Abstract
Accused by the video behemoth of spreading misinformation,
CREDIT: t0ad
We wanted to let you know our team reviewed your content, and we think it violates our medical misinformation policy.
The email said that, although my channel wouldn’t be removed, one of the videos I had uploaded had been taken down. Still, there were words of comfort:
We realise this may be disappointing news, but it’s our job to make sure that YouTube is a safe place for all. If you think we’ve made a mistake, you can appeal this decision.
I did appeal; immediately and in the most emphatic terms. A few hours later I received notice that my appeal had been turned down and that “we won’t be putting your content back up on YouTube”. No further appeals allowed.
So what misinformation had I spread? You can see for yourself - the video is also hosted on Vimeo and remains online at
I had uploaded it to YouTube and it never attracted more than a few hundred views during the three years it was up.
The book dissects a range of “denialisms” (including anti-vaxxing), assesses the future of the phenomenon and speculates on what it would take to make denialism history. As the rapid growth in medical misinformation during the Covid pandemic has shown, the book didn’t stop this phenomenon.
Still, YouTube wasn’t making a judgment on the effectiveness of my arguments but on the content of the video. So what did it find so objectionable?
I don’t know for sure: the email contained no specific information on what the problem was or how the assessment was made. There are a couple of things that might have raised some flags. One was that the voiceover was compiled from a cacophony of voices of denialists themselves - including ones disparaging accepted scholarship on vaccines, Aids, the Holocaust, climate change. The other is that the blurb was taken from the publisher’s website, which begins:
The Holocaust never happened. The planet isn’t warming. Vaccines cause autism. There is no such thing as Aids. The Earth is flat.
So it seems that quotation of medical misinformation - in the second instance, ironic quotation - triggered YouTube’s disciplinary process. Whether it was quotation in the video itself, in the blurb or both remains unclear.
I also don’t know who or what made this judgment. Was it one of those fabled algorithms? Or a (probably underpaid and undertrained) worker making dozens of such decisions every day? Or was the judgment made by a member of the public who reported the video (maybe maliciously)? And was the appeal handled by a human or an algorithm?
Either way, the thing or person making the decisions could not distinguish between quotation/irony and serious intent.
This is a bizarre kind of Turing Test: is YouTube’s incompetence down to an incompetent human or an incompetent algorithm?
The information on the YouTube website does not help clarify matters. The page titled “How does YouTube manage harmful content?” states:
We remove content that violates our policies as quickly as possible, using a combination of people and machine learning to detect potentially problematic content on a massive scale. In addition, we rely on the YouTube community as well as experts in our Trusted Flagger programme to help us spot potentially problematic content by reporting it directly to us. We also go to great lengths to make sure that content that violates our policies isn’t widely viewed, or even viewed at all, before it’s removed. Our automated flagging systems help us detect and review content even before it’s seen by our community. Once such content is identified, human content reviewers evaluate whether it violates our policies. If it does, we remove the content and use it to train our machines for better coverage in the future. Our content reviewers also protect content that has a clear educational, documentary, scientific or artistic [EDSA] purpose.
Such generalities do not help in isolating where the process broke down here.
I am not the first person to be mystified by the capricious application of YouTube’s takedown policy (or, indeed, those of other social networks). In January 2021, talkRADIO’s channel was taken down for “violating YouTube’s community guidelines”. Although YouTube confirmed that some content appeared to violate Covid-19 guidance, the entire channel was restored within a day. At the end of October, the entire channel of the leftwing news website Novara Media was taken down for about two hours. On that occasion, it appeared that some of the channel’s content had been flagged by members of the public (who can report violations of YouTube guidelines), and on review it was reinstated.
In order to try to penetrate the mystery of YouTube’s decision-making process, I decided to conduct an experiment. While I might not be able to discover who or what made the call in my case, I could at least identify whether it was the video or the blurb or both that brought down its wrath.
I created a new YouTube account in a false name and recorded two very short videos. One of them featured a statement that vaccines, including the Covid vaccine, were dangerous and no one should take them. The other featured a statement proclaiming the exact opposite.
I titled them “Vaccines are great! Everyone should take the Covid vaccine!” and “Vaccines are dangerous! No one should take the Covid vaccine!”
The twist was that I applied the title to the wrong videos - so the anti-vax video had a pro-vaccine title and vice versa. That way, if one were taken down and the other stayed online, I’d be able to tell what aspect of the content infringed YouTube’s policies.
Within half a day of posting, one of them had indeed been taken down: “Vaccines are great! Everyone should take the Covid vaccine!” The one with the anti-vax title, though, remains online at the time of writing, nearly a week after posting.
You can view the videos (now safely, for the moment, archived on Vimeo at
Actually, my experiment may have made YouTube’s policies and processes less clear. And that’s the point: they constitute a “black box” to which we have no access and, even if the processes are rigorous and consistent, they seem capricious and incoherent. As in all cases of judgment and disciplining, only full transparency can reassure those being judged that the system is fair.
In the interest of encouraging that transparency, I contacted YouTube’s press office to explain what I had done and share an early draft of this article.
At the time of writing, there has been no response. While in high-profile cases such as Novara Media and talkRADIO, YouTube have issued statements (albeit not very helpful ones), for minnows like me it is almost impossible to engage the company in any kind of dialogue.
I have a lot of sympathy for these companies in that I can appreciate how hard it is to monitor the vast amount of information posted online every second. And it’s likely that quotation, irony and sarcasm constitute a massive challenge in content monitoring.
Even though companies such as Google, which owns YouTube, are almost inconceivably wealthy, the employment and training of sufficient numbers of staff capable of executing fine and nuanced judgments in a consistent way might burn through even their vast profits.
What I am not sympathetic to is the persistent denial by social media companies about who they actually are. Whether they want to be or not, they are publishers and ultimately responsible for the content they produce. Whether they want to be or not, some of them have become de facto public utilities, from whom exclusion is such a weighty matter that it requires the same accountability and transparency as exclusion from any other public good demands.
“Kafkaesque” is a much misused term, but applying it to social media companies seems justified.
In The Trial, Kafka described the disorientation produced by a justice system whose inner workings were mysterious. Treating YouTube et al as if they were the organs of a state might force them to acknowledge their power and to accept the responsibilities that go with it.