Every major fraud claim of the past decade has been investigated. Here is what each investigation actually found.
American elections have seen genuine fraud — the 1876 presidential election was settled by a backroom deal after disputed returns, Tammany Hall stuffed ballots for decades, and a 2018 North Carolina congressional election was overturned for absentee ballot tampering. They have also seen a large volume of claims that investigation did not substantiate. This Dot takes the specific allegations one at a time — impersonation, deceased voters, noncitizen voting, ballot harvesting, machine manipulation, double voting — and reports what audits, prosecutions and litigation found in each category, along with what the studies did and did not examine.
Heritage fraud database
1,412
Proven cases, all types, four decades — the leading fraud-sceptic catalogue’s own count. Heritage calls it "a sampling," not a census.
Documented mail-fraud rate
4
Cases per 10 million mail votes — 0.000043%. Brookings, Nov 2025, using Heritage’s own database and News21.
Post-2020 lawsuits
64
20 dismissed pre-merits, 14 withdrawn by plaintiffs, 30 heard on the merits, 1 plaintiff win. "Lost, Not Stolen," 2022.
Arizona 2021 partisan review
+360
More votes for Biden than the official Maricopa canvass. Commissioned by Senate Republicans, run by Cyber Ninjas.
Georgia 2020
3
Counts — machine, full statewide hand audit, Trump-requested recount. Same winner, 11,779-vote margin, each time.
Before anything else — three different things
Almost every argument about vote fraud is two people using one word for three phenomena that differ by six orders of magnitude. Individual fraud is documented and rare. Administrative error is more common and usually caught. Outcome-changing systemic fraud is an extraordinary claim requiring extraordinary evidence. A true report of the first is routinely offered as proof of the third, and a refutation of the third is routinely offered as a denial of the first. Both moves are wrong, and this page will keep separating them.
Individual fraud
Documented. Rare.
A person casts a ballot they are not entitled to cast — their late mother's, a second one in another state, one in a jurisdiction where they no longer live. This happens every cycle and people are prosecuted for it every cycle. The Heritage Foundation, the leading catalogue built specifically to document it, has assembled 1,412 proven instances spanning four decades and every fraud type. It is real, it is criminal, and it is not in dispute.
Hundreds of documented cases per decade, against roughly 150 million ballots per presidential cycle.
Administrative error
More common. Usually caught.
A clerk fails to update a memory card. A voter file lists someone who moved. A precinct is reported twice on election night. Antrim County, Michigan misstated its unofficial 2020 count by thousands of votes in a county of under 22,000 registered voters — and it was a ballot-programming mistake, confirmed by a forensic examination and a bipartisan hand tally. Error is the ordinary condition of running 8,000 election jurisdictions on volunteer labour. It is also what the canvass exists to find.
Common enough that every state builds a correction process around it; almost always found before certification.
Outcome-changing systemic fraud
The extraordinary claim.
Tens of thousands of fraudulent ballots, coordinated, surviving signature verification, the voter file, the canvass, the audit and the recount, in a statewide or national race. This is a categorically different assertion from the first two, and it requires categorically different evidence. Four American elections in the modern era were overturned for fraud — all of them municipal or single-district races decided by a few hundred votes. None was statewide. None was federal except one U.S. House seat.
No statewide or presidential result in the modern era has been overturned or reversed for fraud.
The six claims, one at a time
Each claim below is stated first in its strongest form — the specific factual basis its proponents actually rely on, not a caricature — and then the finding, and then what the cited investigation did and did not examine. The last of those three is where most of the public argument goes wrong. Select a claim to read it.
Voter impersonation at the polls
Vanishingly rare
Someone shows up and votes in another person's name.
The claim at its strongest
It is the fraud that voter ID laws are designed to stop, and it is genuinely hard to detect: if no one checks identity at the book, a successful impersonation leaves no trace in the data at all. Absence of evidence here is partly a function of not looking. Proponents also point out that the studies below count *prosecutions and credible allegations*, which is a floor, not a ceiling — undetected impersonation would be invisible to every one of them.
What investigation found
Justin Levitt's review of every credible allegation from 2000 through 2014 found 31 incidents against more than one billion ballots cast — roughly one per 32 million. The News21 investigation, which filed public-records requests with election officers in all 50 states for every category of fraudulent activity, built a database of 2,068 alleged fraud cases since 2000 and found 10 involving impersonation. The Heritage database, compiled by the leading fraud-sceptic organisation, contained 10 impersonation cases when the Brennan Center audited it.
What the evidence does and does not cover
These are counts of *known* incidents — prosecutions, guilty pleas, and credible documented allegations. They do not and cannot measure undetected impersonation. What they do establish is that decades of prosecutors, partisan election officials in both parties, and a well-funded organisation looking specifically for this crime have found it at a rate of a few dozen incidents per billion ballots.
Bottom lineDocumented, at a rate of roughly one per 32 million ballots.
Georgia 2020 · ballots alleged vs. ballots verifieddrawn at true scale
AllegedVerified by investigation
Sources: Brennan Center fact-check of the January 2, 2021 call; Georgia Secretary of State; Atlanta Journal-Constitution. The yellow bars are drawn on the same scale as the orange ones and are therefore mostly invisible — that is the finding, not a rendering artefact. Four dead-voter ballots were verified out of a claimed 5,000; every one was a relative casting a deceased family member's ballot, and every one was referred for prosecution. Georgia's certified margin was 11,779 votes.
Cases in the Heritage database
1,412
Proven instances of election fraud, all types, spanning four decades — Heritage's own count. Heritage describes the database as "a sampling," not a census. An earlier vintage recorded 1,328 cases and 1,143 criminal convictions.
Impersonation incidents per billion ballots
31
Levitt's review of credible allegations, 2000–2014 — about one per 32 million ballots. Some may be clerical error.
Documented mail-ballot fraud rate
4 per 10M
0.000043% across the 2016, 2018, 2020 and 2022 generals — Brookings, Nov 2025, using the Heritage database and News21.
Impersonation cases in the Heritage database
10
When the Brennan Center audited it: 10 impersonation and 41 noncitizen cases out of 749 records at that vintage.
Noncitizens convicted of illegal voting
<100
In the Heritage database, 1982 to the present — against a July 2026 DHS assertion of ~278,000 noncitizens registered.
Texas prosecutions from the 2020 election
16
From roughly 22,000 staff hours. All were Harris County residents who gave false addresses on registration forms; none received jail time.
Scope
This Dot is about the evidentiary record — what audits, prosecutions, litigation and academic studies actually found, and what each of them examined. The mail and absentee method — how a mail ballot travels from request to count, and its rejection data — lives next door in Mail Voting, and the state-by-state rules landscape lives in How America Votes. Every figure on this page carries the source it came from, and how far the data runs is declared in the sources drawer below.