The U.S. Census Bureau published a seven-page brief on Aug. 18 concluding with "high confidence" that more than 24,000 of the 128 million 2020 voter records it has checked belonged to noncitizens. President Trump posted it within hours and wrote "I WON THE ELECTION!" The brief says nothing of the kind. It never says who those voters chose, never reports a margin of error, and carries no author's name.
What the document actually says
The brief is "Noncitizen Voting in the 2020 Election, A Beginning Analysis," dated August 2026, seven pages with two appendices. Its PDF metadata shows it was built Aug. 13 and last modified Aug. 18, the day it went online. Page 1: "The Census Bureau's beginning analysis readily determined with high confidence that over 128 million of these voter records represented citizens, and that over 24,000 of these voter records represented noncitizens at the time of the election."
The method is record matching, not a survey. The Bureau ran a commercial 2020 voter file, bought from a Virginia firm the brief calls DataClear, through its Person Identification Validation System to assign each person an anonymous key, then checked those keys against federal noncitizen sources including green cards, visas, ICE files and taxpayer ID numbers issued only to people ineligible for a Social Security number. A record counts as a noncitizen voter when a federal file says noncitizen and nothing anywhere says citizen as of Election Day.
The instinct that this is under-investigated is defensible
Voting in a federal election as a noncitizen is a crime under 18 U.S.C. 611, punishable by up to a year in prison, and grounds for removal. If the number is 24,000, that is 24,000 federal crimes nobody has been charged with. If it is 200, that is still 200. There is also no comprehensive federal database of American citizens, which is why every attempt at this question becomes a records-matching exercise.
When Justin Levitt of Loyola Law School writes that the finding cannot be separated from matching noise, he is also conceding that the noise is larger than the signal, so nobody can rule out a real number in that range either. "We should treat these 24,000 as allegations, not as definitive proof of noncitizens," University of Florida elections researcher Michael McDonald told CNN. Allegations are worth running down. This document was built so that nobody can.
Georgia is the cleanest test, and the two numbers are 20 times apart
Table 1 says 400 noncitizens voted in Georgia in 2020. Georgia's Republican secretary of state has already done this work the hard way. Brad Raffensperger wrote to acting Census director George Cook on Aug. 19, citing a 2024 statewide audit that found 20 noncitizens registered out of more than 8 million records, according to Georgia Public Broadcasting. He asked for the data rather than attacking the report: "I am asking for all the identifying information utilized by your agency to detect the alleged 400 non-citizens who would have voted in Georgia in 2020."
Georgia Secretary of State Brad Raffensperger, whose 2024 audit found 20 noncitizens registered out of more than 8 million Georgia records. The Census brief says 400 voted in Georgia in 2020 alone. Official portrait, 2024. (Wikimedia Commons, uploaded by Mthassinger, licensed CC BY 4.0)
Every noncitizen who voted had to register first, so either Georgia's audit missed 95 percent of them, which would be a state-level scandal, or the federal match is generating false positives at scale. Only one side of that question is refusing to show its work. Nevada's Democratic secretary of state, Francisco Aguilar, whose state got a count of 250, made the same demand to NPR.
The brief cites a paper on false match rates and never reports one
Levitt noticed at the Election Law Blog that one of the brief's five citations is a 2014 Census working paper titled "Estimating Record Linkage False Match Rate for the Person Identification Validation System," the exact system this analysis used. "The vast majority of the errors that paper finds ... are larger than 0.019%," he wrote. The 24,000 figure is 0.019 percent of 128 million. His conclusion: "it's entirely possible that every single purported noncitizen voter could instead be explained by errors in the matching process alone."
John M. Abowd, the Bureau's chief scientist from 2016 to 2022, told CNN: "None of the professional researchers that I knew or worked with at the Census Bureau would make such a claim from that study." Carolyn Carlson, president of the firm that sold the Bureau the file, confirmed to CNN and Votebeat that it carried no Social Security numbers, the one field that makes a match near-certain.
No career statistician put a name on it
Census products carry authors. This one carries none. Former bureau director Robert Santos, who resigned in the opening days of Trump's second term, called it an obvious "hijack by the White House" and told CNN, "I was shocked and dismayed that a document of this sort ... would appear on a Census Bureau website." Amy O'Hara, who ran the Bureau's Center for Administrative Records Research and Applications, told NPR the brief "lacks any sort of measures of variance or discussion of bias." On its closing claim about statistics on "election crime," she added: "even if they observed anomalies in the data, those anomalies aren't necessarily fraud."
The Census Bureau's headquarters in Suitland, Maryland. File photo, March 2007. (U.S. Census Bureau, public domain, via Wikimedia Commons)
NPR reported, citing a Census employee it agreed not to name, that career civil servants did not do the work and that the team included people affiliated with the America First Policy Institute. The Bureau, the White House, Commerce and AFPI answered none of the four outlets that asked.
Nearly 7,000 of the 24,000 have no country of birth at all
Table 2 breaks the total out by country of birth, and the shape does not match the political frame around it. Mexico leads at 3,800, Canada is second at 950, Germany fifth at 700, the United Kingdom sixth at 650. Only 71.2 percent were linked to a USCIS record naming a country, leaving roughly 6,900 records with no country in them at all.
Chart built by PatriotAddict from Table 2 of the Census Bureau's own brief. The single largest category is records with no reported country of birth. The Bureau rounds every count for disclosure avoidance, so the bars do not sum exactly to 24,000.
The 32 million records Trump says will make the number explode do not all exist
Trump's Truth Social post went up at 4:48 p.m. on Aug. 18: "On the first 128,000,000 Voters, the Census Bureau proves that over 24,000 Noncitizens voted illegally! The Census Bureau is going to analyze the next 32,000,000 Voters, and this number will explode. I WON THE ELECTION! We must pass THE SAVE AMERICA ACT." Commerce Secretary Howard Lutnick, whose department oversees the Bureau, posted that it is "just the beginning."
Now run the Bureau's own appendix numbers. Appendix 1 says the DataClear file captures 146,700,000 of the 159,934,200 ballots states certified to the Election Assistance Commission. Subtract the 128 million already checked and roughly 18.7 million records remain inside the file. The "over 32 million" only reconciles against the 160 million ballots actually cast, meaning about 13.2 million of it is ballots that are not in the file and cannot be checked this way. Apply the brief's own rate to what genuinely remains and you get roughly 6,000 more, for a total near 30,000.
There is also nobody to charge. Footnote 1 says the analysis team never saw a name, and 13 U.S.C. 9 bars the Bureau from using protected data for anything but statistics or publishing anything by which an individual can be identified. Raffensperger's request for 400 names runs straight into that statute, and nobody at Commerce has said which way it cuts.
The SAVE America Act is four votes short and out of runway
The House passed the bill 218 to 213 on Feb. 11, 2026, clearing S. 1383 with substitute text from H.R. 7296, per the National Association of Counties. It would require documentary proof of citizenship to register and photo ID to vote. On the narrow question this brief raises, that is the direct fix. The 24,000, if real, registered and voted without anyone checking a document.
The Senate math has not moved. Republicans hold 53 seats, cloture takes 60, and Majority Leader John Thune has said all year the votes are there neither for the bill nor for ending the filibuster, the Associated Press reported. He declined Trump's demand to cancel the August recess and the Senate left on Aug. 7. David Becker told the Washington Examiner that under the Purcell principle a court would block implementation this close to November anyway. The Brennan Center counters that more than 21 million voting-age citizens lack ready access to a birth certificate.
The Census Bureau produced a number that cannot carry the weight the president put on it, and spent institutional credibility that election-integrity conservatives actually need. A Republican secretary of state who has already audited his own rolls asked for the records so he could investigate 400 specific allegations. That is the test. If the names arrive and Georgia opens 400 files, the finding was real. If they never arrive, an unsigned brief with no error rate was never about prosecuting anyone.
Header photo: President Trump with Vice President JD Vance and Commerce Secretary Howard Lutnick in the Oval Office at Lutnick's swearing-in as Commerce Secretary, Feb. 21, 2025. Commerce oversees the Census Bureau. Official White House photo, public domain, via Wikimedia Commons.
