SAVE America Act

A federal requirement to prove citizenship and identity when registering and when voting, brought into force five weeks before the midterm election.

AI evaluation · not yet reviewed by a human

This evaluation was produced and sourced by an AI model; a human review is still pending. Figures and conclusions may still change. The review log is at the foot of the page.How review works →

Every person registering to vote in a federal election, or changing their address or name on the rolls, would have to present documentary proof of citizenship in person. Voting in person would require photo identification, and mail voters would have to enclose a copy of one both when requesting and when returning a ballot. States would be obliged to match their rolls against a federal immigration database and remove records after notice. Existing registrants would keep their registration. This evaluation assumes enactment around 30 September 2026, leaving about five weeks before state registration deadlines, and looks four years ahead.

Balance

Much worse for the future · 0.06 previous scale

Balance on the previous scale. The Bilanz 2.0 simulation is not yet available for this evaluation. The category comes from the share of the debate on the pro side (r).

For 4 · 6 % Against 61 · 94 %

Size class: small Scale of this evaluation: Normalised Impact — unitless, calibrated to this topic. For comparison: one point here is worth roughly 20 million euro per year. One vote is valued at 1,000 euro in this evaluation, derived from what an American election cycle costs per ballot cast, raised by a factor for the legitimacy a ballot carries beyond its administration. How we score →

Arguments for

Arguments against

8 arguments evaluated · Scoring v1.3 Δ absolute −57

Arguments — For

3 arguments

One federal rule replaces twelve state regimes

2.7of 100

Twelve states now impose documentary requirements of differing scope, and two of them are in court over it. A single federal standard removes part of that divergence, though the statute hands the cure processes back to the states.

Value 6 · Administrative qualityImpact 1Plausibility 4.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is the working order of the administrative system that runs elections — staff time, court time, and the predictability of the rules for people who move between states. This does not touch anyone's existence or health, and it does not decide who wins. It affects how reliably and how expensively a public function is carried out. That places it well below the constitutional core and above mere convenience. The value is middling because a well-functioning election administration is a real public good, but the consequence stream here is friction rather than participation.

Without the measure, the map keeps fragmenting: twelve states require documentary evidence in at least some cases, eight of them in all cases ^[10], and courts have struck down parts of these laws in Kansas and, in May 2026, New Hampshire ^[9]. Voters who move between states meet different rules; registration drives operating across state lines carry twelve compliance regimes; litigation runs continuously. A federal standard removes that layer. What it does not remove is the layer beneath: the statute leaves the handling of name discrepancies and the notice process before removal to the states ^[1], so the same divergence reappears in the procedures that decide individual cases. The saving is therefore partial — administrative and legal friction in the low tens of millions per year rather than the elimination of a patchwork. The Impact is small because the friction removed is real but narrow, and because the measure recreates part of it one level down.

The mechanism is straightforward and needs no study: one rule instead of twelve is less to comply with. What is contested is whether the net effect is a reduction at all. The federal requirement is new for 38 states, so the transition itself generates a wave of litigation and administrative divergence that the steady state would later reduce. Within the four-year horizon, most of what is observable is the transition rather than the steady state. There is no precedent for federalising registration requirements at this scale, so nothing anchors the size of the eventual saving. The plausibility is low to middling: the direction is defensible, the magnitude is a guess, and over this horizon the sign itself is not secure.

evidence basis: Mechanism · P ceiling 5

Confidence in the result

1.2of 100

A large majority supports the requirement, and support of this size is itself a fact about legitimacy. But the same survey shows the majority thins once the cost to eligible voters is named, and the confidence gained on one side is partly lost on the other.

Value 6 · Trust in institutionsImpact 0.5Plausibility 4

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is whether people accept the outcome of an election they did not win, which is what makes a democratic system able to transfer power without force. That is a real public good and it sits close to the constitutional core, but the consequence here is diffuse and psychological rather than a concrete change in who can vote. It touches everyone lightly rather than anyone heavily. The value is middling: legitimacy matters greatly, but a marginal shift in stated confidence is a thin slice of it.

Without the measure, a substantial part of the electorate doubts that only citizens are voting, whatever the audits show. With it, that specific doubt has a documented answer. A February 2026 survey found 71 percent of registered voters supporting the act ^[1], which is a level of agreement rare in American election policy. The delta, though, is not that number. The same survey splits 54 to 46 when respondents are asked to weigh the requirement against eligible voters being turned away ^[1], and confidence lost among those who see the measure as exclusionary offsets part of the confidence gained among those who see it as protective. Booked net, the residual is small and positive rather than the headline figure. The Impact is small because what looks like a large effect is largely two movements in opposite directions.

Survey support is measured well; the link from a legal change to durable confidence in results is not. Polls capture stated approval of a policy, not whether people afterwards accept an election they lose. No study establishes that adopting a documentary requirement raises subsequent acceptance of results, and the states that already have one have not visibly diverged from the others on that measure. Running the other way, the loss side is equally unmeasured. Both directions rest on reasonable expectation rather than evidence, which caps how much confidence this argument can carry. The plausibility is low: the survey number is solid, and it is not the thing the argument needs.

evidence basis: Plausibility · P ceiling 5

Ineligible registrations and votes are prevented

0.1of 100

The measure closes a gate that is currently held by a sworn attestation rather than a document. What comes through that gate, on every audit that has looked, is tens to low hundreds of cases against rolls in the millions.

Value 9 · Integrity of the voteImpact 0.0Plausibility 8.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is who decides a federal election. An ineligible ballot cancels an eligible one, which is the same injury the measure's critics name on the other side, arriving from the opposite direction. That symmetry is deliberate here: both this argument and the participation arguments below sit on the constitutional core, at the same weight, so the comparison between them is a comparison of size and evidence rather than of values. Nothing about the smallness of the number reduces the weight of the value it touches. A society that treats election integrity as a constitutional matter treats one improperly cast ballot as serious. The value is high because the integrity of the franchise is a foundational good, and it is exactly as high as the weight given to the eligible voters on the other side.

Without the measure, citizenship at registration rests on a sworn statement backed by after-the-fact audits and, increasingly, voluntary database matching. With it, a document must be produced before registration completes. The question is how many ineligible registrations the document catches that the attestation does not. Every audit that has looked gives a similar answer: Georgia found 20 non-citizens among 8.2 million registrants, nine of whom had ever cast a ballot ^[1]. Kansas, running a comparable requirement, identified 39 non-citizens across nineteen years ^[6]. A count of the Heritage Foundation's own case database yields 77 instances of non-citizen voting nationwide between 1999 and 2023 ^[1]. Scaled to a roll of 211 million ^[2], and adding the removals mandatory database matching would produce, the central estimate is a few hundred prevented ineligible votes per election cycle. The Impact is small because the number of affected cases is small, not because each case matters little — one ineligible vote and one lost eligible vote weigh the same here.

This is the best-evidenced number in the whole evaluation, and what it establishes is how small it is. Three independent sources — a state audit in Georgia, nineteen years of enforcement data from Kansas, and a national case database assembled by an organisation that campaigns for the requirement — converge on the same order of magnitude ^[1]^[6]. They differ on the exact count, which is expected given different definitions and periods, but not on the scale. The one genuine uncertainty runs upward: audits catch what they look for, and a determined effort to register non-citizens at scale would not necessarily show up in them. That is why the upper band is set an order of magnitude above the central estimate. Against that, no such effort has ever been documented in the United States. That the effect exists is beyond dispute; that it is small is nearly as well established.

evidence basis: Precedent · P ceiling 8.5

Arguments — Against

5 arguments · top 3 shown

Eligible citizens blocked at the registration gate

35of 100

Two states have run this requirement long enough to count the people it stops: Kansas blocked 31,089 eligible applicants during its transition, and Arizona still carries 34,933 registrants who never supplied a document after twenty years. Scaled nationally that is roughly 120,000 eligible votes lost per year.

Value 9 · Right to voteImpact 6Plausibility 6.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

What disappears is a person's vote in an election that is held once and cannot be repeated for them. A citizen who moved house in September, files the address change in October and is told to appear in person with a birth certificate she does not have does not get a later chance at that election. The people this falls on are disproportionately those who move often, who married and changed their name, and who were born at home or in states with poor records — none of which has any bearing on their entitlement to vote. This is the same value stream as the integrity argument above, at the same weight, which is what makes the two comparable: one ineligible ballot admitted and one eligible ballot prevented are the same injury to the same good. The value is at the constitutional core because the franchise is the mechanism by which every other political right is defended.

Without the measure, registration runs almost entirely through channels that never see a document: motor vehicle offices handle 55 percent of applications, online registration 14 percent, mail 9.1 percent, and in-person filing 5.9 percent ^[3]. With it, every one of those transactions requires a document presented in person. How many eligible citizens that stops need not be estimated, because two states have run the experiment. Kansas blocked 31,089 eligible applicants against a roll of about 1.8 million in the three years its law operated ^[6]. Arizona, twenty years in and with a driver-licence pathway that satisfies the requirement automatically for most residents, still has 34,933 registrants who never supplied the document, out of 4.4 million active voters ^[12]. Applied to a national roll of 211 million ^[2], that is roughly a million people in the first year and a few hundred thousand each year after. Arizona also shows how many of them would have voted: its federal-only voters turned out at about 19 percent, against 79 percent for everyone else ^[12]. That leaves about 120,000 lost votes per year. The Impact is the largest in this evaluation because a large number of people meet a requirement they cannot satisfy, even though most of them vote rarely.

Two states have run a documentary requirement long enough to measure, and they measure different parts of it. Kansas shows what a transition costs, and the trial court found the law burdened qualified applicants while only nominally preventing non-citizen registration ^[6]. Arizona shows what remains after two decades of adjustment and a fallback ballot this proposal does not offer: still 34,933 people who never supplied the document ^[12]. Arizona also settles what used to be the weakest assumption here, how many of the affected would have voted, and the answer of 19 percent against 79 is well below what is usually assumed ^[12]. Both transfers err in a known direction. Kansas understates, because a national rollout on five weeks' notice allows less adjustment than Kansas had. Arizona understates too, because its voters at least receive a federal ballot while under this proposal they would receive none. What remains uncertain is how much of the national roll resembles these two states, which is why the range still spans a factor of seven. Good middling plausibility: two independent real cases agree on the order of magnitude, and both err toward the conservative side.

evidence basis: Precedent · P ceiling 8.5

Cost and disruption to election offices

11of 100

Fifty states and thousands of counties would have to reprogramme registration systems, retrain staff and handle in-person document checks, on ten days of federal guidance and while primaries are still running. No one has costed it.

Value 5 · Administrative costImpact 4.5Plausibility 5

▸ Show reasoning & sources ▾ Hide reasoning & sources

This stream is public money and public administrative capacity — funds that are raised from taxpayers and, once spent here, are unavailable elsewhere. It touches no one's rights and no one's health directly. Its weight in this evaluation is the standard weight for public expenditure, neither more nor less, and it is deliberately not raised because the spending happens in a constitutionally sensitive area. The value is moderate: public money always counts, and it counts the same here as anywhere else.

Without the measure, election offices run the cycle they have planned for two years. With it, they receive federal guidance within ten days of enactment ^[1] and must rebuild registration intake around in-person document presentation before the deadlines. That means software changes, new forms, staff training, longer counter queues, and a surge of provisional ballots to adjudicate afterwards. Money spent here is money not spent on poll worker recruitment or ballot processing in the same weeks, which is why the disruption matters beyond its price. No agency has produced a cost estimate; the figures in circulation are expert ranges in the tens to hundreds of millions of dollars ^[1]. Assuming a couple of hundred million once and fifty million annually thereafter gives roughly a hundred million dollars per year over the horizon. The Impact is small in money terms and its real weight is the timing rather than the sum.

That a national administrative change costs money is not in doubt; every figure attached to it here is. There is no fiscal note, no state-level estimate, and no comparable national rollout to transfer from. The band spans an order of magnitude, from thirty to two hundred and fifty million per year, and the true figure could sit outside it. What is better founded is the direction and the timing: the guidance window is fixed in the statute at ten days ^[1], and the primary calendar is a matter of public record. Low plausibility on the size, high on the fact that a cost falls in the wrong weeks.

evidence basis: Plausibility · P ceiling 5

Mail voters unable to supply identification twice

8.2of 100

Nearly a third of ballots arrive by mail, and each would need an identification copy at both the request and the return stage. No jurisdiction has ever run that, which is why this is the least predictable part of the measure.

Value 9 · Right to voteImpact 1.6Plausibility 5.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The consequence is the same as at the registration gate — a citizen entitled to vote does not have their ballot counted — and it lands on the same value. It falls hardest on voters for whom mail voting is not a convenience but the only practical route: people who cannot stand in line, who work shifts, who live far from a polling place. For them a failed mail ballot is not a fallback to in-person voting but the loss of the vote itself. The value sits at the constitutional core for the same reason as the argument above.

Without the measure, a mail voter requests a ballot, receives it, and returns it signed. With it, a photo identification copy must accompany both the request and the return ^[1]. Nearly one third of the 112 million ballots at the last midterm were cast by mail ^[3], about 37 million. Two document steps rather than none introduces two points at which a ballot can fail: no access to a copier or scanner, a copy that cannot be read, an identification document that expired. The statute provides an escape: a mail voter may instead give the last four digits of a Social Security number with an affidavit ^[1], which removes the hardest version of the problem for anyone who knows the option exists. Assuming a failure rate between 0.1 and 0.5 percent gives 37,000 to 185,000 lost votes in 2026 and a smaller number in 2028. The Impact is moderate because the exposed population is very large and the per-person obstacle, thanks to the affidavit route, is usually small.

This is the part of the measure with no precedent anywhere. States with strict identification requirements apply them at polling places; none has required an identification copy at two separate stages of a mail ballot at national scale. The estimate is therefore built from a mechanism and a plausible failure rate rather than from any observed case, and the assumed range of 0.1 to 0.5 percent is a fivefold span with nothing anchoring it. Two things pull the number down: the affidavit alternative, and the fact that habitual mail voters are on average older and better documented than the population as a whole. One thing pulls it up: whether voters learn of the affidavit route in five weeks depends entirely on state communication, which is the thing least likely to work under this timetable. Low to middling plausibility: the mechanism is clear, the size is a projection, and a projection with no case behind it cannot carry more.

evidence basis: Projection · P ceiling 6

Voters turned away at the polling place

4.5of 100

The requirement reaches every in-person voter, but this is the part of the measure the research has actually studied, and the studies find small effects. Most voters hold identification and those who do not can cast a provisional ballot and cure it.

Value 9 · Right to voteImpact 0.6Plausibility 8

▸ Show reasoning & sources ▾ Hide reasoning & sources

The same stream again — a citizen entitled to vote does not vote. The particular people involved are those least likely to hold current identification, which correlates with age, income and housing stability rather than with entitlement. That the number is small does not change what each instance is: a person turned away at the door of a public election. The value is at the constitutional core, unchanged by the size of the count.

Without the measure, 36 states already ask for some form of identification and the rest do not. With it, photo identification is required everywhere for in-person voting. This argument counts only the residual after the registration gate: a person already blocked from registering never reaches the polling place, and counting them twice would inflate the total. Between 85 and 95 percent of the voting-eligible population holds valid photo identification ^[8], and a voter without one may cast a provisional ballot and present identification within three days ^[1]. What remains is voters who lack identification, do not know about the provisional route, or cannot return within three days: an estimated 10,000 to 60,000 lost votes in 2026 and fewer in 2028. The Impact is small because the exposed population is enormous but the share who both lack identification and fail to use the cure is very thin.

This is the one component with a real research literature, and it argues against the largest version of this objection. A review of the field found modest turnout effects at most ^[11]. A panel study using administrative registration and turnout records rather than surveys found no significant negative effect of strict identification laws on either ^[7]. A replication of the most-cited contrary finding showed that correcting data errors produces positive, negative or null estimates depending on specification, so no firm conclusion survives ^[8]. The dissent in the literature is real and has not been resolved ^[11], which is why the estimate is not set at zero. Transferring these findings to a nationwide first-time rollout on five weeks' notice also weakens them, since every study observes states that had time to prepare. Good plausibility: several independent studies agree that the effect is small, and the direction of the remaining uncertainty is known.

evidence basis: Study · P ceiling 9

Citizens removed in error by database matching

1.8of 100

The federal immigration database that states would be required to match against refers about one record in five thousand, and county review confirms only a fraction of those. Making the process mandatory and national in the weeks before an election removes the time to sort them out.

Value 9 · Right to voteImpact 0.3Plausibility 6.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

A citizen removed in error learns of it, if at all, when they try to vote. The people this falls on are concentrated among the naturalised, who are being asked to prove something the state has already adjudicated in their favour, sometimes decades earlier. The consequence is the loss of a vote in a specific election, the same stream as the arguments above, with the added weight that the burden falls on a group defined by how they became citizens. That does not raise the value ranking, which is already at the constitutional core, but it is what the argument is about. The value is at the constitutional core because a wrongly removed citizen has lost the same thing as a wrongly blocked one.

Without the measure, states use the federal database voluntarily and at their own pace, with time to review flagged records before an election. With it, matching is mandatory, national, and compressed into the pre-election period. Of 49.5 million registrations checked to date, roughly 10,000 were referred, about 0.02 percent, and county review confirmed only a fraction of those as genuine ^[1]. Texas flagged 2,724 records out of more than 18 million and found citizens among them ^[1]. The system's known weak point is people naturalised after their record was created, whose file still shows their earlier status ^[1]. Applied across 211 million registrations ^[2], mandatory matching implies roughly 40,000 referrals per cycle; if 10 to 40 percent are erroneous and go uncured before election day, that is a few thousand to fifteen thousand citizens removed. The Impact is small because the referral rate itself is very low, and it is not zero because the notice process the statute leaves to the states has the least time to work exactly when it is needed most.

The error direction is documented rather than hypothesised: Texas found citizens among its flagged records, and reporting has traced the naturalisation-timing problem to its cause in how the records are built ^[1]. What is not documented is what share survives county review under time pressure, because the existing use of the system has run with time to spare. The estimate therefore transfers an observed error type to an unobserved operating condition. The range of 10 to 40 percent uncured is wide and unanchored. Middling to good plausibility: that erroneous removals happen is established by real cases, and how many survive a compressed timetable is inference.

evidence basis: Precedent · P ceiling 7

Summary

The problem the measure addresses is real but very small: every audit that has looked for non-citizens on American voter rolls has found them in the tens or low hundreds against registration bases in the millions, and Kansas found 39 across nineteen years of enforcement [1][6]. The measure would close that gap almost entirely, so on its own terms it works. That benefit is the best-evidenced number in the evaluation — three independent sources, including a database assembled by the requirement's own advocates, agree on its scale [1][6]. Against it stands a loss of between 150,000 and 600,000 eligible votes, resting on one strong precedent: Kansas blocked 31,089 eligible applicants while its law was in force, and the trial court found it burdened qualified applicants while only nominally preventing non-citizen registration [6]. The balance falls the way it does because of size, not because of evidence — the objection is roughly three orders of magnitude larger than the benefit, and it would still be larger if every uncertainty were resolved in the measure's favour. Notably, the parts of the measure with the largest exposed populations are not the parts doing the damage: the research on photo identification at polling places finds small effects [7][8][11], while the registration requirement, which touches far fewer people, carries almost all of the loss. One thing the public debate gets wrong in both directions: the requirement is close to party-neutral, since Democrats and Republicans are about equally likely to hold at least one qualifying document [5]. The hinge is the timetable, not the requirement: with the documentary rule commencing for the 2028 cycle instead of five weeks before this one, the largest objection shrinks by roughly an order of magnitude and the balance becomes genuinely close.

Outlook — effect over time

Much worse for the future · 0.06 previous scale

Sources

FactCheck.org: Q&A on the SAVE America Act. factcheck.org
U.S. Election Assistance Commission: Election Administration and Voting Survey 2024. eac.gov
U.S. Election Assistance Commission: Election Administration and Voting Survey 2022. eac.gov
Brennan Center for Justice: Millions of Americans don't have documents proving their citizenship readily available. brennancenter.org
Bipartisan Policy Center: Do documentary proof of citizenship requirements disadvantage one party more than the other?. bipartisanpolicy.org
Prism News: Kansas proof-of-citizenship voting law blocked 31,000 eligible voters. prismnews.com
Cantoni & Pons, NBER Working Paper 25522: Strict ID Laws Don't Stop Voters: Evidence from a U.S. Nationwide Panel, 2008-2018. nber.org
Grimmer, Hersh, Meredith, Mummolo & Nall, Journal of Politics 80(3): Obstacles to Estimating Voter ID Laws' Effect on Turnout. jmummolo.scholar.princeton.edu
Brennan Center for Justice: State proof-of-citizenship requirements and litigation. brennancenter.org
Ballotpedia: Voter registration proof of citizenship requirements by state. ballotpedia.org
MIT Election Data and Science Lab: Voter identification. electionlab.mit.edu
Votebeat Arizona: Who are the Arizona voters without proof of citizenship?. votebeat.org
OpenSecrets: Total 2024 election spending projected to exceed previous record. opensecrets.org

Last reviewed by Claude Opus 5 · August 25, 2026 · 1× AI, not yet reviewed by a human

August 25, 2026AI reviewClaude Opus 5First evaluation
Taken from the archiv/usa-source-2026-08/ delivery (revision final); derived fields recomputed.

Evaluations are produced with AI support and reviewed on a schedule for new developments; human passes are marked separately.How we review →