Can AI disinformation actually swing an election outcome?

Political Capital's post-election report for EDMO/HDMO (Kreko et al., 2026-04-29) examined Fidesz's 2026 Hungarian parliamentary campaign and found that record AI-generated disinformation "reached an unprecedented level" yet was "only partially successful" in swaying opinion — Fidesz lost. Its public opinion survey found 73% of respondents had encountered AI-generated content on social media, 52% of those who saw deepfake videos were unsure whether they were authentic, and 37% had at some point been misled into believing manipulated content was real. Despite that reach, the same survey found 90% of respondents considered AI-generated or manipulated political content "entirely unacceptable," against only 3% who found it "somewhat acceptable."

The report frames AI content's persuasive power in psychological terms: AI-generated video and images "transform political messages into highly persuasive and emotionally powerful visual narratives" because people trust what they see, and the fear, anger, or moral outrage this evokes "reduces critical thinking and increases the likelihood of sharing." But it attributes the campaign's limited success to factors beyond persuasion mechanics: near-universal public distaste for AI-manipulated content; a credible opposition challenger (Péter Magyar) who could fact-check Orbán's claims and was more trusted than the incumbent; voter "fatigue" toward government disinformation; preemptive warnings about expected manipulation that "played an important role in neutralizing disinformation before it could spread"; and a growing independent media audience that kept exposing abuses of power.

This sits against lab-measured findings that generative AI reliably boosts persuasiveness, such as Where does AI's persuasive power actually come from?, and against Is AI shifting from content creation to strategy in influence operations?'s picture of AI as strategic orchestrator: Hungary shows that even heavy, court-defying use of AI-generated disinformation does not translate cleanly into votes once most of a population rejects the practice. It also extends the warning-based findings in Can a simple warning reduce how much LLMs persuade people? and Does telling people they are talking to AI change how persuaded they become? from the lab to a real election — Political Capital's "preemptive communication" is the field analog of an experimental persuasion warning, and it reports a similar dampening role. The 90% rejection figure also complicates Do readers trust unlabeled AI-written messages as much as human ones?: in Hungary, AI origin was widely suspected rather than hidden (52% unsure, 37% misled at some point), and that suspicion tracked with near-universal disapproval rather than indifference.

The report is a single national case from one contested election, produced by a watchdog organization rather than a neutral academic lab, and it cannot isolate how much of Fidesz's loss is attributable to AI-disinformation rejection specifically versus the other named factors — challenger trust, fatigue, independent media. Its survey data are self-reported perceptions of deepfake exposure and attitudes, not measured behavioral shifts in vote choice. The excerpt's own hedge is instructive: Fidesz's "trendsetter role might decline... but deepfakes in the campaigns definitely won't disappear" — implying that public rejection of AI-manipulated content in one election is not evidence the content lacks political effect elsewhere, only that it failed to overcome a specific configuration of opposition trust and fatigue in Hungary in 2026.