AI Election Statistics 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: October 26, 2026 | License: CC BY 4.0
Eight days before the November 3, 2026 midterms, 173.4 million voting-age Americans, or 64.3% of the 269.8 million adults in the 50 states and DC, live in a state with an election deepfake law in force. Axis Intelligence Research calculates that 96.4 million adults in 20 states and DC have none, including Pennsylvania, Illinois, Ohio and Georgia.
Quick Answer
Thirty-one states have enacted election deepfake laws, per the National Conference of State Legislatures (NCSL), and 30 remain in force after Hawaii’s was found unconstitutional. Axis Intelligence Research’s Deepfake Ballot Shield Rate (DBSR), the share of voting-age residents those laws cover, reads 64.3% as of October 26, 2026. At least 164 AI-made political ads have aired this cycle.
Key Findings
- Axis Intelligence Research calculates a Deepfake Ballot Shield Rate (DBSR) of 64.26%: 173,361,689 of 269,763,509 U.S. adults live in a state with an election deepfake law in force on October 26, 2026.
- Axis Intelligence Research finds that 96.4 million voting-age residents in 20 states and the District of Columbia have no election deepfake law, led by Pennsylvania, Illinois, Ohio and North Carolina.
- Five states that passed their first election deepfake law in 2026 (Louisiana, Maine, Maryland, Tennessee, Vermont) added 15.9 million adults and 5.91 points of DBSR coverage, per Axis Intelligence Research.
- In Axis Intelligence Research’s tally of Convovo’s 71 Senate and governor race maps, GPT-6.1 Sol picked the Democrat in 64 races and Grok 4.7 picked the Republican in 62.
- The Wesleyan Media Project identified at least 164 AI-generated or AI-enhanced political ads by September 4, 2026, worth just under $80 million, and 69% carried no AI disclaimer.
axis-intelligence.com/ai-election-statistics/
How Many States Have Election Deepfake Laws in 2026?
Thirty-one states have enacted laws regulating deepfakes in political messaging, according to NCSL’s statute table, updated October 13, 2026. Four of them (California, Maryland, Minnesota and Texas) prohibit certain political deepfakes; the other 27 require a disclosure. The count was 30 when Maryland signed its law in May 2026, per Public Citizen’s May 13 release.
NCSL counts only statutes that explicitly govern AI in campaigns and elections. Broader impersonation or deepfake laws that do not mention candidates or elections are left out, even if they could apply in an election.
Why NCSL says 31 and Public Citizen lists 33
Public Citizen’s state legislation tracker, updated October 21, 2026, lists an enacted election deepfake law in 33 states. Axis Intelligence Research matched the two lists state by state. They agree on 31 states and differ on two:
| State | Public Citizen | NCSL statute table | Why they differ |
|---|---|---|---|
| Massachusetts | Enacted November 2024 as a budget amendment, sunset February 2025 | Not listed | The provision expired |
| Pennsylvania | SB 649, enacted July 2025 | No enacted election statute listed; several election-specific bills pending | NCSL does not treat the enacted bill as an election deepfake statute |
| Result | 33 | 31 | Axis Intelligence Research |
Both lists still include Hawaii, whose law no longer applies. That is why “31 states” is the right count of laws enacted, and 30 is the right count of laws in force.
Which court rulings changed the map in 2026?
NCSL lists six federal cases. Three produced rulings this year, with very different reach:
| State | Case | Ruling | Effect on the law | Source |
|---|---|---|---|---|
| Hawaii | Babylon Bee, LLC v. Lopez | Found unconstitutional, January 2026 | No longer applies | NCSL |
| Montana | Accountability in State Government v. Knudsen | Preliminary injunction, September 16, 2026 | Blocked only against the plaintiffs, a former state legislator and his PAC; the judge declined to halt the law statewide | Reuters |
| Vermont | Poitras v. Clark | Preliminary injunction, September 8, 2026 | Blocked only for one creator and one video; broader challenge pending | VTDigger |
| California | Kohls v. Bonta | One provision found unconstitutional, 2025 | Legislature amended the law; disclosure rules remain | NCSL |
| Minnesota | Kohls v. Ellison; X Corp. v. Ellison | Pending, no ruling on the merits | In force | NCSL |
Judge Susan Watters found Montana’s law likely discriminates by viewpoint, because it allows AI-generated ads that elevate a candidate while barring deepfakes meant to damage one, as reported by Reuters. She still limited relief to the named plaintiffs, citing the Supreme Court’s 2025 limits on universal injunctions.
Axis Intelligence Research therefore keeps Montana and Vermont in the coverage count. A plaintiff-specific injunction narrows who can be pursued; it does not remove the state’s adults from the law’s reach. Readers who treat any enjoined state as uncovered can subtract Montana’s 913,041 and Vermont’s 535,049 adults, which lowers the DBSR to 63.73%.
Analyst note — Sarah Mitchell. The legal fight is not about whether a model can make a convincing fake. That question is settled. It is about whether a state can regulate a political message because of the tool used to make it. Hawaii lost that argument outright; Montana and Vermont lost it for the people who sued. All six cases NCSL lists turn on the First and Fourteenth Amendments, and every disclosure law in the table below is one lawsuit away from the same test. The coverage number is a snapshot, not a floor.
What Share of Voters Is Covered by an Election Deepfake Law? The Deepfake Ballot Shield Rate (DBSR)
The Deepfake Ballot Shield Rate (DBSR) is an Axis Intelligence Research metric: the share of U.S. voting-age residents (age 18 and over) who live in a state where an election deepfake law is in force. It weights each state law by the number of adults it reaches, so California’s 31.2 million adults move the reading far more than any small state can.
Baseline reading, October 26, 2026: 64.26%.
DBSR = 18+ population of states with a law in force ÷ 18+ population of the 50 states and DC × 100 = 173,361,689 ÷ 269,763,509 × 100 = 64.26%
Inputs: the 31 states in NCSL’s statute table, minus Hawaii. Population comes from the U.S. Census Bureau’s July 1, 2025 estimates of the resident population age 18 and older (table SCPRC-EST2025-18+POP, released January 2026).
| DBSR component | Value | Source |
|---|---|---|
| U.S. adults, 50 states + DC (July 1, 2025) | 269,763,509 | U.S. Census Bureau |
| Adults in the 30 states with a law in force | 173,361,689 | Axis Intelligence Research |
| DBSR, October 26, 2026 | 64.26% | Axis Intelligence Research |
| Adults with no election deepfake law in force | 96,401,820 (35.74%) | Axis Intelligence Research |
| Share of states with a law in force | 60.0% (30 of 50) | Axis Intelligence Research |
| Covered adults in prohibition states (CA, MD, MN, TX) | 24.01% of U.S. adults | Axis Intelligence Research |
| Covered adults in disclosure-only states | 40.26% of U.S. adults | Axis Intelligence Research |
| DBSR from laws enacted 2019–2025 only | 58.35% | Axis Intelligence Research |
| Points added by 2026 first-time laws (LA, ME, MD, TN, VT) | 5.91 | Axis Intelligence Research |
| DBSR if Hawaii’s law still applied | 64.69% | Axis Intelligence Research |
| DBSR excluding Montana and Vermont | 63.73% | Axis Intelligence Research |
The gap between 60% of states and 64.3% of adults is the main story in the table. Election deepfake law is concentrated in large states: California, Texas, Florida and New York hold 52.15% of all covered adults. One adult in four (24.01%) lives under a prohibition; the rest of the covered population is protected by a label requirement.
Election deepfake laws by state: coverage table
| State | Adults 18+ (July 1, 2025) | Share of U.S. adults | Law in force, Oct 26, 2026 | Type (NCSL) |
|---|---|---|---|---|
| California | 31,180,511 | 11.56% | Yes | Prohibition |
| Texas | 24,109,307 | 8.94% | Yes | Prohibition |
| Florida | 19,019,796 | 7.05% | Yes | Disclosure |
| New York | 16,097,036 | 5.97% | Yes | Disclosure |
| Pennsylvania | 10,488,801 | 3.89% | No election-specific statute | — |
| Illinois | 10,100,540 | 3.74% | No | — |
| Ohio | 9,368,603 | 3.47% | No | — |
| North Carolina | 8,838,026 | 3.28% | No | — |
| Georgia | 8,796,778 | 3.26% | No | — |
| Michigan | 8,065,114 | 2.99% | Yes | Disclosure |
| New Jersey | 7,557,289 | 2.80% | No | — |
| Virginia | 7,019,802 | 2.60% | No | — |
| Washington | 6,366,184 | 2.36% | Yes | Disclosure |
| Arizona | 6,026,503 | 2.23% | Yes | Disclosure |
| Massachusetts | 5,826,510 | 2.16% | No (2024 law sunset) | — |
| Tennessee | 5,739,349 | 2.13% | Yes | Disclosure |
| Indiana | 5,397,168 | 2.00% | Yes | Disclosure |
| Maryland | 4,928,480 | 1.83% | Yes | Prohibition |
| Missouri | 4,910,413 | 1.82% | No | — |
| Colorado | 4,792,358 | 1.78% | Yes | Disclosure |
| Wisconsin | 4,750,680 | 1.76% | Yes | Disclosure |
| Minnesota | 4,547,092 | 1.69% | Yes | Prohibition |
| South Carolina | 4,421,834 | 1.64% | No | — |
| Alabama | 4,075,161 | 1.51% | Yes | Disclosure |
| Kentucky | 3,590,081 | 1.33% | Yes | Disclosure |
| Louisiana | 3,568,234 | 1.32% | Yes | Disclosure |
| Oregon | 3,461,772 | 1.28% | Yes | Disclosure |
| Oklahoma | 3,165,587 | 1.17% | No | — |
| Connecticut | 2,970,201 | 1.10% | No | — |
| Utah | 2,616,637 | 0.97% | Yes | Disclosure |
| Nevada | 2,603,663 | 0.97% | Yes | Disclosure |
| Iowa | 2,518,739 | 0.93% | No | — |
| Arkansas | 2,416,023 | 0.90% | No | — |
| Mississippi | 2,295,720 | 0.85% | Yes | Disclosure |
| Kansas | 2,294,452 | 0.85% | No | — |
| New Mexico | 1,686,046 | 0.63% | Yes | Disclosure |
| Idaho | 1,557,631 | 0.58% | Yes | Disclosure |
| Nebraska | 1,538,757 | 0.57% | No | — |
| West Virginia | 1,421,798 | 0.53% | No | — |
| Maine | 1,170,629 | 0.43% | Yes | Disclosure |
| New Hampshire | 1,170,277 | 0.43% | Yes | Disclosure |
| Hawaii | 1,151,103 | 0.43% | No (found unconstitutional, Jan 2026) | Disclosure |
| Rhode Island | 916,867 | 0.34% | Yes | Disclosure |
| Montana | 913,041 | 0.34% | Yes (injunction for named plaintiffs only) | Disclosure |
| Delaware | 849,963 | 0.32% | Yes | Disclosure |
| South Dakota | 714,952 | 0.27% | Yes | Disclosure |
| North Dakota | 616,388 | 0.23% | Yes | Disclosure |
| District of Columbia | 569,575 | 0.21% | No | — |
| Alaska | 565,570 | 0.21% | No | — |
| Vermont | 535,049 | 0.20% | Yes (injunction for one plaintiff only) | Disclosure |
| Wyoming | 461,419 | 0.17% | No | — |
| Sources: adults, U.S. Census Bureau (SCPRC-EST2025-18+POP); law type and enactment, NCSL statute table (October 13, 2026); court status, NCSL, Reuters and VTDigger; shares, Axis Intelligence Research. |
Which large states have no election deepfake law?
Pennsylvania (10.5 million adults), Illinois (10.1 million), Ohio (9.4 million), North Carolina (8.8 million) and Georgia (8.8 million) are the five largest states without one in force. With New Jersey and Virginia, the seven account for 62.2 million adults, by Axis Intelligence Research’s sum of the Census figures. NCSL lists pending election deepfake bills in Pennsylvania and North Carolina, among others.
Analyst note — Sarah Mitchell. A state law does not stop a deepfake from reaching a voter; it changes who can be held responsible after it lands. For a model provider or an ad platform, the DBSR is a compliance map: 64% of the adult audience now sits under some state rule, and those rules differ on labeling wording, pre-election windows and who can sue. The 36% outside them is not a safe zone either. There, the default rule is the platform’s own policy.
How Many AI-Generated Political Ads Ran in the 2026 Midterms?
At least 164 political ads generated or enhanced with AI, worth just under $80 million, according to the Wesleyan Media Project’s September 4, 2026 release. The count covers broadcast TV, connected TV and digital ads, with spending estimated by AdImpact. Wesleyan calls it an undercount.
| Measure (through September 4, 2026) | Value | Source |
|---|---|---|
| AI-generated or AI-enhanced political ads identified | 164 (minimum) | Wesleyan Media Project |
| Spending on those ads | just under $80 million | Wesleyan Media Project |
| Ads with no AI disclaimer | 69% | Wesleyan Media Project |
| Spending on ads with no AI disclaimer | 78% | Wesleyan Media Project |
| Ads carrying an AI disclaimer | 31% | Axis Intelligence Research (100 − 69) |
| Count in Wesleyan’s May 2026 report | 33 ads, about $20 million | Wesleyan Media Project |
| Growth in ad count, May → September | 4.97× | Axis Intelligence Research (164 ÷ 33) |
| Growth in spending, May → September | 4.0× | Axis Intelligence Research (80 ÷ 20) |
| Average spend per AI ad (upper bound) | about $487,805 | Axis Intelligence Research (80,000,000 ÷ 164) |
Which sponsors used AI in political ads?
Wesleyan attributes 80% of identified AI ads and 83% of their spending to Republican candidates or pro-Republican groups, and 11% of ads and 14% of spending to pro-Democratic sponsors. The remainder comes from issue ads and independent candidates. These shares describe ads Wesleyan could identify as AI-made; they measure detected use, not total use.
Analyst note — Sarah Mitchell. The disclosure number is the one to hold. Thirty states have a law in force, yet 69% of identified AI ads carried no AI label. Part of that is legal: many state laws cover only deceptive depictions of real people, not an AI-generated backdrop or voiceover, and federal law sets no AI disclaimer rule for these ads, as Wesleyan notes. The rest is enforcement. Detection is still a human team reading ads one by one.
Which Platforms Require AI Disclosure in Political Ads?
Platform policy now does more of the labeling work than state law in the states without one. Three of the largest ad platforms take three different positions:
| Platform | Rule for AI content in political or election ads | Source |
|---|---|---|
| Google (Search, YouTube, Display) | Advertisers must disclose election ads with synthetic or digitally altered content that depicts real or realistic-looking people or events; Google auto-generates the label in some formats | Google Ads political content policy |
| Meta (Facebook, Instagram) | Advertisers must disclose when AI or digital techniques create a photorealistic image, video or realistic audio of a person saying or doing something they did not | Meta newsroom, November 28, 2023 |
| TikTok | Paid political advertising is not allowed; official election bodies may run participation ads | TikTok ads policy |
Google’s policy carves out edits that are inconsequential to the ad’s claims, such as cropping, resizing and color correction. Meta’s rule dates from a November 2023 announcement; Meta’s current policy page could not be retrieved for this update.
Do AI Chatbots Lean Toward One Party When Asked Who to Vote For?
When forced to name one candidate, two leading models split in opposite directions. In a study of nearly 500 House, Senate and governor races run by Convovo, a research studio led by Cornell professor Tapan Parikh, Grok 4.7 recommended the Republican in 92% of answers and GPT-6.1 Sol recommended the Democrat in 97%, as reported by TechCrunch on October 24, 2026. The two models agreed less than 10% of the time.
| Measure (Convovo study, via TechCrunch) | Grok 4.7 | GPT-6.1 Sol |
|---|---|---|
| Overall pick for one party | 92% Republican | 97% Democrat |
| Picks against a district’s lean | Republican in 47 of 72 deep-blue districts (65.28%) | Democrat in 57 of 67 heavily Republican districts (85.07%) |
| Search off, 30-race follow-up | Refused 27; picked the Democrat in 3 | Picked the Democrat in 24; declined 5 |
| Frequently cited sources | X (41% of answers); a Christian conservative voter guide (39%); Turning Point Action (23%) | Wikipedia, the Associated Press, Axios, Reddit |
Percentages in the second row are Axis Intelligence Research calculations from the district counts.
How did the chatbots pick in Senate and governor races?
Axis Intelligence Research tallied every Senate and governor race on Convovo’s published state maps, as displayed on October 26, 2026. Of 71 statewide races, GPT-6.1 Sol picked the Democrat in 64 (90.14%) and Grok 4.7 picked the Republican in 62 (87.32%).
| Race type | Races | GPT-6.1 Sol | Grok 4.7 |
|---|---|---|---|
| Senate | 35 | 30 Democrat; 4 independent or third party (Idaho, Montana, South Dakota, Nebraska); 1 split (West Virginia) | 30 Republican; 1 Democrat (Kansas); 3 split (Rhode Island, New Jersey, Colorado); 1 declined (Arkansas) |
| Governor | 36 | 34 Democrat; 2 Republican (New Hampshire, Wyoming) | 32 Republican; 1 Democrat (Maryland); 1 independent or third party (Rhode Island); 1 split (Colorado); 1 declined (Minnesota) |
| Statewide total | 71 | 64 Democrat (90.14%) | 62 Republican (87.32%) |
Source: Convovo AI Ballot 2026 state maps; tally and rates by Axis Intelligence Research. “Split” and “declined” follow Convovo’s own legend.
The statewide tally runs a few points below the overall rates TechCrunch reported for all races, which include the House. Both patterns held across very different states: GPT recommended Democrats for Senate in states such as Alabama, Oklahoma and Texas, and Grok recommended Republicans for governor in states such as California, Hawaii and Massachusetts.
How the study was run, and what it does not show
The researchers queried both models through their developer APIs with search turned on, using one prompt: name one candidate, without hedging. The consumer apps declined to answer that prompt. Each Grok API request also carried about 1,240 tokens of input the researchers did not write and could not see. OpenAI told TechCrunch the test used a developer API with a prompt that demanded a recommendation, which does not reflect how people use ChatGPT; SpaceXAI had not responded at publication. Convovo’s study page describes the method but does not itself list the percentages.
Analyst note — Sarah Mitchell. The headline numbers measure a forced-choice API call, not a voter’s chat session, and both companies’ consumer apps refused the question. What the study does show is that when retrieval is on, the model’s source mix drives the answer: Grok leaned on X and conservative voter guides, GPT on wire services, Wikipedia and Reddit. The search-off follow-up, where Grok mostly refused, points the same way. Partisan lean here is largely a retrieval property, which means it can change with a ranking tweak between now and November 3.
How many Americans ask chatbots about the election?
About one in ten. Pew Research Center found that 8% of U.S. adults say they are using AI chatbots to find election-related information this year, in a survey of 5,273 adults fielded October 12–18, 2026. The rate is 11% among adults under 50 and 4% among those 50 and older.
| What chatbot users look for (share of users) | Share | Source |
|---|---|---|
| Where candidates stand on issues | 76% | Pew Research Center |
| Accuracy of claims heard elsewhere | 71% | Pew Research Center |
| Candidates’ lives and careers | 65% | Pew Research Center |
| Suggestions for how to make vote choices | 40% | Pew Research Center |
| Registration, polling places, voting by mail | 40% | Pew Research Center |
Axis Intelligence Research estimates that 8% of the Census Bureau’s 269,763,509 U.S. adults is about 21.6 million people, and that roughly 8.6 million of them (8% × 40%) ask chatbots for suggestions on how to make vote choices. Both are estimates built on Pew’s rounded percentages.
Which 2026 Election Deepfake Disputes Have Reached Regulators or Courts?
The new state laws are being tested in real disputes this cycle. The cases below were compiled by Pluribus News and published by the Arizona Capitol Times on July 21, 2026, with later court outcomes added from the sources named.
| State | Dispute | Date | Source |
|---|---|---|---|
| Oregon | Secretary of State investigation into unlabeled AI videos of a member of Congress | July 2026 | Arizona Capitol Times / Pluribus News |
| Michigan | State Senate candidate seeks $150,000 in damages from an activist over alleged AI videos | May 2026 | Arizona Capitol Times / Pluribus News |
| Arizona | Congressional candidate sues a super PAC over unlabeled AI-generated billboards and signs | July 2026 | Arizona Capitol Times / Pluribus News |
| Montana | Complaints over AI-altered mailers showing two candidates with pride flags; on September 16, 2026, a federal judge barred enforcement against the PAC that sent them, but not statewide | May–September 2026 | Arizona Capitol Times; Reuters |
| Vermont | Federal judge bars the attorney general from pursuing a content creator over one AI video; the broader challenge is pending | September 8, 2026 | VTDigger |
| Kentucky | Labeled and unlabeled AI ads on both sides of a U.S. House primary | May 2026 | Arizona Capitol Times / Pluribus News |
| Hawaii | State election deepfake law found unconstitutional (Babylon Bee v. Lopez) | January 2026 | NCSL; Courthouse News Service |
What penalties do the newest laws carry?
Maryland’s law, effective June 1, 2026, makes knowingly or recklessly spreading a deepfake to deceive voters a misdemeanor punishable by up to $5,000, up to 5 years in prison, or both, under Chapter 444 of the 2026 Laws of Maryland. It prohibits the conduct outright rather than requiring a label. Louisiana took the disclosure route for phone campaigns: HB639, Act 559, requires disclosure of AI in telephone campaign communications from August 1, 2026.
Methodology
Axis Intelligence Research assembled this page between October 21 and 26, 2026. Every figure was read in its source document and logged as a row in ai-election-statistics.csv.
State law status. The list of laws comes from NCSL’s table of state statutes on deepfakes in campaigns (updated October 13, 2026), which gives each statute’s citation, year, type and penalties. We matched it state by state against Public Citizen’s tracker (updated October 21, 2026): the two agree on 31 states, and Public Citizen’s two extra entries are Massachusetts (sunset February 2025) and Pennsylvania (SB 649, which NCSL does not list as an election statute). We used NCSL’s prohibition and disclosure labels as published.
DBSR. DBSR = (sum of the 18+ population of states with an election deepfake law in force) ÷ (18+ population of the 50 states and DC) × 100. A state is “in force” if it appears in NCSL’s statute table and its law has not been found unconstitutional; Hawaii is the only state removed. Plaintiff-specific injunctions (Montana, Vermont) do not remove a state; excluding both gives 63.73%. California stays in because, per NCSL, its legislature amended the law after the 2025 ruling and its disclosure requirements remain. The denominator includes DC, which has no law. Population comes from the Census Bureau’s July 1, 2025 estimates, read from the official table (SCPRC-EST2025-18+POP, released January 2026); the 51 jurisdictions sum exactly to the published U.S. total of 269,763,509 adults. The DBSR uses residents age 18 and over rather than registered or eligible voters because it is the only measure published for every state on the same date; noncitizens are included in the base. The reading is recomputed whenever a law is enacted, repealed, found unconstitutional or enjoined statewide.
Convovo tally. Convovo publishes its Senate and governor results as color-coded state maps without a count. Axis Intelligence Research counted every tile on the maps as displayed on October 26, 2026, using Convovo’s legend (Democrat, Republican, independent or third party, declined, split). House results are shown on a grid of every House race that we did not count; for all races combined, we report the rates published by TechCrunch. Convovo’s prompts and API settings are described above.
Other calculations. Growth multiples are later value ÷ earlier value. Average ad spend uses Wesleyan’s total and minimum ad count, so it is an upper bound. Chatbot-user estimates multiply Pew’s national rates by the Census adult population; both measure U.S. adults, but Pew’s percentages are rounded. District pick rates divide TechCrunch’s reported counts.
Source levels. Census, NCSL, state legislatures, Pew, Wesleyan and Convovo are primary for their own figures. Public Citizen’s tracker is a compiled record of state bills from an organization that supports these laws. The overall Convovo rates come from TechCrunch’s report. Court rulings come from NCSL, Reuters and VTDigger; the dispute table comes from press reports and is marked is_primary = no. Meta’s political-ad rule comes from its November 2023 announcement and is flagged as older data.
About This Dataset
ai-election-statistics.csv holds 461 rows in long format, one observation per row: the July 1, 2025 adult population and share of the U.S. total for all 50 states and DC; each state’s election deepfake law status, NCSL law type and Public Citizen enactment year; the DBSR reading and every alternative reading; NCSL and Public Citizen counts; court rulings in Hawaii, Montana and Vermont; Maryland and Louisiana statute details; Wesleyan Media Project AI ad counts, spending and disclosure rates (May and September 2026); Pew Research Center chatbot-use rates (October 2026); Convovo results, both TechCrunch’s reported rates and each model’s pick in all 71 Senate and governor races, state by state; platform ad policies for Google, Meta and TikTok; and dated 2026 disputes.
Columns: row_id, metric, value, unit, as_of_date, geography, segment, source_org, source_document, source_url, retrieved_date, is_primary, axis_calculated, method_note. Values are raw numbers with units in their own column; dates are ISO 8601.
License: Creative Commons Attribution 4.0 (CC BY 4.0). Credit line: Axis Intelligence Research, AI Election Statistics 2026, 2026.
How to Cite This Page
APA: Axis Intelligence Research, & Mitchell, S. (2026, October 26). AI election statistics 2026: Deepfake laws cover 64% of voting-age Americans, $80 million in AI ads and chatbots that split by party. Axis Intelligence. https://axis-intelligence.com/ai-election-statistics/
MLA: Axis Intelligence Research, and Sarah Mitchell. “AI Election Statistics 2026: Deepfake Laws Cover 64% of Voting-Age Americans, $80 Million in AI Ads and Chatbots That Split by Party.” Axis Intelligence, 26 Oct. 2026, axis-intelligence.com/ai-election-statistics/.
Chicago: Axis Intelligence Research, and Sarah Mitchell. “AI Election Statistics 2026: Deepfake Laws Cover 64% of Voting-Age Americans, $80 Million in AI Ads and Chatbots That Split by Party.” Axis Intelligence, October 26, 2026. https://axis-intelligence.com/ai-election-statistics/.
AI and the 2026 Election: Questions Voters, Campaigns and Platforms Are Asking
Is it illegal to post an AI deepfake of a candidate in my state?
It depends on the state. Thirty states have an election deepfake law in force, covering 64.26% of U.S. adults on Axis Intelligence Research’s DBSR. Twenty-six of them require a label rather than banning the content. California, Maryland, Minnesota and Texas prohibit certain deepfakes; Maryland’s carries up to $5,000 and 5 years in prison since June 1, 2026.
Does a campaign have to label an AI-generated ad?
Under 27 states’ laws, a disclosure is the requirement, per NCSL, and on Google and Meta a label is required regardless of state. Google requires a disclosure for election ads with synthetic or digitally altered content, and Meta requires one for photorealistic AI depictions in political ads. Wesleyan found that 69% of 164 identified AI political ads ran without an AI disclaimer as of September 4, 2026.
Can ChatGPT or Grok tell me who to vote for?
The consumer apps declined when researchers asked them to pick one candidate. Through developer APIs, Grok 4.7 picked the Republican in 62 of 71 Senate and governor races and GPT-6.1 Sol picked the Democrat in 64, by Axis Intelligence Research’s tally of Convovo’s maps. OpenAI said the API test does not reflect how people use ChatGPT.
Why do Montana’s and Vermont’s deepfake laws still apply after judges ruled against them?
Both injunctions protect only the people who sued. In Montana, Judge Susan Watters blocked enforcement against a former legislator and his PAC on September 16, 2026, but declined to halt the law statewide, per Reuters. Vermont’s September 8 injunction covers one creator and one video, per VTDigger.
Is there a federal law on AI in political ads?
Not one that requires an AI disclaimer on these ads, as the Wesleyan Media Project notes. Disclosure rules come from the states and from platform policies. TikTok sidesteps the question by banning paid political ads altogether.
How many Americans use AI chatbots for election information?
Eight percent of U.S. adults, per Pew Research Center’s October 12–18, 2026 survey of 5,273 adults: 11% of those under 50 and 4% of those 50 and older. Axis Intelligence Research estimates that is about 21.6 million adults.
Why do some sources say 33 states have election deepfake laws and others 31?
Public Citizen’s tracker lists 33 states; NCSL’s statute table lists 31. The difference is Massachusetts, whose budget-amendment rule sunset in February 2025, and Pennsylvania, whose 2025 law NCSL does not classify as an election deepfake statute. Both lists include Hawaii, whose law was found unconstitutional in January 2026, so 30 states have a law in force.
How much did campaigns spend on AI-generated political ads in 2026?
Just under $80 million across at least 164 ads by September 4, 2026, per the Wesleyan Media Project, up from about $20 million across 33 ads in its May report. Axis Intelligence Research puts the growth at 4.97 times in ad count and 4.0 times in spending.
Related reading: Deepfake statistics · Deepfake fraud statistics · AI regulation tracker · AI in social media statistics · Social media regulation statistics · AI safety statistics · Connected TV advertising statistics