Evaluating How the USCIS SAVE System Performs in Practice:
Evidence from Election Offices
The MIT Election Data and Science Lab helps highlight new research and interesting ideas in election science, including through research grants under our ongoing Practice-Oriented Research program.
Our post today comes from Jack Friedman and Wren Orey, of the Bipartisan Policy Center, based on their ongoing research funded by this program. The information and opinions expressed in this column represent their own research, and do not necessarily represent the opinions of the MIT Election Lab or MIT.
In May 2026, Utah reported the results of a “citizenship audit” of its voter registration rolls that it conducted in April. Of more than 2 million voter records analyzed, just 27 were found to be noncitizens, with another 25 likely noncitizens. The audit’s findings are noteworthy because they confirm what previous studies have consistently shown: that noncitizen voting, which is already illegal in every state for federal elections, is extremely rare. But the findings are notable for another reason too: they were made possible by a resource—the Systematic Alien Verification for Entitlements (SAVE) service, a program run by the U.S. Citizenship and Immigration Services (USCIS)—that was only recently made available to election officials.
After USCIS made a series of changes to SAVE last year, election offices in states across the country began adopting the tool with the purpose of identifying and removing noncitizens from voter registration lists (and preventing noncitizens from registering to vote in the first place). Until a federal court's ruling in League of Women Voters v. U.S. Department of Homeland Security in June 2026, which temporarily blocked the 2025 SAVE expansions (a September 25, 2026 Supreme Court ruling allowed them to resume), many states were using SAVE results to inform decisions about who can register—and who stays registered—to vote. These decisions were affecting millions of voters and potential voters. Yet little is known about how accurate SAVE is, or the administrative burden that it created for officials. Our research project is an attempt to fill that gap.
What is SAVE?
SAVE is an online service operated by USCIS that lets government agencies verify a person’s immigration or citizenship status. It was originally created in 1987 for benefits-granting agencies, such as those providing unemployment insurance or health care, to establish applicants’ eligibility.
Prior to 2025, election officials could, in theory, use SAVE to verify the citizenship status of voters. But in practice, they rarely did so. The following limitations rendered SAVE an impracticable voter registration and list maintenance tool.
- Non-federal agencies like election offices were charged usage fees for each query.
- Queries could only be made on one individual at a time.
- Each query required a DHS-issued immigration identifier, which election offices rarely possess.
- SAVE did not query Social Security Administration databases so did not return information on U.S.-born citizens.
But in 2025, USCIS made a series of changes to the SAVE program that made it a more viable resource for election officials:
- Usage fees for non-federal agencies were lifted.
- A bulk upload feature was added, enabling queries on thousands of records at a time.
- Social Security Administration and State Department databases were made available to SAVE queries.
- Queries were permitted using Social Security numbers (and SSN4), including those of U.S.-born citizens.
When these changes came online, election offices in many states quickly signed up to partner with USCIS to process voter records through SAVE. As of November 3, 2025, 26 states had established (or were in the process of establishing) memoranda of agreement for SAVE voter verification. As of this writing, 101 election offices across 29 states are registered, via a memorandum of agreement, with SAVE for voter registration and/or list maintenance. USCIS reported that state voting agencies had submitted more than 46 million voter verification queries as of October 2025.
In a matter of months, election officials went from using SAVE only on rare occasions to submitting large swaths of their voter registration records.
Research Questions
The widespread usage of SAVE for voter verification is unprecedented; in a sense, a live experiment is being conducted. But the results haven’t yet been systematically analyzed, as we are trying to do. We have four primary research questions.
1. How accurate is the system? The tendency of SAVE to generate false positives—voters flagged as (potential) noncitizens who are in fact citizens—is a well-documented problem. For instance, officials in Texas have described flagged voters who, upon investigation, were found to have already documented their citizenship to the state years earlier. We are asking which kinds of voters tend to be flagged; how often it turns out that flagged voters have already provided proof of citizenship; and, where no such proof has been recorded, how many voters supply it when asked.
2. What makes the system return a usable answer? SAVE does not respond only with "citizen" or "noncitizen." It frequently returns an inconclusive result, which means that election staff must manually investigate further. We are examining whether a conclusive response is more likely if, for example, a full nine-digit Social Security number is provided rather than an SSN4.
3. Under what conditions do voters respond to outreach? When voters are flagged by SAVE as potential noncitizens, election offices typically send them a mailed notice requesting proof of citizenship. The outcome (whether the voter’s registration is ultimately cancelled) often depends on whether and how the person replies. We are trying to determine how often people respond to the notices, what those responses look like, how many provide proof of citizenship, and whether the length of the deadline affects response rates. An office cannot determine who is flagged by SAVE, but it does determine how they notify voters, how much time they give voters, and how to respond to voter action or inaction.
4. What administrative burden does the process impose? Submitting lists to SAVE and managing the ensuing process uses election official time and expenses. Our project aims to document the administrative burden that the SAVE citizenship verification process imposes on election officials, and to identify potential ways to minimize that burden.
Methods and Initial Findings
To answer these questions, we seek to use already-public data, issue public records requests for more data, and hold interviews with election officials.
We have submitted public records requests to 24 jurisdictions across six states that hold agreements with USCIS to use SAVE for voter registration and/or voter list maintenance: Florida, Georgia, Kansas, Missouri, Texas, and Utah.
Our requests seek three categories of records:
- Query-level material. The lists of voters an office submitted to SAVE and the responses it received, with personal identifying information redacted.
- Case-investigation material. Written policies and procedures, the template notices mailed to flagged voters, tracking of who responded and how each case was resolved, and records of any registrations cancelled.
- Aggregate statistics. Counts of voters by SAVE response category, resolution counts, and outreach outcomes.
We have received records from several jurisdictions to date, and a number of records requests remain outstanding. While it is too early to draw conclusions, the records we have received suggest at least one consistent trend: voters flagged by SAVE as potential noncitizens are often removed from voter lists without a definitive determination about citizenship. In one county, for example, 55 of 84 flagged voters were removed, not because they were confirmed noncitizens, but because they did not respond to the mailed notice that the county sends to flagged voters.
To supplement what we learn from these requests, we are also analyzing publicly available reports from states participating in SAVE. Two are especially instructive.
- Utah’s lieutenant governor published a citizenship verification audit (discussed above) in May 2026 covering 2,069,640 voter records. Data from the Utah Driver License Division (DLD) confirmed citizenship for 96% of these records. The lieutenant governor’s office then used SAVE to review 71,256 of the records DLD data could not confirm. Of those, 8,917 could not be confirmed as citizens, and a manual review of those cases identified 27 confirmed noncitizens and 25 probable ones. Together these 52 confirmed/probable noncitizens represent 0.0025% of the total 2,069,640 records reviewed.
- In Texas, the secretary of state’s office processed their entire voter registration list through SAVE, which identified 2,724 potential noncitizens. On September 15, 2026, the secretary of state’s Office reported that a review of Department of Public Safety (DPS) records confirmed citizenship for 578 (21%) of the initial 2,724 flagged voters (citizenship was established for 504 of these 578 records after passport data, previously missing from SAVE, was added). That leaves 2,146 registered voters flagged as potential noncitizens, or 0.01% of Texas’ more than 18 million registered voters.
Several tentative conclusions follow from these reports. First, SAVE can help election officials to verify the citizenship status of many—perhaps most—voters. Second, it also works well as an informal audit of registration records, helping to identify individuals erroneously added to voter lists. These are typically people who indicated they were noncitizens on the relevant (usually paper) form, and were inadvertently registered to vote through no fault of their own. Third, however, proving that someone is not a citizen is difficult to do conclusively, and requires significant manual effort. Fourth, how well SAVE functions depends on coordination and data-sharing across state agencies. Typically, citizenship information is spread across multiple agencies; in Texas and Utah, DPS and DLD, respectively, possessed citizenship records that the election office had not previously consulted.
What Comes Next
Several of our records requests remain open, and we will analyze them as they arrive, alongside the records we have already received. We are also continuing interviews with election officials and reviewing whatever additional information states make public. Through this combination of records, interviews, and public reporting, we hope to produce evidence about how SAVE is and is not working for election offices, and to help answer some of the important questions raised by election officials’ expanded use of SAVE.
