The specter of misinformation, once a slow-burning ember passed between human actors, has found a terrifying new accelerant: AI chatbots. What does this mean for the average voter trying to navigate the complex landscape of an election? It means that the very tools they might turn to for accurate information are, according to a new Demos study, actively misleading them. We’re not talking about subtle nuances here; we’re talking about fabricated scandals, incorrect election dates, and erroneous ID requirements—all presented as fact by services like ChatGPT, Google Gemini, and Replika.
This isn’t a drill.
Demos’ report, titled ‘Electoral Hallucinations,’ paints a stark picture: 34% of the questions posed to free AI tools yielded misinformation during simulations for the recent Scottish election. Consider the market share: a parallel poll found 20% of British adults — roughly 10 million people nationwide — had used AI for election information. The scale of potential delusion is staggering.
The Regulatory Void
The Electoral Commission, through its chief executive Vijay Rangarajan, is rightfully sounding the alarm. Half of voters encountered misleading information during the 2024 general election, a statistic that should send shivers down the spines of anyone who values democratic integrity. Rangarajan’s call for ministers to introduce legislation making AI companies more accountable is not just a suggestion; it’s a desperate plea to shore up a crumbling legal framework.
“Voters want accurate information to help them engage with democracy and it is concerning that AI tools have made the spread of false or misleading information dramatically faster and more accessible than ever,” he said. “The current legal framework should go further.”
What’s particularly galling is the global nature of the problem, yet the deficit in localized regulation. Azzurra Moores from Demos points out the irony: these pervasive tools are developed by US corporations, yet the UK lags behind in legislative protection for its public and its democracy from the fallout of widespread falsehoods. The proposals are concrete: hold AI firms liable under UK defamation and electoral law, mandate accuracy safeguards, and allow independent researchers to scrutinify internal data and training sets.
How Bad Were the Hallucinations?
The Demos investigation laid bare the varying degrees of AI incompetence. Replika, the companion chatbot, performed worst, with a staggering 56% error rate, even inventing a candidate and detailing made-up scandals. ChatGPT, despite its ubiquity, still faltered in 46% of its answers, including significant factual errors about election dates and eligibility rules.
Google Gemini, while better at 22% inaccuracy, still managed to misrepresent a candidate’s stance and wrongly claim an ongoing police investigation. Grok, Elon Musk’s AI, showed the lowest error rate at 9%, but its tendency for irrelevant or poor-quality external links undermines its utility. And Google’s own AI Overviews? It only answered 11% of prompts, making it largely irrelevant for this kind of fact-checking.
Compounding the issue, nearly half of the AI responses failed to cite official sources, and when links were provided, they were often broken or woefully out of date—44% of ChatGPT’s citations were at least a year old. This isn’t just unreliable information; it’s information presented with a veneer of authority that actively obstructs verification.
A Government Playing Catch-Up
The Department for Science, Innovation and Technology offers platitudes about defending elections being a “priority” and that work is “ongoing.” They cite the Online Safety Act, which they claim is closing loopholes to protect users from illegal content. Yet, there’s no concrete commitment to amending existing legislation like the Representation of the People Bill. The sentiment is there: “AI is critical to the UK’s future prosperity and security. But if we want people to seize the benefits this technology promises, they need to be able to trust it.” This is where the government’s strategy feels like a missed opportunity; proclaiming trust is insufficient when the very mechanisms of misinformation are actively being amplified.
Replika’s defense—that it’s a “companion for reflection and self-expression, not as a source of factual or real-world information”—highlights a critical disconnect. While their users might be informed of the chatbot’s limitations, the broader public, encountering these tools through social media or casual search, may not be. This industry-wide issue demands a proactive, not reactive, regulatory approach. Relying on users to discern the fabricated from the factual when dealing with systems designed to mimic human intelligence is a fool’s errand.
This situation is not merely an inconvenience; it’s a systemic vulnerability. The ease with which AI can generate and disseminate plausible-sounding untruths poses a direct challenge to the informed electorate that underpins any functioning democracy. The tech industry’s reliance on self-regulation, or a piecemeal approach to existing laws, has proven utterly insufficient. The time for decisive, legislative action is now, before the next election cycle sees an even more sophisticated wave of AI-driven disinformation wash over the public consciousness.
