch date comes from the Jiemian News and China Business Journal interviews; the roughly 100,000 yuan of compute per drama comes from the China Business Journal interview with director He Xiushan. The 221,900 dramas and the 0.47 percent figure come from an industry article republished by 36Kr on November 10, originally published by Chief Business Review, and are second-hand data. On revenue share, only the Hongguo platform's wording is used — it is not calculated from view counts and is tied more closely to watch time — with no attempt to estimate an amount. No income figure appears anywhere in the article; only the creators' replies that they had not yet calculated returns and would not disclose specifics are kept.