Meta与学术研究人员合作完成的一项研究发表在《科学进展》期刊上，声称用户在该公司旗下社交媒体平台上接触虚假信息的情况极为罕见[1]。这项研究涉及231百万Facebook账户和200百万Instagram账户[1]，显示Facebook用户平均接触不可信来源内容的比例仅为1.1%，Instagram则为0.1%[1]。研究指出，删除此类内容对用户的信念几乎无影响——在2020年选举前后三个月内的干预试验中，虽然约一半用户的虚假信息来源被删除、接触率下降约70%，但追踪的十项指标包括虚假声明信念、真伪辨别能力、极化和媒体信任均未出现显著变化[1]。
然而，这项研究的细节表明其结论的适用范围存在多项重要限制[1]。数据显示，虽然平均暴露率低，但53百万Facebook账户（占23%的用户）接收了不可信来源内容的80%，最高暴露的2.5%用户中，不可信来源甚至提供了约60%的政治和社会内容[1]。此外，研究所有30名作者中有12名是Meta现任或前任员工[1]。Meta已于2025年4月在美国取消了第三方事实核查[1]，而在COVID疫苗内容中，未被标记的误导信息的总影响是被事实核查者发现内容的46倍[1]。
A research collaboration between Meta and academic researchers, published in Science Advances, found that user exposure to content from unreliable sources on Facebook and Instagram is uncommon, with average rates of 1.1% on Facebook and 0.1% on Instagram.[1] The study analyzed 231 million Facebook accounts and 200 million Instagram accounts.[1] However, the distribution of such content was highly concentrated: 53 million Facebook accounts, representing 23% of users, received 80% of all content from untrustworthy sources.[1] Among the most exposed 2.5% of users, unreliable sources provided approximately 60% of political and social content.[1]
The research examined whether removing flagged content altered user beliefs by tracking a subset of approximately 16,000 consenting users over three months surrounding the 2020 election, during which half had their sources of unreliable content removed.[1] This removal resulted in roughly a 70% reduction in exposure to such material.[1] Across ten measured metrics—including false belief acceptance, ability to discern accuracy, political polarization, and media trust—no significant changes were detected.[1]
The study's authors included 12 current or former Meta employees among 30 total contributors, raising questions about potential bias in the research design and conclusions.[1] Meta discontinued its third-party fact-checking program in the United States in April 2025.[1] In the context of COVID vaccine misinformation specifically, the total reach of unmarked misleading content was 46 times greater than that of fact-checked material, highlighting gaps in the platform's content moderation approach.[1]
