Donald Trump and his allies quickly seized on an announcement from Mikie Sherrill, the New Jersey governor, on Tuesday that 6,600 noncitizens had registered to vote because of a software error, and fewer than 400 of them voted, using the disclosure to escalate false claims that there are scores of noncitizens on the voter rolls.

Trump seized on the news days after he said in a primetime speech that the Department of Homeland Security identified 278,000 noncitizens on the voter rolls in New Jersey and three other states. But a Guardian review of the letters the DHS sent the states shows the agency expressed less certainty about the number and has offered no explanation of its methodology.

Since 2018, New Jersey has had automatic voter registration, which means that any time someone interacts with a motor vehicle office they are automatically registered to vote, unless they opt out. Between 2023 and 2024, some 6,600 people had indicated they were not citizens, yet nonetheless had been registered to vote. About 400 of those people cast a ballot, Sherrill said. Harmeet Dhillon, the head of the justice department’s civil rights division, sent a letter to Sherrill earlier this week requesting the names of those who were registered and voted and the department is reportedly mulling prosecutions. Sherrill suggested on Thursday the state wasn’t going to turn over the information on voters and said earlier in the week it would be unfair to prosecute them for an error that was the state’s fault.

The number represents a small fraction of the more than 6.6 million registered voters in the state. But within hours of the announcement, Trump and other election deniers inflated the number of people who registered to vote because of the error, and held it up as yet another data point that “illegals” are stealing American elections on behalf of Democrats.

Trump said in a post on Truth Social: “35,152 is the new number just released by Homeland Security of non-citizens registered to vote in New Jersey. These are just the ones that got caught. The real numbers will prove to be many times this amount. They are all Dumocrats!.” There were Democrats and Republicans erroneously registered, Sherrill said on Tuesday, with the majority unaffiliated with a party.

The incident is far from proof of the widespread fraud that Republicans allege, experts say.

“It’s not good, I don’t want to suggest otherwise. I don’t want to tell anyone that they’re stupid for thinking it’s less than perfectly optimal. But it’s not really an example of what they’re talking about,” said Stephen Richer, a Republican who oversaw elections in Maricopa county, Arizona, from 2021 to 2025. “These people were trying to do the right thing and acknowledge that they’re not citizens, and then the vast majority of them, even though registered and presumably [were] sent election mail materials, ignored that and didn’t vote.”

The number of non-citizens the state says were registered represents a “minuscule fraction” of New Jersey’s active registered voters in 2024, said David Becker, an election administration expert and the executive director of the Center for Election Innovation and Research. “The number of potential ballots cast from those incorrect registrations would represent roughly 0.009% of all ballots cast in New Jersey’s 2024 general election, when Kamala Harris defeated President Trump by more than 250,000 votes.

“While administrative errors such as New Jersey’s do infrequently occur, they are very rare and affect a very small fraction of voters. The states entrusted to run our elections are best positioned to detect and correct these errors, as they have access to substantially better data than the federal government,” he said.

The firestorm over the New Jersey voter rolls came as Trump escalated his incendiary rhetoric about voting ahead of the midterm elections. Last week, in a primetime address from the White House, he offered what seemed to be an eye-popping statistic: the Department of Homeland Security (DHS) had found 278,000 non-citizens on the rolls in just four states – California, Nevada, Pennsylvania and New Jersey. That, too, is a misleading figure.

In its letters to election officials in those four states DHS expressed less certainty about the 278,000 number, saying it was a ceiling of how many noncitizens could be on the rolls. The agency also offered little to no explanation of its methodology for identifying noncitizens on the rolls. It gave the states two weeks to respond.

“It’s a laughable letter,” Becker said. “Imagine someone said to you: ‘Some of the items in your fridge are poisonous. Write us back in two weeks and tell us what you’re going to do about that.’” The first question one would ask, Becker said, is “which items?”

“They’re not serious. They don’t want the states to even maintain their records correctly. If they did, they would be completely transparent, and they would have acted immediately and said we absolutely need you to do this,” he added. “This is designed to create a false argument post-election that the election was stolen. This has nothing to do with actually cleaning up the lists.”

The noncitizen registration claim was part of a slew of misleading information Trump offered in a speech last week designed to weaken confidence in the US election system ahead of November midterm elections in which his party is expected to lose seats. It also comes as Republicans are pushing legislation that would impose new proof-of-citizenship requirements on voter registration, even though several studies have shown noncitizen voting is extremely rare.

Last week,both Trump and Markwayne Mullin, the homeland security secretary, cited the 278,000 number as evidence that states should share sensitive voter information with the federal government, with Mullin threatening potential prison time for those who don’t comply. At least one Republican has cited the number as a bona fide example of voter fraud. “We know there are roughly 278,000 noncitizens registered to vote. The last three presidential elections were decided by fewer votes,” Marsha Blackburn, a senator of Tennessee, said on Tuesday.

“We can affirm that on its face, we refute these claims. These numbers are wildly speculative at best and the Department of Homeland Security hasn’t shared anything that backs it up,” said Cisco Aguilar, a Democrat serving as Nevada’s secretary of state. As of Wednesday afternoon, the DHS had not provided Nevada with a list of suspected noncitizens in the state.

In a letter to New Jersey officials last week, the DHS said there may be as many as 35,152 noncitizens on the state’s voter rolls. The agency said it had matched 19,497 voter records to a noncitizen in its system, but did not offer an explanation for the rest.

Sherrill said she had been informed of the voter registration issue in her state and begun investigating before Trump’s speech on Thursday evening. She said the DHS had provided no information to back up its claims.“We have asked them for any evidence of this, where they got that number, who this is. We’ve received nothing back yet,” she said on Tuesday.

Understanding the methodology the DHS used is critical to assessing the veracity of its claims, experts say. Matching records across different databases is difficult, and the system the DHS uses to try and identify noncitizens – Systematic Alien Verification for Entitlements (Save) – is known to be error-prone. The DHS said it relied on publicly available voter information from the four states to try and identify noncitizens, information that probably lacks the level of detail needed to get a solid match between voter records and DHS data, Charles Stewart III, a professor at MIT and an expert in election administration, wrote in a post on Substack.

“DHS likely lacked the identifiers needed for high-quality matching. Save checks generally require name, date of birth, and a unique immigration or Social Security identifier, and public voter files ordinarily do not contain the most sensitive identifiers,” he wrote. “Anyone with experience in these data-matching exercises should expect those numbers to melt as quickly as ice cream in a Phoenix August.”

Even though the agency claimed there could be up to 278,000 noncitizens on their rolls, the DHS said in its letters to the states that it was able to match the name, date of birth, address and social security number to a noncitizen in its system for about 118,000 people. The agency offered no explanation for why it believed there were an additional 160,000 noncitizens on the rolls

The DHS did not respond to multiple requests for comment on its methodology.

“I am about as certain as I can be, without them coming up with some magical data technique that nobody has ever seen before, that these numbers cannot be even remotely credible,” Becker said. “What I can’t tell you is whether they just made them up. Because they have not been transparent about that. And it’s possible that they just made them up, because we just haven’t seen any methodology whatsoever.”

Claims of widespread noncitizen registration and voting often start with a headline-grabbing statistic, and then the numbers often whittle down once election officials investigate. In 2019, for example, Texas announced it suspected there were nearly 100,000 noncitizens on its rolls, 58,000 of whom may have voted. The state later admitted the claim was based on bad data and agreed not to cancel anyone’s voter registration based on it.