Between its re-introduction as X’s flagship approach to tackling disinformation by Elon Musk and Mark Zuckerberg’s announcement that Meta is adopting the same method in lieu of fact-checkers, ‘Community Notes’ appears to be a new alternative approach increasingly seen as the answer to “unbiased” fact-checking.
However, Community Notes is not the magic bullet it is being touted to be, and perhaps more importantly, it is not a one-size-fits-all solution to online disinformation, harassment and hate speech.
What are Community Notes?
Before it was expanded and rebranded to ‘Community Notes’ by Elon Musk in November 2022 after he acquired the platform, it was first introduced as Birdwatch in 2021 by Twitter’s former management.
It is a feature where individuals can provide context, such as fact-checks, below a picture, video, or message posted on X. Based on a crowdsourced method, it is a community-driven content moderation system designed to offer relevant and informative context. According to X, it “aims to create a better-informed world by empowering people on X to collaboratively add context to potentially misleading posts.”
Notes that appear beneath potentially misleading content are ranked by a bridging-based algorithm that doesn’t take majority rule into account but focuses on agreement from users on different sides of the political spectrum. The “notes” generated by users carry information about tweets flagged as potentially misleading. These annotations are made visible after the programme’s algorithm deems them “helpful” based on different factors, including positive ratings and bipartisan consensus. Conversely, an already-published note can disappear if its rank is reduced low enough.
The algorithm, according to X, is unique from other social media algorithms in that it tries to form consensus between groups with different perspectives as a way to avoid bias, pile-ons and abuse by including ratings from diverse contributors.
This approach to moderating online content drew both initial support and backlash from across political leanings and among experts, journalists and researchers.
Are They Any Good?
Community Notes have been regarded as a legitimate approach to content moderation, not the least because it is a crowd-sourcing system (like Wikipedia) which is more likely to engender trust among users, particularly because people feel better knowing the decisions going into labelling or removing tweets are based on actions of collective actions rather a top-down one. Even Elon Musk himself has been subject to Community Notes fact-checking as well as prominent US and UK politicians’ accounts and government accounts like the White House and FBI, among others.
Studies on Community Notes have also found the following:
- Based on a study of COVID-19-related notes in 2022-23, there was a 97% accuracy in the context provided by contributed (according to medical professionals), suggesting that Community Note fact checks can be high quality.
- According to a different study, adding annotations to erroneous tweets increases the likelihood that the original author would remove the post by 80% and decreases retweets by half. This is similar to what Twitter’s research found.
- Data made available by the Community Notes programme can be useful for researchers and fact-checkers. featuring the notes that were contributed, their rating system, and their history of status, such as whether they were deemed helpful and displayed to the public or useless and concealed, updated regularly. This can prove useful for fact-checkers who can search for claims there.
- Community Notes are very good at identifying false advertising as well as AI-generated images as well as low-stakes misleading tweets relating to pop culture, football, etc.
However, after several months of its full roll-out, several shortcomings backed by evidence have emerged highlighting its limits in combatting disinformation online. These shortcomings range from how slow the feature reacts to disinformation, its low visibility, its limited access, its vulnerability to manipulation, its enablement of disinformation spread, and its general inconsistency of operation.
Slow to Act
One of the main criticisms that Community Notes on X has faced is its markedly feet-dragging speed of dealing with countering disinformation on the platform. According to this Bloomberg report on misinformation during the Israel-Hamas conflict found that notes relating to the misleading tweets typically took more than seven hours to show up, with some taking up to 70 hours. One key trait of online disinformation is that it spreads quickly and can cause harm within a relatively short time, especially on X. However, because notes take time to be written and voted on, they react slowly more often than not and will likely usually show up after the fact.
Low Visibility and Access
Several analyses and demonstrated instances have also indicated that most users on X never see Community Notes correcting dis/misinformation. For example, this analysis by Alexios Mantzarlis and Alex Mahadevan found that only 29% of fact-checkable tweets in their sample carry a note rated ‘helpful’ and, of the helpful notes, only 67% were assigned to a fact-checkable tweet. This was corroborated by a WIRED report, which noted that in the backend of the database, most notes remain unpublished simply because they failed to receive enough upvotes.
Another investigation by Matt Binder found that misleading tweets receive way more views than fact-checks and that Community Notes are not being seen by most users. He demonstrates this using several verifiable examples of misleading tweets about critical subjects like the Israel-Palestine conflict, among others. Additionally, they found that in the three days preceding the US 2024 election, less than 6% of the approximately 15,000 notes attained helpful status.
Vulnerable to Manipulation
Several reports have also highlighted that despite the bridging-based algorithm and the safeguards put in place to eliminate the effects of bias and partisanship on the contributions of community members, there are indications that the tool is vulnerable to manipulation by malicious actors.
First, WIRED found that Community Notes contributors actively coordinate daily to upvote or downvote particular notes. They quoted an anonymous contributor saying: “We have a group so we can coordinate who writes what Community Notes, and if you write one, you can coordinate so you get one voted as helpful, and there’s no validation on any of that,” The contributor was also reported claiming that Russian embassies engaged in similar tactics and “will downvote anything that’s anti-Russian.”
Another contributor tells WIRED that he has two X accounts both of which have been approved for access to the Community Notes system – an indication that the system can be manipulated at an even larger scale by malicious actors. This is further highlighted by another study which points out the risks that clusters of bad-faith actors can skew the algorithm by attacking sources that are usually regarded as credible.
Spreads Disinformation
Ironically, for a tool whose primary function is countering disinformation on X, credible analyses are suggesting that Community Notes may be helping spread disinformation on the platform. These include false claims about Taylor Swift, bodyguard, one discrediting a legitimate tweet regarding the Israel-Palestine conflict and another one appended to a CNN tweet on the demography of Black fathers in the US.
According to Mahadevan, it was a racist note based on faulty, outdated data which added no substance to the CNN report and was seen by millions of people before it was eventually removed several days later.
The situation is worsened by the fact that many of the misinformation ‘super-spreaders’ on the platform have blue checkmarks, making them eligible for ad payout under Musk’s Ad Revenue Sharing scheme. Earlier in October 2023, Musk had said on X that posts labelled with corrections from Community Notes would be excluded from the ad revenue scheme. However, subsequent studies have revealed that ads continue to be shown underneath tweets that have received notes and that many misleading tweets from blue checkmark users remain unflagged by community notes.
Other Issues
There have also been several other issues raised by researchers, journalists and fact-checkers that are worth knowing about. These include:
- Community Notes is based on a narrow and outmoded understanding of disinformation where true-false, real-fake binaries are standard, the potential of content harmfulness is rarely considered, and a mechanism for addressing disinformation delivered through humour (e.g. satire) does not exist.
- According to Jonathan Warden, expertise is sometimes discounted in the bridging-based algorithm of Community Notes in order to break polarisation. He explains that the algorithm may identify and ignore a dimension other than political leaning if it is the principal source of polarisation in the user community (for example, between naive users and experts).
- Consensus is prioritised over facts. The algorithm relies on a cross-ideological agreement on a subject before notes go up. For example, similar numbers of contributors on the left and right have to agree before a note goes public on a tweet. In the words of Mahadevan: “That does not work anymore because 100 people on the left and 100 people on the right are not going to agree that vaccines are effective.”
What It Means for the West African Ecosystem
So far, Community Notes contributors exist in just over 70 countries, which includes the US, UK, EU countries, India, and a handful in Africa (Morocco, Tunisia, Algeria and South Africa). This is according to the announcements regularly posted on X’s Community Notes handle, which also suggests that there are still no contributors from the West and East African region, even as the dedicated content moderation team in Africa was fired when Musk took over the company in 2022. Meanwhile, online disinformation poses an increasingly significant threat to African countries, especially in relation to elections and democracy, but also concerning issues like insurgency and social cohesion.
More so, if Community Notes become accessible to West African contributors, it is unlikely to solve all of the disinformation challenges as has been the case with its roll-out in other countries. Therefore, journalists, fact-checkers, and ecosystem stakeholders need to continue promoting a multiple-pronged approach to combatting various forms of disinformation and online harassment, especially as other social media platforms such as Meta move to ditch third-party fact-checking in favour of X-styled Community Notes starting with the US.
As aptly stated by Yoel Roth, the former head of Twitter’s Trust and Safety, Community Notes “was never supposed to replace the curation team, but instead intended to complement it.” Stakeholders in the ecosystem must adopt the same attitude to Notes if we’re to continue mitigating the harms of digital disinformation on our societies.
