What it actually takes to get a video taken down — by platform, in days.

Time-to-removal is the number rights holders feel in their gut. A video is up, it is earning someone else views, and every day it stays up is a day of value you do not capture. So the speed of a takedown is not a vanity metric. It is the difference between catching demand while it is hot and catching a corpse.

The problem is that almost nobody publishes their real numbers. Platforms describe their processes in policy pages and transparency reports, but the lived experience — days from a valid notice to a confirmed, verifiable removal — is rarely shown. We are publishing ours, with the caveats stated plainly, because the spread is the most useful thing a rights holder can learn before choosing where and how to file.

What we measure, and what we do not.

The metric is simple to define and hard to fake: days from the moment we file a complete notice to the moment the platform confirms the content is removed. Not “submitted.” Not “under review.” Confirmed gone, by the platform, in a way we can verify. That confirmation requirement is why our sample is smaller than our filing volume — we only count what we can prove.

The legal backbone here is the notice-and-takedown system in the Digital Millennium Copyright Act, specifically the safe-harbor provisions at 17 U.S.C. § 512. A platform that wants to keep its safe harbor must remove or disable access to identified material “expeditiously” upon receiving a compliant notice. The statute does not put a number on “expeditiously,” which is exactly why measured medians are worth publishing.

Filing carries obligations too. Section 512(c)(3) requires a statement of a good-faith belief that the use is not authorized, and the Ninth Circuit’s decision in Lenz v. Universal held that a sender must consider fair use before sending. Our standard process includes verifying ownership or authority and reviewing context before a notice goes out. We do not measure speed by cutting that step.

The data we have so far.

Over the most recent 90-day window, our confirmed-removal coverage is concentrated where we can verify outcomes end to end. In that window, the platform with confirmed tracking shows a median of 3.2 days from filing to confirmed removal, across 19 confirmed cases. That is a small sample, and we label it as such. With n below 30, the number is directional, not definitive.

The tight band around the median tells its own story. The 25th percentile sits at 3.2 days and the 75th at 3.3 days, so the bulk of confirmed cases cluster in a narrow range rather than scattering across weeks. For a rights holder, predictability has its own value: a process you can plan around beats an occasional fast result surrounded by uncertainty.

One absence matters. If a platform does not appear in our numbers, that means we do not yet have confirmation coverage there. It does not mean the platform fails to remove content. We are deliberate about this distinction because the alternative — presenting silence as a verdict — would be both wrong and unfair.

Reading speed against each platform's own commitment.

The only fair benchmark for a platform’s measured time is that same platform’s own published commitment. Comparing two platforms head to head invites the wrong conclusion, because scale, content type, queue design, and factors we cannot observe from the outside all move the number. So the right question is narrow: how does a platform’s measured median sit next to what it says about itself?

On that basis, a 3.2-day median reads comfortably inside the “expeditious” standard the statute requires and the responsive posture described in published platform copyright policy, such as the process documented in the YouTube Help Center copyright-removal pages. We make no claim about any platform’s intent. We are reporting a measured number against a published policy, and nothing more.

Figure 2 is the one most likely to be misread, so read it carefully. In the quarter beginning April 2026 we filed 161 notices and independently confirmed 19 removals inside the 90-day window. That is not a 12% removal rate. The gap is mostly cases still moving through review at the snapshot date and platforms where we do not yet capture confirmation. Confirmed removal is the strictest bar we can hold ourselves to, and the strictest bar produces the smallest count.

None of these figures forecast what will happen on your video. They are backward-looking, small, and bounded by where we currently have confirmation coverage. We publish them so you can plan with real distributions instead of marketing adjectives, not as a representation of any future outcome.

Why speed is an asset question, not a cleanup question.

Here is the implication most rights holders miss. A fast, predictable removal time is not just tidy housekeeping. It changes the economics of your catalog. When unauthorized use comes down in days rather than weeks, the cost of using your content without permission becomes predictable, and predictable cost is what makes licensing the rational choice for the people using your work. We have argued this at length in enforcement is asset-building: the goal is not to make a number go down, it is to convert proven demand into paying relationships.

Speed also closes a specific leak. Every day a video sits up unaddressed is value moving to someone else’s account, and that value rarely comes back. We mapped where it disappears in the video value leak. Time-to-removal is one of the few levers that acts directly on the leak rate. A 3-day median and a 30-day median are not the same business.

What to do with this.

First, demand measured numbers, not adjectives. Anyone handling your enforcement should be able to show days-to-confirmed-removal as a distribution, with sample sizes attached. If the only answer is “fast,” you have no way to plan and no way to audit.

Second, separate filing volume from confirmed outcomes. A big notice count says nothing about what actually came down. Ask specifically for confirmed removals and how confirmation is verified. The strict definition is harder to hit and far more honest.

Third, benchmark each platform against its own published policy, never against another platform. The fair question is whether a platform’s measured median sits inside what it has committed to publicly. Read the policy, read the U.S. Copyright Office’s Section 512 materials, and judge the number against the commitment.

We will keep publishing this data as our confirmation coverage widens. The numbers will move, the samples will grow, and some assumptions here will not survive contact with a larger dataset. That is the point of putting them in public: a number you can check is worth more than a promise you cannot.