This piece examines one question only: after content is reuploaded, who handles it and what exactly are they handling. Figures were traced to primary sources wherever possible — platform transparency reports and rule text, judicial interpretations and judgments, regulator announcements, and original papers. Every conclusion carries one of three grades: [Solid] = traced word-for-word to a primary source and not materially shaken; [Narrowed] = the direction holds, but the circulating version was found to contain a specific error, and what appears here is the corrected form; [Judgment] = mechanically coherent but without direct empirical support. How the money flows through this chain, and what AI changed, are the other two halves — separate pieces.
1. What happens after they catch it
[Solid] YouTube processed 2,502,941,368 Content ID claims in 2025, up 14% year over year. Rightsholders chose to monetize over 90% of them rather than block. The video stays on the site; the ad revenue is redirected to the rightsholder. As of December 2024, Content ID had paid out more than $12 billion.
So the real cost of being caught is that this video stops earning. The channel is still there. Nobody pays damages.
It is a revenue-rerouting pipe. The platform loses no traffic, the rightsholder receives money it would not otherwise have had, and only the reuploader's expected return is squeezed toward zero. All three parties can live with that, which is why the arrangement is stable.
Almost nobody walks the dispute path. Of those 2.5 billion claims, only 12,840,608 were disputed by uploaders — 0.51% — and of those who did, 67.42% won, with a 75% success rate on further appeal. A system where objecting gives you better than even odds sees objections in half of one percent of cases. The likelier explanation is not that the claims are all correct; it is that the ladder is long, walking to the end risks a copyright strike, and most people absorb it instead.
"YouTube has the strongest copyright detection in the world" is a true statement. The question is who gets to use it. [Solid] Only 7,626 rightsholders worldwide hold Content ID access, and 4,454 actually used it. The official eligibility page adds that mashups, compilations, remixes, gameplay footage, and non-exclusively licensed material generally cannot serve as reference files — precisely the highest-volume category for content farms, both as raw material and as finished product. The substitute tool offered to ordinary creators looks only for full or near-full reuploads, may not surface a short clip, and scans only videos uploaded after yours.
The strongest detection existing, and you having recourse, are two different things.
The platform has explained why the gate is there. More than 5% of removal requests submitted through the public webform were assessed by its review team as abusive — likely false assertions of copyright ownership. For tools with eligibility requirements, the abuse rate is 0.3% or lower, a difference of more than twentyfold. Among the abuser types it lists is the Backdater, who pretends to have published first in order to take down the actual author. The platform's response was to restrict who may use the tools, not to make detection more accurate.
[Narrowed] On audio fingerprinting under real conditions, the figure usually cited is the ISMIR 2022 BAF dataset result that "no algorithm achieves an F1 above 47%," with per-second recall as low as 0.02 to 0.32. Those were 2022-era open-source systems used out of the box. Since 2024, systems retrained on broadcast data and preceded by music-speech separation reach F1 0.76 to 0.91 on the same benchmark.
What remains unsolved is text. Put an 11B-parameter paraphrasing model in front of the detectors and DetectGPT's detection rate at a fixed 1% false-positive rate falls from 70.3% to 4.6%; watermarking and the various classifiers are evaded too, with semantics essentially unchanged. The same paper supplies the counterexample: semantic retrieval against a corpus of the originals still finds 80% to 97%. This explains every platform's technical choice. Effective originality detection compares against content that already exists; it does not ask whether a passage reads like it was written by a machine.
2. Seven years on, the platforms reversed themselves
In October 2019 a Facebook internal report described the mechanism of the day very plainly: the platform did not penalize entirely unoriginal content, and something that had gone viral once would probably go viral again. The report was published in full by the press in 2021. Its figures: as much as 40% of US Page views went to pages featuring primarily unoriginal content, and Balkan content-farm pages reached 140 million Americans a month, 75% of whom had never followed any page.
Over the following years Google moved first. In March 2024 it shipped three new spam policies at once, of which "scaled content abuse" is officially defined as generating many pages primarily to manipulate rankings rather than help users — no matter whether the content is produced by automation, by humans, or by some combination. The test moved from how it was made to whether it adds anything.
The real turn came in July 2025, when two platforms moved in the same month. On 14 July, Meta announced it would remove monetization and recommendation from accounts that repeatedly reuse others' work, disclosing that it had acted on roughly 500,000 accounts in the first half of 2025 and removed about 10 million profiles impersonating creators. The next day, YouTube renamed its "repetitious content" policy to "inauthentic content," stating that content made with generic or unoriginal templates, or giving the impression of mass production, is ineligible for monetization — and the action lands on the whole channel, not a single video.
By 13 March 2026, Meta had written originality directly into ranking, with a number attached: views and watch time on original Reels roughly doubled in the second half of 2025 versus the same period in 2024. Its definition of unoriginal is more concrete than anyone else's — watching along, reacting with an emoji, stitching several clips together, reading out text already on screen: none of it counts, and all of it gets demoted in Feed and Reels.
Same company, reversed seven years later.
What this shift really changed is who initiates. A copyright complaint requires the rightsholder to discover the infringement and make the case; monetization review is executed unilaterally by the platform, and nobody whose work was taken needs to appear at any point.
Cross-language reuploading is the clearest case. The original author is typically in another language market, behind another wall, and was never going to appear in the process. Now the platform itself initiates, so enforcement runs faster than it used to. The cost is that the standard is defined unilaterally, there is no independent appeal, and collateral damage can only be worked through the platform's internal process.
One boundary is frequently misread. [Solid] What YouTube changed was "templated mass production," not "reuploading." The company clarified that reused content with significant original commentary, substantive modification, or educational and entertainment value remains monetizable; what was cut off is series of story videos differing only in minor details, slideshows with identical narration, and verbatim copies with no substantive change. Reporting this as "YouTube bans reuploads" yields the wrong prediction about the industry.
3. Damages differ by three orders of magnitude
In the Yunnan Chong Gu case, Douyin's operator was ordered to pay Tencent 32.4 million yuan. In a rewriting case before the Shenzhen Intermediate Court, the award was 800 yuan. Both are reuploading.
| Defendant | Typical award | Basis |
|---|---|---|
| Platform | 32.4M yuan (Yunnan Chong Gu) | Contributory infringement, 2M yuan × 16 episodes + costs |
| Platform | 30M yuan (Kuang Biao) | Infringer's profits exceeded the statutory ceiling; discretionary damages applied |
| Platform | 60M / 29.1M yuan | Guangdong and Chongqing High Courts, final, including punitive damages |
| Individual | median 3,000 yuan | 175 Yangtze Delta commercial-enforcement judgments; mean 17,951 yuan, 19.5% of claims supported |
| Individual | 800 yuan | A rewriting case before the Shenzhen Intermediate Court |
[Solid] Civil litigation barely deters an individual reuploader. Of the short-video copyright cases the Beijing Internet Court concluded between September 2018 and February 2022, 2,026 in total, only 233 produced a judgment; 1,793 were settled or withdrawn — 88.5%. The Yangtze Delta sample contains an even more telling structure: 85.2% of plaintiffs were companies and over 90% of defendants were too. Enforcement in China is in practice a business-to-business activity; individual creators are neither the main plaintiffs nor the main defendants.
The real constraints are two others, and neither is civil damages.
[Solid] The first is criminal. Judicial Interpretation No. 5 of 2025, effective 26 April 2025, treats fees collected through advertising and similar means as unlawful gains, with 30,000 yuan constituting a "relatively large amount." The heavier provision concerns the burden of proof: where the works are numerous and the rightsholders dispersed, if there is evidence of unlawful dissemination and the defendant cannot produce evidence of a licence, the court may find the use unauthorized — shifting the burden of proving authorization work-by-work onto the defendant.
[Solid] The second is administrative. Twenty years of the Jianwang campaign produced 11,545 investigated online piracy cases and 1,214 referrals for criminal prosecution; a single Cyberspace Administration campaign disposes of accounts by the million. The administrative line runs three to four orders of magnitude above the judicial one. Being banned is far likelier than being sued.
Fair use, as a fallback, essentially does not exist in China. [Solid] Article 24 of the Copyright Law is a closed enumeration — twelve specific situations plus a catch-all that no administrative regulation has ever filled — and "transformative use" is a US-law concept that Chinese courts rarely apply. Across 47 short-video fair-use judgments in the Wolters Kluwer database, exactly 1 (2.1%) found fair use.
The United States built a low-cost channel specifically for individuals, and the numbers there are no better. [Solid] The Copyright Claims Board opened in June 2022; through March 2026 it received 1,920 filings and issued 47 final determinations (2.4%), with 41% dismissed at compliance review. An empirical study did the arithmetic for the first two years: contested determinations awarded claimants $4,000 and default determinations $54,150, for a total of $58,150 — against a requested operating budget of $2.2 million a year, $4.4 million over the same period.
4. What you actually have left
On reuploading, the original creator is the only party with no means of enforcement. Detection is tiered by identity, and the strongest tier requires you to be an institution holding exclusive rights. Disposition sits with the platform, on a standard it defines alone. Median judicial damages are too low to cover the cost of pursuing them. And the channels that genuinely bite — criminal and administrative — are initiated by prosecutors and regulators, not by you.
What remains is limited but not nothing. Three actions, in order of return:
- Spend the time on platform complaints, not litigation. The two channels differ by six orders of magnitude in throughput. Google Search has received roughly 17.9 billion URLs in copyright removal requests since 2011 and acts on about 97.5% of DMCA notices; over four years the US Copyright Claims Board issued 47 determinations. Litigation's realistic function is to add leverage to a complaint, not to convert into money.
- Keep the original materials rather than buying a timestamping service. Article 16 of the Supreme People's Court Online Litigation Rules presumes only that blockchain-stored evidence was not altered after being written to the chain; Article 18 states plainly that where a party argues the data lacked authenticity before it was written, the court shall examine that. Notarization-grade evidence lowers your cost of collection, not your odds of winning. What helps is project files, camera originals, and draft versions — the things that speak to what happened before the chain.
- Know which tier you cannot reach. Content ID requires exclusive rights to a substantial body of frequently uploaded original material. The substitute tool explicitly does not scan three situations, one of which is "the video already has a Content ID claim" — the other party asserted ownership first, which is the most common shape reuploading takes. Rather than retrying that path, put the effort into the one you can verify: find it, report it, keep the proof.
One last observation, easy to miss. [Judgment] This arrangement is stable because it works for the platform, for rightsholding institutions, and for the advertising system — and fails only for individuals, whose losses appear in no transparency report and constitute no one's cost. As long as that loss stays off the books, no party has any reason to fix it.
Primary sources: YouTube's copyright transparency reporting and TorrentFreak's compilation of the 2025 figures; YouTube's official Content ID eligibility page, Copyright Match Tool documentation, and channel monetization policies (including the 2025-07-15 update); Meta Newsroom, "Cracking Down on Spammy Content on Facebook" (April 2025) and "Rewarding Original Creators on Facebook" (March 2026); the Facebook 2019 internal report published in full by MIT Technology Review; Google Search Central's three March 2024 spam policies and accompanying FAQ; the ISMIR 2022 BAF dataset paper together with the later benchmarks in ETRI Journal (2024) and Multimedia Tools and Applications (2025); Krishna et al., "Paraphrasing evades detectors of AI-generated text" (NeurIPS 2023); the full text of Judicial Interpretation No. 5 of 2025 of the Supreme People's Court and Supreme People's Procuratorate; Articles 24 and 54 of the Copyright Law of the PRC (2020 revision); the Beijing Internet Court's report on short-video copyright cases; Wu Yueting, "Commercial Copyright Enforcement in the Yangtze River Delta"; Articles 16 to 19 of the Supreme People's Court Online Litigation Rules; twenty-year figures from the National Copyright Administration's Jianwang campaign; and official Copyright Claims Board statistics (June 2022 to March 2026) alongside "Assessing the CCB After Two Years." Every load-bearing conclusion was independently re-verified; what appears here is the corrected version, with grades and staleness risk marked line by line.