AI 网文出版全景图 · 写作的任务化先行者 · 2026-07 · 文字内容工业Web Fiction & Publishing × AI · The Assembly Line of Words · Jul 2026
最工业化的写作,最先被接管
The most industrial writing falls first
网络文学是人类有史以来最工业化的写作形态——在 AI 到来之前,它就已经把创作流水线化、任务化、KPI 化了;因此它也是最容易、最先被 AI 接管的创作形态。而这场接管暴露的定律是:内容通胀,筛选升值,关系为王
Web fiction is the most industrialised writing humanity has ever practised — long before AI arrived it had already turned creation into an assembly line of tasks and KPIs; which is why it is the first form of writing AI takes over. And the takeover exposes the law: content inflates, curation appreciates, the relationship is king
本图相信:最小单元不是「一本书」,是「一章更新」——日更是作者与读者之间的契约;「黄金三章」(第一章冲突、第二章金手指、第三章打脸)本质是为推荐算法与读者耐心阈值定制的留存工程。当写作早已是有套路库、有工序、有计件报酬的任务系统,大模型只是接管了一条现成的流水线(B)。This map believes: the atom is not «a book» but «one chapter’s update» — the daily update is the covenant between author and reader; the «golden three chapters» (conflict, cheat-power, face-slap) are retention engineering tuned to the algorithm and the patience threshold. Where writing was already a task system of trope libraries, work orders and piece rates, the LLM merely took over a ready-made line (B).
本图不相信:「AI 将消灭作者」——反 AI 的第一枪恰恰在最工业化的网文圈打响(国内首例创作者集体反对 AI 训练条款);读者抵制「AI 味」、平台单月拒签低质书逾 10 万本。任务化的部分被 AI 拿走,人格化的部分被 AI 反向凸显:AI 擅长生成文字,不擅长成为一个被追随的人。It does not believe «AI abolishes the author» — the first shot against AI was fired precisely in this most industrial quarter (the first collective creator revolt against AI-training clauses); readers police the «AI flavour» and one platform rejected 100k+ low-grade books in a single month. AI takes the taskified part and throws the personal part into relief: it generates prose well; it cannot become someone worth following.
主脊:双轨生命周期七环节(网文轨「开书→签约→日更→变现→IP」与出版轨「选题→编辑→发行」并轨对照),②③④标红;三条结构带:免费 vs 付费的模式战争、AI 垃圾书洪水、筛选升值与确权;判断层:IP 与人设 ＞ 文笔,社群关系 ＞ 单本销量,筛选信用 ＞ 生产能力。姊妹分工:传媒公信力(同构)→media,翻译出海→translate,任务化劳动→gig,网红人格→creator,影视改编(在建)→film。
The spine: a twin-track lifecycle in seven links (the webfic track «launch → contract → daily update → monetise → IP» run against the publishing track «acquisition → editing → distribution»), ②③④ flagged; three bands: the free-vs-paid model war, the AI slop flood, curation’s appreciation and rights; the judgment: IP and persona ＞ prose, community ＞ unit sales, curatorial credit ＞ production capacity. Siblings: media credibility (isomorphic) → media, translation → translate, taskified labour → gig, creator persona → creator, film adaptation (in build) → film.
3119.8 万人
中国网络文学作者规模(官方报告,A);对照诚实层:有稿费收入者约 80 万(≈2.45%)、月入 5000 元以下(含无收入)占 68.7%、月入 2 万+ 仅 4.1%(二手一致引用,C)——幂律极端的计件劳动市场China’s web-fiction author population (official report, A); the honest layer: ~800k earn any royalties (≈2.45%), 68.7% make under ¥5,000/month (or nothing), only 4.1% clear ¥20k (consistently relayed, C) — a piece-rate labour market of extreme power law
3549 部/日
低成本推理模型爆火后,番茄男频「首秀」新书的单日峰值(平时千部级);平台上线 AI 审核后回落至 1000–1500 部/日;2026-06 单月拒签低质书逾 10 万本(B)Daily debut novels on Fanqie’s male channel at the post-cheap-model peak (vs ~1k normally); AI review pushed it back to 1,000–1,500/day; June 2026 saw 100k+ low-grade books rejected in one month (B)
2985.6 亿元
2024 中国网文 IP 改编市场规模(+14.6%,B);微短剧 504 亿元超院线票房、一平台年授权改编短剧超 7000 部——即使正文可被量产,世界观资产与粉丝共识仍是上游水源China’s 2024 web-fiction IP-adaptation market (+14.6%, B); micro-dramas hit ¥50.4B — past the box office — and one platform licensed 7,000+ drama adaptations in a year: even if prose can be mass-made, world-assets and fan consensus remain the headwater
1.7 万+ 部
一出海平台 AI 翻译作品累计量(2024 年新增约 3200 部→2026 累计破 1.7 万),畅销榜 TOP100 中 AI 译作占 42%、效率 ×9、成本 −90%+(B)——出海翻译是全链 AI 化最彻底的一环Cumulative AI-translated titles on one international portal (~3,200 added in 2024 → 17k+ by 2026); 42% of its top-100 bestsellers are AI-translated, at 9× speed and −90% cost (B) — the most thoroughly AI-fied link in the chain
⚠️ 口径裁判(先读):① 一免费阅读系「年收入 300 亿/600 亿元」为媒体测算,其母公司明确否认——绝对值存疑,但「免费系体量远超付费系」方向可由月活数据独立佐证;② 「80 万有稿费收入/68.7%/4.1%」为多篇报道一致引用但未定位到原始统计表,标 C;③ AI 垃圾书洪水规模之争并列呈现:一检测机构称某自助类目 77% 疑似 AI vs 行业观察者反驳「约十分之一有显著 AI 贡献」——方向确凿、规模高度不确定;④ 一头部主播 GMV 有自然年独立口径(102 亿)与财年合并口径(143 亿)之分,不可混用;⑤ AI 同质化与长文本一致性的学术结论多基于英文语料与特定模型,迁移到中文网文需谨慎;⑥ 代写报价、「48 小时 500 万字」等工业化描述来自单一/非官方信源,置信度偏低;⑦ 四份深度研究交叉整理(一份约 50 个带源核验为主力);渗透%为编辑估值。
⚠️ Basis rulings (read first): ① the free-reading group’s «¥30B/60B revenue» is a media estimate its parent explicitly denies — the absolute number is doubtful while the direction (free dwarfs paid) stands on MAU data; ② the «800k earning / 68.7% / 4.1%» distribution is consistently relayed but never traced to a primary table — graded C; ③ the slop-flood size dispute is juxtaposed: one detector claims 77% of a self-help shelf is likely AI vs the trade’s «about one in ten new titles with significant AI» — direction firm, magnitude wide open; ④ one top book-streamer’s GMV comes in a calendar-year solo basis (¥10.2B) and a fiscal consolidated basis (¥14.3B) — never mix; ⑤ the homogenisation and long-context studies mostly use English corpora and specific models — port with care; ⑥ ghostwriting rates and «5M words in 48 hours» claims are single-source; ⑦ cross-compiled from four deep-research reports (one ~50-source-verified as backbone); penetration %s are editorial.
◆ 中心装置 · 任务化先行者The central device · the taskification forerunner
最工业化的写作,最先被接管;也最先反弹The most industrial writing falls first — and revolts first
当大模型出现时,网文早已把「写作」重构成有套路库、有工序、有交付节奏、有计件报酬的任务系统:代写工作室千字 5–50 元流水线量产,大纲分「300–600 字简版(供决策)」与「2 万字详版(供写作)」两级工序;头部作者曾维持 3 小时 4000 字的日更节奏(B/C)。AI 接管的不是「文学」,是一条现成的流水线。When the LLM arrived, web fiction had already rebuilt «writing» into a task system with trope libraries, work orders, delivery cadence and piece rates: ghostwriting shops mass-produce at ¥5–50 per thousand characters, outlines split into a 300–600-character decision brief and a 20,000-character production spec; a top author once held a 4,000-characters-in-3-hours daily pace (B/C). What AI took over was not «literature» but a ready-made line.
▸ 接管这一侧(生成端坍缩)The takeover side
编辑自述投稿「两三成」含 AI 润色;低成本模型爆火后单日新书峰值 3549 部;AI 有声书日均产能 500 万字、成本 −90%;出海 AI 翻译累计 1.7 万部、畅销榜占 42%;一「全自动生成平台」宣称 48 小时产出 500 万字(单源 C)——写作、翻译、审校、有声化的边际成本正压向零。
Editors report two-to-three of every ten submissions carrying AI polish; daily debuts peaked at 3,549 after the cheap-model shock; AI audiobooks run 5M characters a day at −90% cost; AI translation totals 17k titles and 42% of the bestseller list; one «full-auto platform» claims 5M characters in 48 hours (single-source, C) — the marginal cost of writing, translating, proofing and voicing presses toward zero.
▸ 反弹这一侧(人格与守门)The revolt side
国内首例创作者集体反对 AI 训练条款(2024-07 平台上线「解除条款」入口;上百作者称解约后被限流,平台否认,无第三方裁定);最严平台推「AI 辅助六级分类」违规锁章禁榜;监管落地:AI 生成内容 2025-09-01 起强制显式+隐式标识;番茄 2026-06 单月拒签逾 10 万本——读者用「AI 味」投票,平台被迫当守门人。
The first collective revolt against AI-training clauses (an opt-out shipped in Jul 2024; a hundred authors allege post-opt-out throttling, the platform denies, no third-party ruling); the strictest platform runs a six-tier AI-assistance taxonomy with chapter locks and chart bans; regulation lands — explicit-plus-implicit labels mandatory from Sep 1, 2025; Fanqie rejected 100k+ books in June 2026 — readers vote with the «AI flavour» and platforms are conscripted as gatekeepers.
◆ 判词Verdict内容通胀,筛选升值,关系为王。AI 把内容边际成本压向零→内容通胀→价值从「生产」迁移到「筛选」与「关系」。任务化的部分(码字、初翻、初校)被拿走;人格化的部分(被追随、被信任、可背书)被反向凸显——这条定律贯穿网文轨与出版轨。Content inflates, curation appreciates, the relationship is king. AI presses marginal content cost to zero → inflation → value migrates from production to curation and relationships. The taskified parts (drafting, first-pass translation, first-pass proofing) are taken; the personal parts (being followed, trusted, vouched for) stand out in relief — one law across both tracks.