This study assessed the conduct of social media platforms during the first two months of the 2023 Israel-Hamas War (October 7-December 7, 2023) by examining the methods and quality of their responses to reports from the Israel Internet Association’s Internet Safety Hotline about malicious, harmful or false content. The study performed a comparative analysis of the response times and characteristics, as well as the treatment in practice of the content reported by the Hotline, considered a Trusted Flagger by most of the major platforms and many online services.
Poor Handling of Harmful Content in War
The main conclusion that emerges from the findings is that the platforms failed to adequately deal with the flood of harmful content in the first two months of the war. In many cases, the platforms did not take proper enforcement actions, their response times to most reports were long and unreasonable in relation to the emergency situation and the severity of the content, and they did not provide a satisfactory response on weekends, even at the beginning of the war. The study’s findings show that the general functioning and response of the platforms during the period under review was deficient. The study found that:
Most platforms did not take proper enforcement actions (removal, flagging or sufficient explanation as to why enforcement action was not taken) in response to the Hotline reports, and in fact there are disturbing gaps between declared content policy and actual conduct. The findings indicate differences between response rates depending on the platform and the type of harmful content. Facebook stands out sorely, with 25% of the Hotline’s reports receiving inadequate handling from the platform (that is, only 75% of the reports received proper handling)–a much higher percentage than the other platforms, whose rate of inadequately handled reports hovers around 11-13% (that is, 87-89% received proper handling).
The response time of the major platforms for Internet Safety Hotline reports in the first two months of the war was about 5 days on average. This is an unacceptably long time even in routine scenarios, let alone during a sensitive and volatile emergency situation like the ongoing war. Reports sent in the context of war, some of which support terrorism and incitement to violence or propagate false information, require a swift response. The platform with the slowest response in the first two months of the war was Facebook, which had an average response time of about 7 days–a long and unreasonable time for users exposed to harmful, alarming and anxiety-inducing content during a period of war and emergency, some of whom are currently enduring life-threatening circumstances.
In general, almost no response was received from the platforms during weekends (Saturday-Sunday). The lack of enforcement actions is particularly evident when contrasted with the constant flow of harmful content on the platforms, seven days a week, both in regular times and certainly during wartime. It seems that the scant resources allocated by the platforms to weekend content moderation are incompatible with the need that surfaced, contradicting their public statements about special measures being put into place to address problematic war-specific content.
TikTok stood out as the only platform that responded to inquiries and reports during the weekends (Saturday-Sunday), and even tended to respond quickly. However, it should be noted that when assessing the general pool of reports to TikTok over the course of the week, many reports remained unanswered as of the finalizing of the present data, which weakens the average response time and overall handling of inquiries demonstrated by the platform.
X (Twitter) is a conspicuously absent from the research data–during the months examined, many reports were forwarded to it by the Hotline, but the reporting mechanism for recognized reporting entities has been deactivated, and the platform’s automatic and inadequate responses do not allow the tracking of response times nor do they allow the assessment of the nature of the platform’s response.
Poor Handling of False Information and Content Requiring Human Intervention
The evaluation of the platforms’ responsiveness and their methods of handling reports showed consistent and evident differences between the handling of different types of content. Reports that can be handled simply using automatic and computerized scanning–such as graphic content and terrorism–were handled relatively quickly and efficiently. On the other hand, content that requires examination by trained personnel and human decision-making, based on local understanding and context–for example, false or misleading information and hate speech and incitement–suffer from insufficient and slow enforcement.
A comparative examination of the response provided to reports across the various types of harmful content shows that the content category that received the lowest level of response was false or misleading information. According to the summary of the data, overall the platforms failed to properly handle 42% of the reports on false information. The percentages of poor handling were also relatively high (12-19%) for reports of impersonation and invasion of privacy, hate speech, and incitement and calls for violence. In contrast, almost all reports of graphic content and terrorism were handled properly (95-99%).
An in-depth examination of the content reported by the Hotline as false information shows that in practice the Meta Company’s platforms failed to properly handle 57% of the reports on Facebook and 36% of the reports on Instagram. In contrast, TikTok failed to properly handle only 10% of reports of false and misleading content.
The reporting of harmful content can span multiple content categories at the same time. When reports of false information only were examined, omitting additional categories (such as hate speech or incitement), no substantial difference was observed between Instagram and TikTok, but on Facebook the percentage of reports where no satisfactory enforcement actions were taken climbed to 72%.
The findings indicate that Meta does not take sufficient enforcement actions regarding false information published on its platforms, both in terms of the manner of response and the time taken to respond. Meta has stated that it relies on independent fact-checkers, who check and rate the degree of accuracy of content,[1] and forwards to them, among other things, inquiries and reports from Trusted Flaggers. The fact-checkers are independent and Meta is unable to coerce them into checking any given content. However, the findings of this study raise concerns about Meta’s decision to transfer responsibility to external fact-checkers when it comes to the Israeli arena. Of all the 57 Hotline reports forwarded to Meta regarding false information, only one post was tagged with a comment by these fact-checkers. The intention here is not to criticize the external fact-checkers, who are held to strict journalistic standards and lack the authority to address conspiracies or false information framed as opinions or questions. Meta, however, does possess this authority and can intervene if the content is deemed to pose real-world harm to users, yet the company fails to take sufficient action. The method by which Meta handles disinformation confirms the study’s conclusions regarding Facebook’s particularly long response times to reports of disinformation. As mentioned, Facebook’s average response time to reports of false information was over 11 days, compared to about 3 days for TikTok and about 6 days for Instagram. It seems that this can be attributed to the “diffusion of responsibility” effect created by Meta’s policy of transferring Trusted Flagger reports of misinformation to external fact-checkers who act as an independent third party.
This fact emphasizes the need to invest in the hiring of moderators adapted to the region and language, and highlights the need for more transparency as to the resources allocated by platforms to local safety teams and their degree of availability.
Lack of Transparency in Platform Moderation
Alongside the enforcement flaws found in all the platforms reviewed, the platforms also fail to provide sufficient information and transparency regarding steps taken with relation to content reported by official and recognized Trusted Flaggers. The responses to Trusted Flaggers are generic, templated and fixed, do not refer explicitly to the content or subject of the report, and in cases of removal or flagging do not provide substantial information, but often suffice with the laconic wording “we have taken appropriate action.” Sometimes the platforms simply declare that they have removed content or, in other cases, claim that they did not remove the content because it and the account that posted it have not violated any of the community rules.
A comparison of the response times between the platforms revealed that on YouTube response times were relatively long, but this platform should nevertheless be noted favorably for making a considerable effort to create transparency, detailing the steps it has taken and replacing the removed content with a statement about what happened to it and why. YouTube’s reply to the reporting party also refers explicitly to the content of the report and specifies the policy section that justified the removal, while personally addressing the reporting party, and in most cases the message also includes the personal signature of one of the platform’s employees. This kind of detailed response greatly contributes to the ability of Trusted Flaggers to follow up on reports, make sure they are handled properly and keep the user community updated. Furthermore, a detailed explanation clarifying the decision and the steps taken can stimulate dialogue, allow a more intimate understanding of the local context if necessary, and improve the quality and accuracy of future reports submitted by Trusted Flaggers.
An examination of the links forwarded to the platforms in the original reports reveals what a removed content item or account looks like post-removal. Here too, the platforms do not practice transparency – most of them do not indicate that an enforcement action was taken against the content, let alone the relevant policy section. This lack of transparency makes it difficult for Trusted Flaggers to follow the treatment of their inquiries, to know what happened to their reports and improve their future reports. Similarly, users harmed by content are not given the benefit of knowing how their appeal has been handled and what the fate of the reported content was.
A satisfactory and prompt response is made possible with the help of proper resources and consistent enforcement of policy
At the outbreak of the war, the platforms rushed to announce their extensive efforts to tackle harmful content and pledged their commitment to improving its removal. The research findings show that platforms are, in fact, capable of promptly responding (within 24-72 hours) in emergency scenarios like times of war or regional conflict, provided there is a financial or reputational incentive to do so. Indeed, analysis of the platforms’ response times during the examined period reveals that in the first week of the conflict, they managed to respond to the Hotline’s reports within 24-72 hours, mirroring Meta’s similar average response time to reports from recognized reporting bodies during the Russia-Ukraine war in 2022. These data underscore the attainability of swift and satisfactory responses to official and well-known Trusted Flaggers in times of crisis, provided the platforms devote the proper resources and manpower to achieving this goal. This was made clear by the Russia-Ukraine war, and by the platforms’ actions at the onset of the Israel-Hamas war when they announced the opening of special operation and command rooms and removed hundreds of thousands of content items. However, in most of the subsequent weeks since the outbreak of the war and through December 7, 2023, the overall handling of reports lapsed into a state of inadequacy.
In general, the findings suggest that despite the challenges and threats confronting the platforms following the October 7 massacre and the subsequent war, they neglected to allocate the required resources for a satisfactory and comprehensive response to policy violations and for the protection of user safety during this emergency. In contrast with their promises, most of the platforms provided only a partial, slow and inconsistent response, even when the reports were labeled urgent and verified by the Internet Safety Hotline, a reliable and professional reporting entity recognized by the platforms. These are particularly alarming figures when one considers how the platforms must have treated reports submitted by the general public.
The obvious general conclusion is that this manner of conduct is a conscious choice and a product of policy, and not simply an inevitable product of circumstance. In the absence of adequate policies and poor allocation of resources, the response times, days of response, and response quality are insufficient.
[1] Meta’s third-party fact-checking program | Meta Journalism Project. https://www.facebook.com/formedia/mjp/programs/third-party-fact-checking