Imagine trying to vote in a presidential election while false videos, manipulated images, and outright lies flood your social media feeds—and the companies that run those platforms do almost nothing when asked to remove them. That, in essence, is what happened in Ireland’s 2025 presidential election. A new report from the European Digital Media Observatory (EDMO) Ireland paints a troubling picture of an information environment left to fend for itself. Over the course of the campaign, EDMO researchers identified roughly 177 posts or pieces of activity that appeared to breach social media companies’ own rules. The categories read like a catalogue of modern digital harms: 85 cases of election-related disinformation, 26 examples of AI-generated election content, 13 incidents of harassment and abuse, two cases of incitement to violence, 49 examples of illegal election-day content, and two issues relating to political advertising. The most damning finding is this: out of those 85 disinformation cases, social media companies acted on just 12. That is only about one in seven. Voters were left to sort truth from lies on their own, and the platforms responsible for hosting the content largely stood by. These numbers may seem small in absolute terms, but elections are won and lost on narrow margins; a single viral lie can alter the atmosphere of an entire campaign. The report suggests that democracy was not protected by the very companies that now serve as the public square.
These statistics represent something deeper than a compliance failure. They reflect a refusal to take responsibility. The report notes that even when posts were brought to platforms’ attention as potential violations of their own terms of service, companies were repeatedly unable to explain why they applied rules so inconsistently. EDMO participated in a “Rapid Response System” in which platforms agreed to respond quickly to flagged content and to provide a rationale for their moderation decisions. Yet the social media companies rarely took the advice on board. One particularly telling example involved a fabricated quote attributed to a High Court judge, questioning the constitutional validity of the presidential nomination process. The story originated on Facebook. EDMO and The Journal, a digital news site partnered with Meta on fact-checking, flagged six variants of the claim to Meta. The Journal even provided a fact-check demonstrating the quote was false. Meta repeatedly declined to take action, insisting the posts did not violate its terms, but offering no further explanation. The report dryly noted: “Meta does not seem to have applied The Journal’s factchecks to obvious cases of disinformation.” This is like employing a firefighter and then ignoring the alarm while your house burns. It is not just about one false story; it is about a structural failure. When platforms own the infrastructure of public debate, their decisions—and their inaction—carry civic weight. The damage is not just to truth but to citizens’ ability to make informed choices, and that damage is hardest on ordinary people who do not have the time or resources to fact-check every alarming post themselves.
Some of the flagged content was genuinely sinister. A week before the presidential election, deepfakes of then-presidential nominee Catherine Connolly announcing her withdrawal from the race spread widely. Imagine seeing a video of a candidate quitting—except she never said it. In an election already short on time, such a lie could have changed votes or caused needless chaos. EDMO also flagged photographs of ballot papers posted on election day, many with the faces of the three candidates defaced. In Ireland, taking and sharing photos of marked ballots is illegal; such images can easily be used to intimidate voters, facilitate vote-buying, or cast doubt on the secrecy of the ballot. The report also highlighted a video of former Taoiseach Leo Varadkar being harassed in Dublin, a reminder that online abuse is not virtual but physical, and that public figures can be endangered by a culture of dehumanisation that platforms allow to fester. There were AI-generated depictions of candidates, which blur the line between reality and fabrication, and false claims about the mechanism of the presidential nomination process—claims that might make citizens wonder whether the entire election was legitimate. These are not abstract policy problems. They are real moments of manipulation and fear. The people targeted are not just political figures; they are human beings whose reputations, safety, and right to stand for office were hung out to dry on platforms that refused to intervene. When a voter opens a feed and sees a fake video, a defaced ballot, or a harassment clip, the psychological effect is not neutral. It creates suspicion, cynicism, and a sense that the system is corrupt. The report’s findings show that social media companies were not absent—they were present but passive, and passivity in the face of manipulation is a choice with consequences.
The EDMO report also exposes the failure of voluntary self-regulation. Under the EU Digital Services Act, platforms are encouraged to direct users to official election information, demonetise disinformation and reduce its visibility, tackle fake accounts, and help people evaluate what they see. These are not radical demands; they are basic public-interest obligations. Yet the report found that signposting to authoritative sources, corrective information, and other proactive information integrity efforts were poor across all platforms. The Rapid Response System, which was supposed to be a safety net, became another piece of paper. X had already pulled out of the commitment, and among those who remained, engagement was patchy. Participating platforms promised to file an expeditious response to issues flagged by EDMO and to provide the reasoning behind their moderation decisions, but they rarely delivered. The report observed: “While items were reported as potential breaches of platforms’ own terms of service, platforms were repeatedly unable to explain inconsistencies in the application of their own rules.” Facebook and TikTok emerged as the most frequently cited offenders: Facebook had 54 flagged cases but acted on only 44% of them, and TikTok had 50 cases but acted on only 54%. LinkedIn, meanwhile, had seven cases flagged and did not act on one of them. The inconsistency was not accidental; it was systemic. If rules exist, they must apply to everyone and be enforced consistently. Otherwise, they are little more than public-relations gestures. The platforms seemed to treat their guidelines as optional guidelines rather than binding commitments, and the election—and the public—paid the cost.
Platforms responded with defensive statements, but their answers only deepen the concern. TikTok said that 97.5% of violating content in the lead-up to the election was removed before it was first reported, and that it continues to work with trusted external partners such as fact-checkers. YouTube said it reviewed all reports received and removed everything that violated its policies. LinkedIn said it received three reports containing a total of nine pieces of content, removed one, and judged the other eight to comply with company policies. Meta did not directly respond to the Irish Examiner when contacted for comment. These claims are difficult to reconcile with EDMO’s evidence. If TikTok was so efficient, why were 50 cases flagged during the election, with only about half acted upon? If YouTube reviewed everything, why did EDMO still see an ocean of disinformation? The gap between corporate reassurances and on-the-ground reality is precisely the problem. Voluntary efforts may work for obvious spam or illegal content, but they break down when dealing with subtler disinformation, especially when it is dressed up as political opinion. Platforms also claimed to rely on fact-checkers, yet in the case of the fabricated High Court judge quote, fact-checks were handed to Meta on a silver platter and ignored. This is not a technical limitation; it is an institutional failure. The report calls out the “poor” proactive efforts across all platforms, and one gets the sense that researchers were not just disappointed but exhausted by the stonewalling. In any other area of public safety—from car safety to food hygiene—such a record would be considered unacceptable. Why should the information ecosystem be any different?
This report is not just an Irish story. It is a warning to every democracy. If the same dynamics play out in a relatively small presidential election, imagine the scale of manipulation in a crowded national campaign, a referendum, or a continental election. The report’s authors are not calling for the abolition of social media; they are calling for accountability. The EU Digital Services Act provides some framework, but it needs teeth, funding, and political will. Citizens, too, have a role: media literacy, verification, and a willingness to look beyond the algorithm. Journalists who fact-check in real time should be supported, not ignored. Platforms must feel public and regulatory pressure to act before the next election, not after. The human cost of inaction is immense. It is felt by candidates like Catherine Connolly, forced to deny a fake withdrawal; by campaigners who see harassment videos of their colleagues go viral; by election officials trying to reassure the public that ballots are secret and sacred; and by millions of voters who simply want reliable information to decide who should lead them. At the heart of all these numbers and examples is a simple human question: in a democracy, can we trust what we see? Ireland’s 2025 presidential election sadly suggests that, too often, the answer is no. But the report gives us evidence to build a better response. It reminds us that democracy is not automatic; it requires vigilance, and it requires institutions—public and private—to do their part. The next election is already on its way, and the voters deserve a chance to hear the truth.