China's micro-drama boom is already trading prompt know-how

2026-10-14 · Christoph Heuwieser

Video: How Can Society Withstand This, Atrioc on his Big A clips channel, 13 October 2026

Neon illustration: a long log of faint, mostly failed attempts, one bright final line at its foot, and a copy of only that line lifted onto an amber price tag, next to the post title.

Atrioc's latest Marketing Monday, How Can Society Withstand This (published 13 October 2026 on his Big A clips channel), is mostly about micro-dramas: vertical series with episodes of one to two minutes, a cliffhanger at the end of each, and a paywall right after it. About 23 minutes in, he reaches a detail I keep coming back to. In China, where the format started, people now sell courses on how to make these shows. Some sell the exact words to type.

What the video shows

Atrioc builds his case in three steps. First the format: low budgets and very fast shoots. Second, AI. He reports that Chinese output jumped to hundreds of thousands of titles this year, nearly all of them AI-made, and that mid-tier actors are losing work. Third, what follows from that. In his words, "a new gold rush has popped up of people selling courses on how to create your own".

He shows people livestreaming lessons on how to type exact words to get an exact result. One seller offers the exact prompt for a crying scene, timed to the tenth of a second, and keeps the full version back until you pay. Another sells sets of expression words, such as "one-sided smirk, slightly raised eyebrow, steady gaze". His phrase for it is "sample packs of words".

He isn't cheering. He calls a lot of the output slop, spends real time on the actors who lost their income, and says that what comes next "doesn't seem good".

How big is this, really?

The numbers depend on who counts, and what they count.

China, 2024 baseline. The Chinese Academy of Cyberspace Studies put the audience at 662 million and the market above 50 billion yuan as of December 2024 (reported by China Daily on 2 January 2026). The video's China figures match this earlier baseline.
China, 2025. In February 2026 an official of the National Radio and Television Administration (NRTA) said the domestic market had more than doubled in 2025, to over 100 billion yuan (about US$14 billion), with nearly 700 million users.
China, 2026 so far. In September 2026 the NRTA said users had passed 800 million. It also said 430,000 micro-dramas were released from January to August, 13 times the 2025 total, and that more than 90% of them were AI-generated. The video's "400,000 in six months, 95% AI" is in the same range, but not the same figure.
Distribution. Much of the domestic audience sits on ByteDance properties. Its free, ad-funded Hongguo app had 168 million daily and 356 million monthly users in July 2026, according to QuestMobile data reported in September.
Abroad. Sensor Tower counted US$2.98 billion in in-app revenue for short-drama apps in 2025, up 115% on the year. In the first quarter of 2025, ReelShort (about US$130 million) and DramaBox (about US$120 million) were the top two, and the US brought in 49% of the revenue. These are App Store and Google Play purchases only, without ads or web payments.

A few caveats. The 100-billion-yuan figure includes ad-funded viewing, so it can't be compared directly with app-store revenue. And I found no independent audit of the official numbers. I'd read all of them as orders of magnitude.

The know-how market

The course side is easy to confirm. Shanghai Open University offers a five-day, 40-hour AI short-drama class for ¥3,580 to ¥3,980 (enrolment notice, 22 April 2026). In June 2026 the Beijing Municipal Radio and Television Bureau and the Xicheng District government ran a two-day training for 170 practitioners. Commercial schools advertise prompt writing (one example, 12 August 2026) next to scripting, storyboards and editing, often with salary promises I'd treat as marketing.

There are free materials too. One Chinese prompt handbook, published on 7 October 2026, teaches reusable character and scene cards, prompts for the first and last frame of a shot, and a fixed order for fixing a failed generation: character, then action, then camera, then duration, then style, changing one thing at a time.

What I couldn't confirm independently is the trade in paid per-scene prompt packs. The video shows examples, and they fit everything else, but I have no figures on how many sellers there are, what they charge or how much they make.

A prompt is the last line of a long log

Look at what is actually for sale. A course is a teacher's summary. A prompt pack is the final words that worked. Neither contains the place the knowledge came from: the many generations where the tear ran the wrong way, the face drifted between shots, or the camera ignored the instruction. That handbook needs a troubleshooting order because failure is the normal case. Whoever has worked through those failures knows which change fixed which problem, and which promising idea went nowhere.

That fuller record is what a4sx exchanges. An agent session is the whole trajectory: the goal, every attempt, the errors, the dead ends, the corrections and the result. Today most sessions on a4sx are coding-agent work, not video, but the shape of the good is the same. Someone who acquires a session can read why the final approach won and continue from there, instead of copying the last instruction and hoping.

What I take from it

I don't want to overstate this. China's micro-drama market is large, fast-moving and state-shaped. Its headline figures come from officials and data firms, not audits. Part of the course boom is plainly hype, and the evidence for the prompt-pack trade is mostly anecdotal so far.

The direction still looks clear to me. When generating content becomes cheap, the scarce thing is knowing how to get one specific result, and people start to pay for that knowledge. In China that market already exists, in courses and packs of words. We think the most complete form of that knowledge is the full session, not the final prompt.

Sources

Atrioc (Big A), How Can Society Withstand This, 13 October 2026. Quotes from the video's captions.
China Daily, High stakes in short takes: Micro-drama industry levels up, 2 January 2026.
Bastille Post, China's micro-drama market doubles in 2025, reaching nearly 700 million viewers, 7 February 2026.
Global Times, China's micro-drama users surpass 800 million, 17 September 2026.
BigGo Finance, Hongguo Short Drama's daily active users top all four major long-form platforms, 24 September 2026, citing QuestMobile.
Marketing Dive, Microdrama apps stand out on mobile, 20 February 2026, citing Sensor Tower's State of Mobile 2026.
Sensor Tower, State of Short Drama Apps 2025 Report, July 2025.
Shanghai Open University, AI short-drama creation class, enrolment notice, 22 April 2026 (in Chinese).
Sohu (credited to Shenzhen News Network), AI short-drama training programme advertisement, 12 August 2026 (in Chinese).
Duanju Spot, 2026 Beijing AI micro-drama training concludes, 24 June 2026 (in Chinese).
Director Ready, AI short-drama prompt handbook, 7 October 2026 (in Chinese).

Signing in with Google or GitHub is used to identify your account (your name, e-mail address and profile picture) and to send you receipts and service notices. Browsing the marketplace needs no account.