Misinformation
Misinformation is false or misleading information that spreads regardless of whether the person sharing it knows it is wrong or intends harm. The word covers an honest mistake, a misremembered statistic, an old photograph captioned with a new event, a satirical headline taken literally, and a rumor passed along out of concern for friends. It is usually contrasted with disinformation, which is false content created or pushed deliberately to deceive, and the distinction matters because the remedies differ: misinformation is mostly a problem of accuracy and attention, while disinformation is a problem of adversaries, incentives and campaigns.
The standard vocabulary comes from the 2017 report "Information Disorder" by Claire Wardle and Hossein Derakhshan for the Council of Europe, which separated three categories by falsity and intent. Misinformation is false but not intended to harm, disinformation is false and intended to harm, and malinformation is genuine information used to harm, such as leaked private material or a true statistic stripped of context. The categories are not fixed properties of a piece of content, because the same false claim can be disinformation when a propagandist seeds it and misinformation when ten thousand sincere people repost it. Researchers therefore often describe a flow: an originator with some motive, a message in some format, and an audience that reinterprets and re-shares it with entirely different motives of its own.
Falsehood does not travel only because of bots; the best-known measurement suggests that people themselves do much of the spreading. In a 2018 study in Science, Soroush Vosoughi, Deb Roy and Sinan Aral analyzed roughly 126,000 stories tweeted by about three million people between 2006 and 2017 and found that false news reached more people, and did so faster, than true news, with false stories about 70 percent more likely to be retweeted. Automated accounts spread true and false stories at similar rates in their data, so the difference was attributed to humans, with novelty a leading hypothesis. The finding is about Twitter and about stories verified by fact-checking organizations, so it should not be stretched to every platform or every kind of falsehood, and a separate body of work suggests that the volume of misinformation consumed by an average person is smaller and more concentrated in a minority of heavy users than headlines imply.
Generative AI changes the economics of misinformation more than its psychology. A language model can produce fluent, plausible, endlessly varied text at near-zero marginal cost, an image model can fabricate a photograph-like scene in seconds, and voice-cloning systems can imitate a speaker from a short sample. Deepfakes, a term that began with face-swap videos on Reddit in 2017, are the audiovisual form of this. Documented episodes include a crude video that appeared in March 2022 purporting to show Ukrainian president Volodymyr Zelenskyy telling soldiers to surrender, quickly denounced and removed; an AI-generated image of an explosion near the Pentagon that circulated on social media in May 2023 and was publicly debunked by officials; and robocalls in January 2024 that imitated President Joe Biden's voice to discourage New Hampshire primary voters, after which the US Federal Communications Commission ruled in February 2024 that AI-generated voices in robocalls count as artificial voices under existing telephone law. A pre-AI precedent is the hacked Associated Press Twitter account that falsely reported White House explosions in April 2013, which briefly moved stock indexes, a reminder that cheap fabrication was never the only ingredient.
The second-order harm of synthetic media is that real evidence becomes easier to deny. Legal scholars Danielle Citron and Robert Chesney named this the liar's dividend in 2019: once the public knows that convincing fakes exist, a politician or executive caught on genuine audio or video can claim it was fabricated and gain some benefit from the doubt. The practical consequence is that a world flooded with fakes does not simply produce more false beliefs; it also erodes the evidentiary weight of true recordings, which is one reason that proposals for content provenance focus on authenticating real material rather than only hunting for fakes.
Common misconceptions make the problem look easier than it is. One is that a correction backfires and entrenches the false belief; large replication efforts, such as Thomas Wood and Ethan Porter's 2019 study across dozens of political claims, found that backfire effects are rare and that corrections usually help modestly, though the effect fades with time. Another is that detection by AI will settle the matter, when detectors are unreliable and attackers adapt. A third is to treat every false statement as a deliberate campaign, which overstates coordination, while the opposite error treats organized influence operations as imaginary. Credible mitigations are layered: platform friction and labeling, prebunking exercises that expose people to weakened manipulation techniques in advance, source transparency, and legal accountability for fraud and impersonation, none of which is sufficient alone.
Read misinformation as a property of how a claim travels rather than of the claim alone: intent separates it from disinformation, human attention and platform design explain its reach, generative AI lowers the cost of fabrication while raising the cost of trusting any single clip, and the durable defenses are calibrated skepticism, verifiable provenance and accountable distribution rather than any single detector.