Election Auditing in the United States

Fifty-One Ways to Almost Check an Election

Roughly 2900 jurisdictions ran some kind of tabulation audit following the 2024 General Election. The nature of these audits depends on state law and vary widely and counties within these states interpret laws in different ways. In 2024, a few states ran statistically rigorous audits but the majority of jurisdictions used low fixed-rate (1-2% typically) hand counts compared against vote totals reported by the election system. A few more used election recounts by analyzing images of cast paper ballots. In September 2026, National Conference of State Legislatures (NCSL) stated “All 50 states and Washington D.C. conduct some type of post-election audit” but several of these are procedural audits that do not check vote-counting accuracy, so there are a handful of states that did not conduct any tabulation auditing at all.

In the interest of further characterizing audits in the U.S, as well as an interest in learning about AI systems, I used AI agents to assist in surveying the audit landscape. These agents did remarkably efficient data collection from all 51 jurisdictions. Despite this capability, several states required public records requests to figure out what they actually did in 2024. This website expresses the state of audits just after the 2024 General Election.

The variations in approaches to audits are perhaps not too surprising given that each state legislature enacted audit laws, in most cases without looking seriously at what other states were doing. And many states enacted their audit laws before the more rigorous methods had been developed or were widely known. In 2019, 34 states had some type of post-election tabulation audit, only Colorado, Rhode Island and Virginia required risk-limiting audits, and Maryland was the only state requiring a ballot image audit statewide (Ryan, 2020). By November 2024, 46 of 51 jurisdictions ran a tabulation audit of some kind, 8 states ran an audit with an explicit risk limit, and South Carolina had joined Maryland in re-counting every ballot image. The less intuitive result of this survey was the large number of factors that determine the utility or effectiveness of a tabulation audit.

Why election audits are hard

Tabulation audits can be difficult to execute accurately. That fact is probably not appreciated by the general public, or even the legislatures. So I think it’s worth a few words on why election audits differ from other routine audits such as banks, taxes, and casinos.

Elections involve anonymous casting of ballots. Once a ballot is cast, the link between the voter’s identity and their choices is gone. This contrasts with other formal auditing where a third party can confirm the transactions.
Elections lack auditor independence. Public company audits require a legally separate firm with appropriate licensing. Only 3 of the 51 jurisdictions require an independent 3rd party. In most cases, election audits are run by the people that ran the election.
Statistical assurance. Financial institutions and casinos have had confidence standards for decades that are almost uniform across these industries. This project found just 8 of 51 states run anything with an explicit, stated numerical risk limit.
Election audit findings are generally not binding. Financial audits have real teeth. But only 12 of 51 jurisdictions have a mechanism by which the audit can control the election outcome and 12 states run “informational only” audits. There are 4 states where the audit was run after the election was certified.
Elections are one-shot and perishable. If banks or casinos have a less-than-stellar audit, they can fix their errors and they will be re-audited. Once an election is in the books, there is no looking back and most election records are held for only two years.
Election auditing has no widely used standards. Financial institutions and casinos are controlled by rules set by a small number of standard-setting organizations. The Election Assistance Commission (EAC) proposed voluntary auditing standards in early 2026 but they did not attempt to standardize tabulation auditing, are silent on statistical rigor, and have not been formally adopted. An open comment period revealed many shortcomings of the EAC document.

What determines how useful an audit is

Finally, here is a list of the various dimensions of an audit that determine overall utility of a tabulation audit:

Mandate: how state law specifies requirements or provides local discretion.
Detection capacity: what evidence is examined, paper ballots, ballot images, ballot marking device bar codes, etc. and which contests are audited; how the evidence is examined, by humans or machines.
Software independence: whether an undetected software bug could produce an undetectable outcome error.
Statistical power: whether the audit supports a supportable confidence level based on a rigorous mathematical derivation. Assigning a risk-limit is one such approach.
Diagnostic power: which kinds of error the audit can detect, how far it can pin a discrepancy down and find its cause, whether it confirms the outcome, and whether it checks the audited unit against the canvass.
Selection integrity: how the races are picked, how the sample is drawn, and when, whether all ballot-casting methods (mail-in, precinct, vote-center) are included.
Consequence: whether the audit can change the result, its timing relative to certification, and when it must expand if a discrepancy threshold is triggered.
Independence of execution: who directs the audit, who counts, and who holds the ballots.
Execution: whether it was completed as designed.
Public verifiability: who can observe the audit, whether results are published and in what form.

These dimensions are not entirely independent from one another, but these are common topics among those that advocate for improved election audits.

The graphics linked from the main page attempt to compare states based on selected audit dimensions that clearly reveal the differences among states. By clicking on any state icon, the user can find details for each as well as links to source documents.

NCSL perspective: https://www.ncsl.org/elections-and-campaigns/post-election-audits

Ryan, T. W. (2020), “Risk-Limiting and Ballot Image Audits,” ch. 20 in B. E. Griffith and J. H. Young (eds.), America Votes! Challenges to Modern Election Law and Voting Rights, 4th ed., pp. 319–334, American Bar Association: https://catalog.jenkinslaw.org/record=b1067672

EAC draft audit regulations: https://www.regulations.gov/document/EAC-2026-0067-0001