It is the never-ending battle for YouTube.
Every minute, YouTube is bombarded with videos that run afoul of its many guidelines, whether pornography or copyrighted material or violent extremism or dangerous misinformation. The company has refined its artificially intelligent computer systems in recent years to prevent most of these so-called violative videos from being uploaded to the site, but continues to come under scrutiny for its failure to curb the spread of dangerous content.
In an effort to demonstrate its effectiveness in finding and removing rule-breaking videos, YouTube on Tuesday disclosed a new metric: the Violative View Rate. It is the percentage of total views on YouTube that come from videos that do not meet its guidelines before the videos are removed.
In a blog post, YouTube said violative videos had accounted for 0.16 percent to 0.18 percent of all views on the platform in the fourth quarter of 2020. Or, put another way, out of every 10,000 views on YouTube, 16 to 18 were for content that broke YouTube’s rules and was eventually removed.
“We’ve made a ton of progress, and it’s a very, very low number, but of course we want it to be lower,” said Jennifer O’Connor, a director at YouTube’s trust and safety team.
The company said its violative view rate had improved from three years earlier: 0.63 percent to 0.72 percent in the fourth quarter of 2017.
YouTube said it was not disclosing the total number of times that problematic videos had been watched before they were removed. That reluctance highlights the challenges facing platforms, like YouTube and Facebook, that rely on user-generated content. Even if YouTube makes progress in catching and removing banned content — computers detect 94 percent of problematic videos before they are even viewed, the company said — total views remain an eye-popping figure because the platform is so big.
YouTube decided to disclose a percentage instead of a total number because it helps contextualize how meaningful the problematic content is to the overall platform, Ms. O’Connor said.
YouTube released the metric, which the company has tracked for years and expects to fluctuate over time, as part of a quarterly report that outlines how it is enforcing its guidelines. In the report, YouTube did offer totals for the number of objectionable videos (83 million) and comments (seven billion) that it had removed since 2018.
While YouTube points to such reports as a form of accountability, the underlying data is based on YouTube’s own rulings for which videos violate its guidelines. If YouTube finds fewer videos to be violative — and therefore removes fewer of them — the percentage of violative video views may decrease. And none of the data is subject to an independent audit, although the company did not rule that out in the future.
“We’re starting by simply publishing these numbers, and we make a lot of data available,” Ms. O’Connor said. “But I wouldn’t take that off the table just yet.”
YouTube also said it was counting views liberally. For example, a view counts even if the user stopped watching before reaching the objectionable part of the video, the company said.