Over the weekend, the Republican presidential candidate shared a pair of posts on his social network, Truth Social, that included AI-generated images: one depicting a hammer-and-sickle flag over a Soviet-style Harris rally, another showing young women in “Swifties for Trump” T-shirts. On Monday, he reposted the Harris image to X and sent an email to supporters calling it “the photo Kamala doesn’t want you to see.”
What’s noteworthy isn’t just that Trump is turning to generative AI to blur the truth. It’s the casual, almost mundane way he’s using it so far: not as a sophisticated weapon of deception, but as just another tool in his rhetorical arsenal.
Experts have been warning for years of AI’s potential to fuel political misinformation. For many, the nightmare scenario centers on a sophisticated, AI-generated or manipulated video that tricks millions into believing a bombshell falsehood. Imagine a video that appears to show election workers secretively trashing ballots or a candidate accepting a bribe.
While there have been a few examples like that, especially abroad, Trump’s use of AI over the weekend illustrates a simpler — and so far, much more common — application of the technology.
The images of Swifties for Trump and Harris as a communist leader don’t appear intended primarily to fool people, though of course some might be taken in. The Swifties image, part of an X post of which Trump shared a screenshot, carried a “satire” label. The Harris image was unlabeled but patently unrealistic.
Rather, the images seem to function more like memes, meant to provoke and amuse. They’re visual parallel to the nasty nicknames Trump calls his opponents and the evidence-free claims he often makes on the trail. For a politician whose favored rhetorical mode is the unconfirmed anecdote — “Many people are saying … ” — generative AI offers a handy new way to illustrate his stories.
This becomes evident when you view the images Trump posted over the weekend in the context of his other Truth Social posts. Far from standing out, the AI depictions of Harris and Swift nestle right into a timeline replete with satirical songs, photoshopped images, jabs at “Comrade Harris” and dubious claims about rally crowd sizes.
The same day Trump posted the AI image of Harris with the hammer-and-sickle banner, the New York Post’s front page portrayed Harris speaking from a dais adorned with a giant hammer-and-sickle logo, under the banner headline “Kamunism.” Trump posted that too.
The distinction between AI memes and deepfakes might seem trivial at first. Both uses of the technology serve to smear political rivals and muddy the informational waters. Both also contribute to the so-called “liar’s dividend,” in which the proliferation of fakes and falsehoods makes it easier for people to dismiss the truth as a fabrication — as Trump did when he suggested that Harris had “A.I’d” a genuine photo of a large crowd greeting her in Michigan.
But a convincing deepfake is hard to make, at least for now, whereas generating less realistic AI images is easier than ever. Even as leading AI image tools such as OpenAI’s Dall-E 3 and Midjourney have sought to prevent users from generating images that could be associated with political misinformation, Elon Musk’s X last week debuted a new version of its Grok AI tool that has few such limitations. A report on Monday by NewsGuard, a company that tracks misinformation, found that Grok complied with 16 of 20 test requests to produce such images; Midjourney complied with nine and Dall-E two.
And while a deepfake needs careful debunking, AI memes are impervious to fact-checks — the intended audience doesn’t care whether they’re literally true. The fake images feel true on some level, or at least it’s enjoyable to imagine that they might be. And if the other side gets righteously riled up about it, the joke is on them. (Democrats troll too; think JD Vance and the couch.)
We might yet see more sophisticated attempts to dupe people with realistic AI fakes in this election cycle, or in others to come. But we also shouldn’t be surprised if AI continues to be used more for propaganda than trickery. As Mike Caulfield, an information literacy expert and author of the book “Verified” put it in a recent blog post:
The primary use of “misinformation” is not to change the beliefs of other people at all. Instead, the vast majority of misinformation is offered as a service for people to maintain their beliefs in face of overwhelming evidence to the contrary.
In reality, any involvement in the election from Swift would probably work in Harris’s favor. But the world of AI offers Trump and his backers an appealing alternate reality in which the Harris campaign is in shambles and even Swifties are turning against her.
Whether playing pretend in this way will help or hurt Trump at the polls in November is unclear. But if his plan B is to question the results of the election, a supporter base accustomed to living in fantasyland could come in handy.