Artificial intelligence has advanced to the point where it can fabricate hyper-realistic images, videos, and audio with unsettling ease, blurring the line between genuine content and machine-manufactured deception. For the average person—and, increasingly, for those who have spent their careers studying digital media—telling what is real from what is generated has become a guessing game. That reality was graphically illustrated in a recent exercise involving Lydia Ray, a professor of computer science at Columbus State University and a cybersecurity researcher who has long focused on election security. Ray took a quiz created by the platform Real or Not, which presents users with a carefully curated mix of AI-generated and authentic materials and challenges them to classify each item as real or fake. She attempted the quiz multiple times, hoping to improve her score with each try. Her best result was only 53 percent—barely better than a coin flip, and far below what she would have expected from her professional training. For someone who spends much of her working hours examining synthetic content in the context of political misinformation, the failure was both humbling and alarming. Ray explained thatthe quiz was deliberately constructed to expose the limits of human perception in the face of advanced generative models. “I think the point of this quiz was to demonstrate that even if we are professionals, we are very limited in our abilities to correctly determine whether an image is a fake image or whether a video is a fake video or not, or fake audio or not,” she said in an interview with the Ledger-Enquirer. That statement carries particularly heavy weight as the 2026 midterm elections approach, when massive volumes of political advertising, social media posts, and manipulated clips will flood the information ecosystem. If a trained researcher specializing in deepfake detection cannot reliably distinguish fabricated content from genuine footage, ordinary voters scanning their feeds will be even more vulnerable to schemes designed to deceive them, suppress their participation, or erode their confidence in the democratic process. The quiz was not meant to embarrass test-takers, Ray emphasized; rather, it served as a wake-up call aboutthe futility of relying on visual inspection alone in an era where AI can produce content that is aesthetically, structurally, and contextually indistinguishable from reality.
Ray began studying election misinformation in2016, when foreign interference in American politics was first moving into the mainstream public consciousness. In the years since, her research has evolved to focus on the more ominous capabilities of generative AI: synthetic audio that clones acandidate’s voice, video that swaps faces onto real bodies, online articles that mimic legitimate journalism, and highly realistic images of events that never occurred. The threat is no longer hypothetical.) AI tools are now widely accessible, affordable, and easy to use, meaning that nearly anyone with an internet connection can create convincing fake media with minimal effort. Ray noted that human beings are naturally inclined to believe what they see and hear, especially when the content aligns with their expectations or emotions. But in the current environment, that instinct is dangerous. “It is really difficult for our human eye to figure out whether an image or a video is AI-generated,” she said. “You have to be very critical.” Her point is echoed by other cybersecurity researchers who warn that simply looking at a piece of media is no longer a sufficient method of verification. Fine-grained details—shadows, reflections, skin texture, blinking patterns, inconsistent lighting—can be telltale signs, but modern generative models are rapidly correcting those imperfections. At the same time, AI can also be used to enhance authenticity, with systems that intentionally introduce natural imperfections to counterfeit content, making it even harder for analysts and fact-checkers to spot red flags. For election officials, journalist, advocacy groups, and voters, the shifting technological landscape means that old assumptions about proof and evidence are crumbling, and new habits of verification must take root quickly.
The specific dangers of AI-powered election misinformation are twofold,according to Ray. First, false claims can be manufactured to undermine public confidence in the integrity of election results. This includes fabricated videos purporting to show voting machines being rigged, audio recordings of election workers conspiring to steal votes, or images of ballot boxes being stuffed with fraudulent ballots. Second, false information can be created to discourage or confuse voters, particularly those from marginalized communities that already face barriers to participation. This could take the form of convincing but entirely invented announcements about last-minute polling location changes, altered voter identification requirements, new eligibility rules, or incorrect election day deadlines. Such disinformation is especially insidious because it often circulates through trusted community networks, religious groups, or ethnic media outlets, making it difficult for recipients to question its authenticity. AI also lowers the cost of targeting individual candidates with character assassination campaigns. A doctored video of a candidate saying something incendiary could be produced in minutes and spread across platforms before the campaign can issue a denial, and eventhen, the damage to public perception may already be done. For Ray, however, the greatest long-term danger is not any single piece of fake content but the cumulative erosion of trust in all information. Widespread, persistent misinformation damages voters’ confidence not only in individual politicians but in the election system itself,and once that confidence is broken, it is extraordinarily difficult to restore. Moreover, the very existence of increasingly realistic AI-generated content provides a convenient excuse for dismissing inconvenient truths. Politicians under scrutiny can wave away authentic recordings as deepfakes; journalists covering real events may see their reporting dismissed as suspect; voters may become so cynical that they tune out all political information entirely. “We don’t know who to trust,” Ray said. “The fact that we don’t know where to get the correct information, that overwhelms me.” That sense of overwhelm, she warned, could have profound consequences for voter turnout and the legitimacy of democratic governance in the United States, particularly in a high-stakes midterm cycle where control of Congress, state legislatures, and governorships will be determined.
In southwest Georgia, Muscogee County election officials have so far not encountered any AI-generated or manipulated content directly related to their local elections, according to Nancy Boren, director of Muscogee County Elections and Registration. But my sparents and absence of incidents does not mean they are unprepared. The county’s elections office has quietly been building a defensive posture against the new wave of election disinformation. Boren confirmed that her staff has received specialized training on AI-generated misinformation, deepfakes, and the spread of false election information from multiple trusted sources, including the Multi-State Information Sharing and Analysis Center, the Center for Internet Security, the U.S. Election Assistance Commission, and the Georgia Secretary of State’s office. These organizations provide election administrators with threat intelligence, detection protocols, and response playbooks specifically tailored to the unique challenges of the AI era. The training covers how to recognize common patterns of AI-powered interference, how to respond when false content about an election emerges, and how to communicate effectively with the public to correct the record before misinformation goes viral. Boren said that if false information about the election were to arise—for example, a fabricated announcement about a polling place relocation, a made-up claim about extended voting hours, or a doctored video of a poll worker acting improperly—the office would activate its communication plan. That plan includes issuing statements through local news media, posting authoritative notifications on the city’s website, sharing updates on official social media channels such as the Muscogee County Elections Facebook page, and coordinating with the Secretary of State’s office to ensure that voters receive consistent, trusted guidance. Election officials understand that in an information emergency,speed and credibility are everything. A slow or muddled response can allow lies to take root and spread, while a fast, transparent,multi-channel correction can limit harm. The county’s preparation reflects a broader trend in election administration across Georgia, where officials are increasingly treating disinformation as a security threat on par with cyberattacks or physical intimidation at the polls. Because AI makes it possible to generate credible-looking fakes in a matter of minutes, election workers must be ready to respond at a moment’s notice—often before news organizations have even had the chance to independently verify the content. The office’s proactive stance provides a model for how local jurisdictions can defend democratic integrity in an environment forensics tools alone cannot solve.
For voters trying to navigate this treacherous landscape, Ray and Boren offer a common piece of advice: do not rely on shallow visual cues to determine whether something is real. Instead, voters should focus on verifying the source of the information and cross-checking claims against authoritative, established institutions. Ray shared a list of practical recommendations for Georgians planning to vote in the November 3, 2026, midterm election. First, she said, go directly to your state or county election office’s website for anything jurisdiction-specific, such as voting deadlines, polling places, ballot rules, and identification requirements. Do not enter a query into a general search engine or ask a chatbot, because AI tools can return outdated, incorrect, or even maliciously crafted answers that appear authoritative. Second, pause before sharing anything online that provokes a strong emotional reaction—whether outrage, fear, joy, or triumph. Disinformation campaigns are designed to exploit emotional impulses, and people who share content in the heat of the moment become the unwitting amplifiers of a hoax. Third, if a claim is unfamiliar or extraordinary, verify it against an independent fact-checker accredited by the International Fact-Checking Network, or IFCN, before reposting. Fact-checking organizations specialize in tracing the origins of viral claims and separating verified reality from coordinated falsehoods. Fourth, treat AI-generated-looking content skeptically by default, especially material that depicts election workers, ballots, or fraud in progress. But crucially, Ray warned, do not assume you can identify such content just by looking at it; the absence of visible artifacts does not mean the content is genuine. Voters should always ask who produced the material, whether it has been reported by reputable news outlets, and whether the account sharing it has a history of trustworthy posts. Boren echoed that advice in more concise terms. She urged voters to check whether a social media account is authentic before treating its content as reliable, to confirm that dates and locations in a post actually match real election procedures, and to seek independent confirmation from trusted sources—such as official election offices and established news organizations—before acting on any unfamiliar claim. “Check it before you spread it,” she said. This simple mantra may be one of the most effective defenses against the growing tide of synthetic media, because a piece of misinformation only becomes dangerous when people believe it and pass it along. By making verification a habitual first step, each voter can help stem the viral spread of AI-generated falsehoods, protecting both personal decisions and the health of the broader electoral process.
As the 2026 midterm elections draw closer, the challenge posed by artificial intelligence will almost certainly intensify. Generative models are becoming more sophisticated by the month, capable of producing persuasive content that was impossible to create just a few years ago. The same tools that enable creative innovation also enable malicious actors—domestic and foreign—to interfere with elections at unprecedented scale, speed, and subtlety. There is no technical silver bullet that will restore the old verities of visual proof. Deepfake detection algorithms exist, but they are in a constant arms race with generation techniques, oftentimes lagging behind the latest advances. Consequently, the burden of maintaining epistemic order inevitably falls on people, institutions, and democratic communities. Election officials can train, prepare, and respond; social media platforms can attempt to label and remove synthetic content; journalist can investigate that has been posted, but it remains ultimately up to voters to adopt a more disciplined, deliberate approach to consuming and sharing information. The advice from Ray and Boren is straightforward yet profound: trust verified sources over viral sensations; prioritize official information over emotionally compelling anecdotes; takea moment to think before clicking share. Lydia Ray’s experience with the Real or Not quiz exposedjust how unreliable human intuition has become in detecting AI-generated media. But it also reinforces why critical thinking, source verification, and institutional trust are more important than ever. In an era when seeing is no longer believing, the act of checking becomes a civic duty. Voters who arm themselves with these habits and who demand accountability from politicians, platforms, and themselves, will be better equipped to preserve the integrity of their elections. The question is notwhether AI will test the system—it already does—but whether the public can learned to navigate a world where both genuine truth and convincing falsehoods demand scrutiny. As Ray put it, the overwhelm of not knowing whom to trust can be paralyzing. But the antidote is not despair; it is vigilance, methodical verification, and a shared commitment to protecting democratic governance from the corrosive effects of synthetic deception. In the end, the midterm elections will be decided not just by who votes, but bywhether enough people stop to ask one simple question before trusting what they see: Where did this actually come from, and how can I confirm it is true?
