Elections in the USA and Exit Polls: How They Work

On election night, before the actual votes are counted, television networks already show who voted for whom, what was most important to voters, and how women, young people, or voters with higher education behaved. This data comes from exit polls, which are surveys conducted among people leaving polling places. Few have ever seen a pollster with a clipboard outside a polling place, yet the results can be surprisingly close to the final counts.

In this guide, we first explain how elections work in the United States: who organizes them, why the president is elected by the Electoral College, what midterms are, and how the media "declares" winners, even though official results come weeks later. Then, step by step, we show how an exit poll is created, why it can be so accurate, and when it has been wrong. Finally, we describe exit polls in Poland, including election silence, the 9:00 PM announcement, and late polls, as well as the rules for voting by the Polish diaspora from the USA in Polish elections and exit polls in other countries. We do not provide any results or forecasts for the 2026 elections; we only explain how the system works.

How Elections Work in the USA

A Pole in America usually expects that elections are governed by a central institution similar to our National Electoral Commission. Such an institution does not exist here. There is no single ballot for the entire country, no single voter list, nor a single office that announces the results on election night. Elections are organized by states, and in practice, primarily by counties and municipalities.

States and Counties, Not Washington, Organize Elections

The foundation is provided by the Constitution itself. Article I, Section 4 states that the time, place, and manner of holding elections for Congress shall be prescribed in each state by its legislature, although Congress may at any time by law make or alter such regulations. The state decides on registration, voting equipment, and mail-in voting. A county official, often called a county clerk or supervisor of elections, designates polling places, hires election boards, prints ballots, and counts votes.

The federal government has a limited role. Congress established a common day for federal elections and can change state rules by law, while the Federal Election Commission (FEC) oversees the financing of federal campaigns. However, the elections are conducted by thousands of local offices. Official results of state and local elections are published by state and local election offices, while the FEC provides aggregate official results of federal elections from past years, publishing aggregate reports every two years.

Tuesday After the First Monday in November

The formula sounds like a riddle: federal elections are held on the Tuesday after the first Monday in November. In practice, this means that Election Day always falls between November 2 and 8. In 2026, it will be on November 3. Congress established the common date for presidential elections in 1845 to prevent the outcome in one state from affecting the selection of electors in another, which was a real problem when states voted on different days.

Why Tuesday and why November? This is explained by the realities of agricultural America in the mid-19th century. November came after the harvest but before winter, when roads became impassable. Sunday was a day of worship, on Monday one could travel by wagon to the county seat, and on Tuesday vote and return before Wednesday, when farmers typically went to market. It is worth noting that Election Day is not a federal holiday. Only some states, including Delaware, Hawaii, Illinois, Kentucky, New York, and Virginia, provide a day off on this day.

Who Can Vote

In federal elections, only citizens of the United States who are at least 18 years old on election day and meet their state's residency requirements can vote. A Green Card is not sufficient. The government portal usa.gov states explicitly that foreigners, including permanent residents, cannot vote in federal, state, or most local elections. The exception is a few municipalities that allow foreigners to vote only in local elections.

Voting rights can be restricted by states. Some individuals convicted of serious crimes (felonies) lose their voting rights for the duration of their sentence or longer, and some states restrict the voting rights of individuals deemed incompetent. The rules vary in each state. Citizens living in U.S. territories do not vote in presidential elections.

Almost everywhere, one must first register as a voter, and the registration deadline is set by the state. According to the National Conference of State Legislatures (NCSL), 24 states and the District of Columbia allow registration on election day or during the early voting period. There is also one exception to the registration rule: North Dakota does not maintain a voter registration list at all. An eligible citizen comes with a valid form of identification and receives a ballot.

ID? It Depends on the State

To a Polish voter who always shows an ID to the commission, American rules may seem surprising. According to NCSL, 36 states have laws that request or require showing identification at the ballot box. The remaining 14 states and the District of Columbia verify voters' identities in other ways, most often by comparing signatures or other data with the registry. Before the election, it is advisable to check the rules on your state election office's website.

Each State Has a Different Ballot

Since elections are organized by counties, the ballot in Chicago looks different from that in Houston, and even different from that in the neighboring county. A single ballot usually contains multiple votes at once: for a congressperson, sometimes a senator and governor, for state legislators, judges, and also referendum questions. The method of casting a vote also varies. In many places, voters fill in boxes on a paper ballot, which is then read by a scanner; elsewhere, devices with screens are used. The state and county decide what equipment is available at the polling place.

Early and Mail-In Voting

"Election Day" in America has long been more of an "election period." According to NCSL, 47 states, as well as the District of Columbia, Guam, Puerto Rico, and the Virgin Islands, allow all voters to vote in person before election day. Only three states do not: Alabama, Mississippi, and New Hampshire, although even there some voters can vote by mail as absentee. Eight states–California, Colorado, Hawaii, Nevada, Oregon, Utah, Vermont, and Washington–along with the District of Columbia conduct all elections primarily by mail: ballots are sent to every registered voter by mail.

The scale of change is well illustrated by data from the Census Bureau. In the 2022 midterm elections, nearly one-third of voters (31.8 percent) voted by mail, compared to 23.1 percent in 2018. Almost half (47.1 percent) voted before election day, compared to 37.8 percent four years earlier.

Counting Votes and Why It Sometimes Takes Days

In the evening, after polling places close, election boards send results to the county office, and counties gradually publish them on their websites. The most excitement surrounds mail-in ballots. Each envelope must be checked before the ballot goes to the scanner.

NCSL compiles when officials can start this "processing" of ballots. In 44 states, the District of Columbia, Puerto Rico, and the Virgin Islands, this can be done before election day. Six states, including Pennsylvania and Wisconsin, allow it to start only on election day, in Pennsylvania from 7 AM, in Wisconsin after polling places open. Where there are many mail-in ballots and work cannot begin earlier, the full result appears later. Additionally, there are provisional ballots issued when a voter's eligibility is in doubt, and their validity is checked after voting. Slower counting is therefore due to regulations and the number of envelopes.

Certification, or When the Result Becomes Official

What we see on election night are unofficial results. They become official only after a canvass, which is an official summary and verification of all protocols, and then after certification, first at the county level and then at the state level. The deadlines again depend on the state. According to NCSL, in 15 states, counties must conduct a canvass within a week of the election, while others have two, three, or even four weeks. State deadlines are equally varied: from the Friday after the election in Oklahoma to the second Tuesday of December in Missouri. Recounts and court disputes can extend this calendar.

Presidential Elections and the Electoral College

The most difficult thing to explain in American democracy is this: in presidential elections, Americans do not formally vote for the president. They vote for electors, who then elect the president and vice president.

538 Electors and the Magic Number 270

Each state has as many electors as it has representatives in the House of Representatives, plus two, because that is how many senators it has. The House has 435 members, the Senate has 100, and the 23rd amendment to the Constitution granted three electoral votes to the District of Columbia, which is not a state. This totals 538 electors. To win, an absolute majority of 270 votes is needed.

According to the National Archives, in the 2024 and 2028 elections, California has the most electoral votes (54), followed by Texas (40), Florida (30), New York (28), and Illinois and Pennsylvania with 19 each. The least, three, are held by Alaska, Delaware, North Dakota, South Dakota, Vermont, and Wyoming, as well as the District of Columbia. These numbers are based on the 2020 census.

The Winner Takes All, with Two Exceptions

In 48 states and the District of Columbia, the "winner takes all" principle applies. The candidate who receives the most votes in a state, even if by just one vote, gets all of that state's electoral votes.

The exceptions are Maine and Nebraska, which use the congressional district method. Two "statewide" electors go to the winner of the entire state, and one elector goes to the winner in each congressional district for the House of Representatives. As a result, these states can split their votes. Nebraska did this in 2008, Maine in 2016, and both states in 2020 and 2024.

Faithless Electors and the 2020 Ruling

Neither the Constitution nor federal law requires electors to vote according to the results in their state. Therefore, from time to time, so-called faithless electors appear, who vote differently than they promised. In 2016, there were ten of them: seven votes were cast (four in Washington state, one in Hawaii, and two in Texas), and three attempts were invalidated in Colorado, Maine, and Minnesota. Faithless electors have never changed the outcome of an election.

The dispute over whether a state can punish such an elector reached the Supreme Court. On July 6, 2020, in the case of Chiafalo v. Washington, the court unanimously ruled 9 to 0 that a state can enforce an elector's commitment to support their party's candidate, and thus the selection of electors from that state. According to compilations of state laws, 33 states and the District of Columbia have provisions to prevent faithless electors from voting.

From December Voting to January 6

Voters cast their votes in November, but electors meet only in December, each group in its state, on the first Tuesday after the second Wednesday in December. Then, on January 6, Congress counts the electoral votes in a joint session of both houses. The Vice President presides, and their role is, as the National Archives puts it, "strictly ministerial," meaning purely formal.

After the disputes surrounding the 2020 elections, Congress changed the procedure. The Electoral Count Reform Act, which President Joe Biden signed on December 29, 2022, as part of the budget bill, explicitly stated that the Vice President has no power to resolve disputes over electors. It also raised the threshold for objections: instead of one senator and one congressperson, an objection to a state's votes must now be signed by at least one-fifth of the members of each house. The certification of electors is generally provided by the governor, and the date for the December meeting of electors has been moved forward by one day.

What If There Is a Tie 269 to 269

With 538 votes, a tie is arithmetically possible. If no candidate receives 270 votes, the newly elected House of Representatives chooses the president from among the three candidates with the most electoral votes. However, not every congressperson votes separately: each state delegation has one vote, and a majority of delegations is needed for selection. The Vice President is then chosen by the Senate from among the two candidates with the most votes, requiring 51 senators for selection.

The House has chosen a president twice. In 1800, Thomas Jefferson and Aaron Burr had a tie in the Electoral College, and the House needed 36 votes to select Jefferson on February 17. The 12th amendment was then adopted to avoid a repeat. In 1824, Andrew Jackson received the most electoral and popular votes but did not achieve a majority, and the House selected John Quincy Adams in the first vote.

When the Winner of the Popular Vote Lost

Since electors count, not the total votes nationwide, a candidate can receive more popular votes and still lose. According to historians of the House of Representatives, this has happened five times:

Year	More Popular Votes Received By	President Elected
1824	Andrew Jackson	John Quincy Adams (election by the House)
1876	Samuel Tilden	Rutherford B. Hayes
1888	Grover Cleveland	Benjamin Harrison
2000	Al Gore	George W. Bush
2016	Hillary Clinton	Donald J. Trump

The Census Shifts Votes Between States

The number of electors is not fixed. Every ten years, the census determines the population of states, and based on this, the 435 seats in the House of Representatives are allocated. This number was frozen by a law in June 1929 at the level established after the 1910 census. After the 2020 census, seven seats shifted among 13 states. Texas gained two, while Colorado, Florida, Montana, North Carolina, and Oregon each gained one. California, Illinois, Michigan, New York, Ohio, Pennsylvania, and West Virginia each lost one seat.

The National Popular Vote Interstate Compact

For years, there has been a debate about whether to replace the Electoral College with a direct election. Changing the Constitution would require huge majorities, so proponents of direct election devised a workaround: the National Popular Vote Interstate Compact. States that adopt it commit to awarding all their electoral votes to the candidate who wins nationwide. However, the compact will only take effect when states with a total of at least 270 electoral votes join.

As of September 2026, 18 states and the District of Columbia have adopted the compact, which together have 222 electoral votes, or about 82 percent of the required minimum. The most recent to join were Minnesota (2023), Maine (2024), and Virginia (2026). Therefore, 48 votes are still needed to activate it.

Midterms, or Midterm Elections

Today's voting, November 3, 2026, is known as the midterms, the elections held halfway through the president's four-year term. There is no presidential candidate on the ballot, but the stakes are high, as voters decide the composition of Congress, which will pass laws, approve the budget, and oversee the government for the next two years.

The Entire House of Representatives

The Constitution states that members of the House of Representatives are elected "every two years." The term lasts only two years, and in every election, whether presidential or midterm, all 435 members of Congress are elected. Each member represents one congressional district, and the candidate who receives the most votes wins in that district. States redraw district boundaries after each census (redistricting). Depending on the state, this is done by the legislature or a special commission.

One Third of the Senate

There are one hundred senators, two from each state, regardless of population. Wyoming has the same number of senators as California. Senators serve six-year terms and are divided into three classes. One class, or roughly one-third of the chamber, is elected every two years. In 2026, regular elections will include 33 seats from the so-called Class II, and there will also be special elections in Florida and Ohio, totaling 35 seats.

Governors, Legislatures, and Referendums

Midterms are not just about Congress. According to the NCSL and election summaries, in 2026, governors will be elected in 36 states and three territories. The ballots also include elections for state legislatures: according to the NCSL, there are 6139 out of 7386 state legislative seats up for grabs in 88 legislative chambers, which is about 83 percent. Voters will also decide on referendums (ballot measures). In September, the NCSL counted 151 such votes qualified for 2026, with 132 scheduled for November. Local offices, from mayors to school boards, are also on the ballots.

Why Midterms Matter

The outcome of the midterms determines whether the president will have allies or opponents in Congress. The Constitution grants the House of Representatives the exclusive right to impeach an official, while the Senate has the authority to try them, with conviction requiring a two-thirds majority of those present. Revenue laws, or taxes, must originate in the House, and without the consent of both chambers, the budget cannot be passed.

Presidents Usually Lose

The history of midterms shows a pattern: the party of the sitting president almost always loses seats in the House of Representatives. The American Presidency Project at the University of California, Santa Barbara, includes 23 midterm elections from 1934 to 2022. In 20 of these, the president's party lost seats in the House. It gained seats only three times: in 1934 under Franklin D. Roosevelt (9 seats), in 1998 under Bill Clinton (4), and in 2002 under George W. Bush (8). The largest losses were 72 seats in 1938 and 63 in 2010. In the last midterms, the president's party lost 41 seats in 2018 and 9 in 2022. Researchers propose various explanations for this phenomenon. However, it is a historical pattern, not a law of nature, and does not guarantee anything about the outcome of specific elections.

Fewer Voters than in Presidential Years

In midterm elections, significantly fewer people vote. According to the Census Bureau, which has been asking Americans about their voting participation for decades, in 2020, during the presidential election, 67 percent of eligible citizens voted. In the 2018 midterms, the turnout was 53 percent, 11 percentage points higher than in 2014. In 2022, the turnout was 52.2 percent, the second highest for midterm elections since 2000.

These numbers come from the Current Population Survey, where people self-report their voting participation. Other sources, calculated using different methods, provide slightly different values.

Primaries and Candidates

Before the Democratic and Republican candidates face off in November, each party must select its candidate. In Poland, party authorities decide on the lists. In America, in most cases, voters decide in primaries, organized by the states, usually in the spring and summer of the election year.

Open, Closed, and "Top-Two" Primaries

The rules are set by each state. The NCSL categorizes states into several groups. In eight states, primaries are closed: only individuals registered as members or supporters of the party can vote. Nine states have partially closed primaries, where the party can allow independent voters to participate. In eight states, primaries are open to voters not registered with any party, in five they are partially open, and in 15 they are fully open, meaning that voters can choose which primaries to participate in on the day of the primary.

Five states are categorized by the NCSL as having unusual systems. The most well-known are the "top-two" primaries in California and Washington. All candidates, regardless of party, run on a single ballot, and the top two candidates who receive the most votes advance to the November elections. Alaska uses a variation where the top four candidates advance to the final.

Caucuses

An older method of selecting candidates is through caucuses, which are meetings of party members held in schools, parish halls, or fire stations, where party supporters discuss and vote on-site. In some states and for some parties, caucuses still serve to select candidates, especially in the race for the presidential nomination. The specific rules are determined by the parties themselves.

Two Parties and Third Parties

American politics has been dominated by two parties, the Democratic and Republican parties, for a century and a half. This is often explained by the electoral system: when one candidate wins in a district by receiving the most votes, and one list collects all the electors in a state, small parties have a tremendous difficulty gaining any seats. Third parties and independent candidates exist but rarely win. Theodore Roosevelt in 1912, as a candidate for the Progressive Party, won in six states and finished second. This was the last time a candidate outside the two major parties finished second in a presidential election. George Wallace in 1968 won in five states, receiving 13.5 percent of the votes, and since then, no third-party candidate has won in any state. Ross Perot in 1992 garnered about 19 percent of the votes but did not win in any state.

Election Night: Who and How Announces Winners

In the evening, a graphic appears on television: candidate X wins in state Y. Many viewers think this is the result announced by the state. Nothing could be further from the truth. In the United States, there is no federal office that collects votes from across the country on election night and announces winners. What is shown on television are projections, and in journalistic terms, "race calls," made by editorial decision teams (decision desks).

AP Has Been Counting Since 1848

The oldest and most important "announcer" is the Associated Press. It has conducted its own vote counting in every presidential election since 1848, the election in which Zachary Taylor became president. In the 2020 general election, the agency announced winners in over 5000 contests, and in the 2024 primaries in over 2500.

A call by AP is not a prediction. The agency announces a winner only when it believes that the losing candidates can no longer catch up to the leader. AP uses vote counting results, data on how many and what types of votes remain to be counted, and its own surveys of voters. The agency does not announce a winner until after the scheduled closing time of all polling places in that state.

When the margin is very small, AP waits. Since 2019, it has formally designated some contests as "too close to call." It may withhold an announcement when the two leading candidates are separated by less than half a percentage point. It also happens that AP announces a winner before a recount if the vote margin is too large for a recount to change it. For example, in Wisconsin in 2020, Joe Biden led by less than a percentage point but by over 20,000 votes, and the recount gave him only 87 more votes. AP declares a standard of 100 percent accuracy. In the last general elections, the accuracy of its calls exceeded 99.9 percent.

Television Decision Teams

Major television networks, including NBC, ABC, CBS, CNN, and Fox News, also have their own decision teams. Each makes decisions independently, which is why one station can announce a winner in a given state several minutes or hours earlier than another. Networks use vote counting results, statistical models, and voter surveys, including exit polls.

Florida 2000: The Night Television Made Two Mistakes

The most famous lesson in humility for decision teams was the night of November 7-8, 2000. According to NBC News, its team first awarded Florida to Al Gore around 7:49 PM Eastern Time, and at 9:55 PM, CNN retracted that projection, followed by other networks. Around 2:16 AM, Fox News announced George W. Bush's victory in Florida, and within minutes, the other networks did the same. Gore called Bush to concede, but soon after, he retracted that concession. Around 4 AM, the networks also retracted their projection for Bush and declared that the outcome could not be determined. The rest is a recount, disputes over manual counting, and finally the Supreme Court ruling in Bush v. Gore on December 12, 2000, which by a vote of 5 to 4 effectively ended the manual recount. Bush's final official margin in Florida was 537 votes.

Recounting Votes

When the result is very close, a recount comes into play. The rules are again state-specific. According to compilations based on NCSL data, about half of the states have provisions for automatic recounts when the margin falls below a statutory threshold, most often half a percentage point, and in some states, a quarter point. The cost of such a recount is borne by the state or county. In states without an automatic threshold or where the margin is larger, the losing candidate or voters can request a recount. The aforementioned example from Wisconsin in 2020 shows that a recount can shift the result by only a few dozen votes.

Official Results Come After Weeks

Announcements from election night, television graphics, and winners' speeches have no legal force. Legal authority comes only from the certification of results by county and state offices, and in presidential elections, also from the electors' voting in December and the counting of their votes by Congress in January. AP announces that a party has taken control of a chamber of Congress only when it declares 218 winners of that party in the House of Representatives or the appropriate number in the Senate. In many states, counting ends after a few days, and the official result comes after several weeks.

What is an Exit Poll

An exit poll is a survey conducted among people who have just cast their votes. The name says it all: the interviewer waits at the exit of the polling place and asks those leaving to fill out a short questionnaire. The American Association for Public Opinion Research (AAPOR) describes it as follows: in randomly selected polling places, at least one interviewer stands who, according to a predetermined key, approaches every third or fifth voter leaving, provides them with a questionnaire on a clipboard, and the voter fills it out themselves, folds it, and drops it into a special ballot box.

A pre-election poll asks whether and for whom someone intends to vote, so it always contains an element of guessing. NBC News, in an article from November 3, 2026, puts it this way: pre-election polls predict whether respondents will vote, while exit polls capture the views of those who actually voted. AAPOR adds that talking immediately after leaving the polling place largely eliminates the problem of people who claim in a phone survey that they voted when they did not. Exit polls also capture those who decided at the last minute. According to exit polls from 2016, 13 percent of voters decided who to vote for only in the last week before the election.

What Exactly Does an Exit Poll Measure

A typical questionnaire collects three types of information. First, who the person just voted for. Second, their views: which issues were most important to them, how they assess the state of the country, which party they identify with. Third, demographic data: age, gender, race or ethnicity, education. NBC reports that in this year's nationwide survey, voters are asked whether they voted for the Democratic or Republican candidate for the House of Representatives in their district, and in many states, also about the Senate and gubernatorial elections. There is also the classic question about the most important problem facing the country.

The result itself will be counted by the election commissions. The exit poll aims to answer the question of why it is what it is: who came to vote, how young and older voters, women and men, graduates and non-graduates voted. NBC states that its Decision Desk, the team announcing winners, uses some data from the exit poll in projections, but primarily it serves to tell the story of the election.

Why Exit Polls Can Be So Accurate

How can tens of thousands of surveys say something about over a hundred million voters? The answer lies in several overlapping layers of statistics.

• Random Selection of Polling Places. The polling places where interviewers stand are not chosen arbitrarily. Researchers randomly select them as a so-called stratified probability sample. The Pew Research Center described in 2016 that nearly a thousand points were selected from over 110,000 polling places in the country, and the process of selecting polling places began over a year before election day.
• Systematic Selection of Voters. Interviewers do not approach those who look friendly. They count those leaving and approach every n-th person, for example, every third or fifth, throughout the day from opening to closing of the polling place. This way, every voter at that polling place has a similar chance of being included in the sample.
• Large Samples. This year, NBC announces a nationwide sample of over 75,000 voters.
• Adjustment for Refusals. Interviewers record the approximate age, gender, and race of those who refuse, so it is known who is missing from the sample, and adjustments can be made.
• Weighting to Actual Results. When official results come in from the commissions, the exit poll data is adjusted to match them. We will return to this topic, which is very important for a fair assessment of accuracy, later on.

How Exit Polls Are Created Step by Step

An exit poll is a large logistical operation that begins long before election day and ends after the last votes are counted. We describe it step by step.

Step 1: Random Selection of Polling Places

Researchers divide polling places in the state into groups, or strata, and randomly select polling places from each of them so that the sample reflects the diversity of the entire state. The methodological statement of the National Election Pool for the 2024 Republican primaries states simply that the survey was conducted using a stratified random sample of polling places within the state. AAPOR emphasizes that historical results from selected polling places, such as past turnout and vote distribution, are then used in models. This allows for comparison on election night to see if more or fewer people are voting in the same locations than usual and whether the results are shifting toward one party or the other.

The scale can vary. In 2004, the state samples from Edison and Mitofsky International included 1,480 polling places. Joe Lenski from Edison Research, who has been working on the national exit poll since 1988, stated in a 2021 interview with the American Enterprise Institute that in presidential and midterm elections, interviewers typically face 700–1,000 polling places. This year, NBC reports that interviewers will appear at over 450 voting locations in 15 states with key Senate races, with at least 30 randomly selected polling places in each state.

Step 2: Interviewer at the Polling Place

On election day, the interviewer stands in front of the designated polling place, usually from its opening. They must adhere to local regulations that specify how close to the entrance they are allowed to stand. This is not a trivial matter. AAPOR reminds us that media outlets have sued over half a dozen states that tried to push interviewers back 100 feet (about 30 meters) or more, and each time courts have overturned these restrictions. However, in some states, distances of up to 75 feet were enforced, which clearly reduced the percentage of people willing to complete the survey. The Edison and Mitofsky International report summarizing the 2004 elections stated that when an interviewer must stand more than 50 feet, or about 15 meters, from the polling place, both the number of responses and the accuracy of the data decrease.

The interviewer receives what is called an interval, which is the number of people after which they should approach a voter. An interval of 1 means every person exiting, while an interval of 10 means every tenth person. According to AAPOR, this is predetermined based on the expected number of voters at that polling place. NBC, for example, reports intervals of every third or fifth person.

Step 3: Refusals Count Too

Not everyone wants to spend a few minutes with a stranger holding a clipboard. Lenski mentioned in 2021 that in the last decade, 40–45 percent of approached voters completed the questionnaire, while in 1988 it was around 60–65 percent. Pew reported around 45 percent in 2016.

What the interviewer does in the case of a refusal is crucial. They record the estimated age, gender, and race of the person who declined to respond, as well as those they did not manage to approach. The NEP methodology for 2024 states explicitly that these characteristics are estimated for non-respondents and used in the adjustment for non-response. If it is known that older voters tend to refuse more often, their responses in the sample receive correspondingly greater weight. Lenski recalled that this principle was developed by Warren Mitofsky in 1967 during the first exit poll prepared for the media.

Step 4: A Piece of Paper and the Secret of Voting

The voter receives a paper form. NBC reports that this year's form has about 25 questions. Other sources mention slightly shorter questionnaires: AAPOR writes about usually fewer than 25 questions and a completion time of under five minutes, while Lenski in 2021 described a form measuring 5.5 by 8.5 inches, or roughly A5, containing 18–20 questions. To ask more questions, several versions of the questionnaire are used: in 2020, the national exit poll had four versions.

The form is anonymous. The voter fills it out themselves and drops it into a box, so their answers, like their actual vote, remain confidential. In mail, phone, and online interviews, as NBC assures, no personal data is stored. The questionnaire can be completed in English or Spanish.

Step 5: Three Calls Throughout the Day

Interviewers do not wait until evening for results. According to the report summarizing the 2004 elections, they call data centers three times during the day: first shortly before noon local time, second in the late afternoon, and third in the last hour before polling places close. The exit poll is not complete until the data from the third call is received. NBC describes the same principle this year: results are transmitted to the data center three times during the day and combined with data from the pre-election survey.

Step 6: Voters Who Are Not Present at the Polling Place

The biggest change in recent decades is early and mail-in voting. AAPOR reports that the share of voters casting ballots before election day rose from about 16 percent in 2000 to an estimated 42 percent in 2016. Lenski mentioned 10–15 percent in 2000 and 70 percent in 2020. Someone who mailed their ballot two weeks earlier cannot be approached at the polling place.

Therefore, today’s exit poll is actually a combination of several studies. In 2024, NEP conducted a survey among early voters based on lists of registered voters, conducted via landlines, cell phones, text messages, and emails, and in states with a high share of early voting, interviewers also stood at early voting locations. The results were combined to reflect the proper proportions of voters voting early and on election day. This year, NBC describes a pre-election survey conducted through mail invitations, phone and text interviews, and online surveys. Respondents are then contacted again to confirm that they indeed voted and for whom they ultimately cast their vote.

Step 7: Weighting and Adjusting Throughout the Night

The report summarizing the 2004 elections outlines what is included in the weighting: the probability of selecting a polling place and a voter, adjustments for the age, race, and gender of those who refused, the best estimate of the outcome in various regions of the state, and the share of votes cast early. NBC states that demographic data is used to ensure the poll reflects the composition of the electorate as well as the election results. As interviewers transmit additional data and commissions publish results, the poll is updated and adjusted in real-time.

The final stage occurs after the votes are counted. AAPOR states clearly that the results of the exit poll are then matched to the actual election results.

Margin of Error: How Much Trust to Place in the Numbers

Every poll is an approximation. If different polling places or voters were selected, the result would be slightly different. The margin of error indicates how large such variations can be purely due to sampling. NEP publishes a table for a 95 percent confidence interval. This means that out of the intervals defined this way, 95 out of 100 include the value that would be obtained by asking all voters using the same methods. For the 2024 Republican primaries, the table indicates, for example, that for a characteristic held by about half of the respondents:

Number of People in the Group	Margin of Error (plus or minus, in percentage points)
100	15
201–500	7
951–2350	4
over 5250	2

Illustrative Example: if in the state exit poll, 50 percent of about two thousand surveyed women declare a vote for a given candidate, the interval from 46 to 54 percent encompasses with 95 percent confidence the value that would be obtained by asking all voting women. However, when looking at a smaller group, for example, 300 people in one age category, the margin increases to about 7 points in either direction. NEP also notes that the error is larger for characteristics concentrated in a few polling places, such as racial minorities living in specific neighborhoods, and smaller for characteristics distributed evenly, like gender. Additionally, there are non-sampling errors, including refusals, which NEP indicates likely increase the overall error.

How Many People Are Asked in 2026

According to NBC, this year's nationwide sample will include over 75,000 voters surveyed at polling places, online, and by phone. In addition to the national survey covering all 50 states, there are state surveys in 45 states with important Senate or gubernatorial races. NBC lists 16 states where the sample is expected to be around 4,000 voters: Arizona, California, Florida, Georgia, Iowa, Maine, Michigan, Minnesota, Nebraska, Nevada, New Hampshire, North Carolina, Ohio, Pennsylvania, Texas, and Wisconsin. In Alaska, the sample is expected to include about 3,000 people and also includes interviews at early voting locations. In the remaining states with results, it will be from 500 to 3,000 voters.

However, NBC also states that interviewers will stand in front of polling places in 15 states, and the list of states with a sample of about 4,000 people includes 16 entries. The Associated Press lists 15 states with in-person interviews: Alaska, Florida, Georgia, Iowa, Maine, Michigan, Minnesota, Nebraska, Nevada, New Hampshire, North Carolina, Ohio, Pennsylvania, Texas, and Wisconsin. Arizona and California are not included, but Alaska is. According to AP, the pre-election survey runs in all states from October 22 (in Alaska from October 19) until election day and includes about 75,000 voters, with interviews before polling places adding about 25,000 conversations.

Who Conducts the Exit Poll in 2026

NBC reports that this year's exit polls are conducted by the research firm SSRS on behalf of a group of media: NBC News, ABC News, Associated Press, CBS News, CNN, Fox News, and MS NOW. All receive the same data, and each newsroom analyzes and reports it independently. AP confirms this in its explanation, although it uses a different name: the survey is called The AP Voter Poll and is conducted by SSRS for AP as well as ABC News, CBS News, CNN, Fox News Media, NBC News, and MS NOW. The name thus depends on who reports the results, but the data comes from one study.

This is a significant change, as just in 2024, two competing systems were in operation. According to AP, the Voter Poll combines methods from AP VoteCast, used in general elections from 2018 to 2024, with methods from traditional exit polls.

AP VoteCast: A Different Approach

Associated Press and Fox News left the exit poll consortium after the 2016 elections. AP explained its departure from the consortium due to the increasing share of early voting: according to AP, about 43 percent of voters voted before election day in 2018 and about 70 percent in 2020. The new survey, prepared by NORC at the University of Chicago for AP and Fox News, debuted in the 2018 midterm elections.

VoteCast did not place interviewers at polling places at all. According to the 2022 methodology description, it combined three sources: a random sample of registered voters selected from state voter rolls, contacted by mail and phone, interviews from a probabilistic panel called AmeriSpeak conducted by NORC, and non-random internet samples. In 2022, around 115,000 interviews were planned, conducted from October 31 until polling places closed on November 8. The non-random part was statistically matched to the random part, and after the elections, the entire sample was weighted to the actual number of votes. Lenski criticized in 2021 that the majority of the VoteCast sample is non-probabilistic and that this approach relies heavily on modeling.

History of Exit Polls in the USA

Warren Mitofsky and Kentucky 1967

Warren Mitofsky, a statistician who previously worked at the Census Bureau, is considered the father of the American exit poll. According to the Roper Center, in 1967 he joined CBS News and soon conducted a pilot exit poll in the Kentucky gubernatorial election. In 1972, Mitofsky led the first national exit poll, and in 1976 he co-founded a joint polling center for CBS and the New York Times. He headed CBS's election division from 1967 to 1990. He passed away on September 1, 2006, in New York.

The priority is not undisputed: Dutch sociologist Marcel van Dam claims credit for the exit poll from the elections in the Netherlands in February 1967.

1980: The Night That Changed the Rules

On November 4, 1980, NBC announced Ronald Reagan's victory over Jimmy Carter at 8:15 PM Eastern Time, based on exit polls. It was only 5:15 PM on the West Coast, and polling places in California were still open for nearly three more hours. According to reports, the projection was based on exit polls covering 20,000 voters. Carter himself conceded defeat before the polls closed in the West. Sources differ on the minutes: "Time" reported 9:45 PM, while other reports mentioned 9:50 PM or just before 10:00 PM Eastern Time.

Congress held hearings to determine whether early projections depressed turnout. A University of Michigan study, described in 1981 by the Christian Science Monitor, indicated that about one-fourth of those who planned to vote later that day withdrew after hearing the network projections or Carter's concession speech. In 1984, both chambers of Congress passed a non-binding resolution urging media outlets to voluntarily refrain from characterizing or predicting results before all polling places closed in a given election. A statutory ban was deemed impossible, as exit polls are protected by the First Amendment to the Constitution.

From Competition to a Joint Consortium

By the end of the 1980s, each network conducted its own exit poll. Lenski recalled that the 1988 elections were the last with true competition. After that, in the name of cost savings, stations joined forces. According to AAPOR, from 1989 to 1992, exit polls were conducted by the Voter Research and Surveys (VRS) consortium, which Lenski recalled was led by Mitofsky and Murray Edelman. In 1993, VRS merged with the News Election Service, an organization counting votes for networks since 1964, and thus the Voter News Service (VNS) was created. It included ABC, CBS, NBC, CNN, Fox News, and the Associated Press.

Florida 2000 and the Fall of VNS

The election night of November 7, 2000, went down in television history. According to Time, networks using VNS data declared Florida for Al Gore around 8:00 PM, then around 2:00 AM awarded it to George W. Bush, and about two hours later concluded that the result was too close to call. It is worth noting that the first announcement came before the polls closed in Florida's Panhandle, which is in the Central Time Zone. Ultimately, Bush's margin was less than half a percent, and the state had to conduct a recount.

In February 2001, executives from ABC, CBS, NBC, Fox, CNN, AP, and VNS testified before a House committee. According to CBS News, they argued that the errors were due to technical problems, not bias, and promised among other things not to declare a winner until all polling places in the state had closed. In November 2002, VNS failed again: a malfunction of the new computer system prevented the quick delivery of data, and state exit polls were not published. In January 2003, VNS was dissolved.

National Election Pool and Edison Research

In 2003, ABC, CBS, CNN, Fox, NBC, and AP formed the National Election Pool (NEP). Exit polls were taken over by Edison Research, initially together with Mitofsky International, and from 2006 on its own. The first study for NEP was the California gubernatorial recall election in 2003. Sources vary on the start year of this collaboration: AAPOR and Pew mention 2003, while other reports mention 2004, likely because the first presidential election Edison handled was in 2004.

2004: The Mystery of Kerry's Lead

On election day 2004, early, incomplete data from exit polls leaked online, and bloggers announced that John Kerry was winning. George W. Bush won. In a report dated January 19, 2005, Edison and Mitofsky compared the polls with official results in 1,460 polling places. On average, the polls overstated Kerry's lead by 6.5 percentage points. This was the largest such error in the last five presidential elections, greater than in 1992 when it was 5.0 points. The authors identified the main reason as Kerry's voters being more willing to complete surveys than Bush's voters. The report also pointed to the role of the interviewers' age and the distance they were kept from polling places.

2016 and the Education Problem

Lenski acknowledged that in 2016, exit polls faced criticism for errors in specific states. Individuals with college degrees were more likely to complete surveys, and this difference took on a partisan dimension. According to him, since 2016, weighting procedures have been changed to account for this, which has reduced, though not entirely eliminated, partisan differences.

From Edison to SSRS

After AP and Fox left in 2017, Edison also took over vote counting previously conducted for the AP consortium. On October 15, 2025, SSRS, a research firm from Philadelphia, announced its acquisition of Edison Research. On the same day, AP announced that ABC, CBS, NBC, and CNN would use AP data on results in the 2026 and 2028 elections. According to Inside Radio, two weeks later, SSRS launched The Voter Poll, which combines Edison and NEP exit poll methods with AP VoteCast and Fox News Voter Analysis methods. The first test was the elections in November 2025 in New York, Virginia, New Jersey, and California, with Joe Lenski as a co-director of the project. AP and Fox then returned to a joint study.

How the media use exit polls on election night

Nothing before the polls close

The principle that no winner is announced in a state until the polls close there was developed after 1980. On January 17, 1985, the three major networks, ABC, CBS, and NBC, voluntarily committed to not announcing projected results in any state until the polls closed there. This is described by the "Washington University Law Review," citing the "Wall Street Journal." Reuters stated more strongly in 2012: Congress required networks to promise that they would not use exit polls to indicate a winner in a given state or to describe who is leading while voting is still taking place there.

This year, NBC announced that it would not release any exit poll results before 5:00 PM Eastern Time on election night, and no results that could indicate the outcome of the race before the scheduled closing of all polls in a given state. AP reports that the first Voter Poll results will be available at 5:00 PM Eastern Time on Tuesday.

Quarantine room

According to AAPOR, since 2006, to prevent leaks, NEP restricts access to exit poll data to a small group of analysts in an isolated location until a few hours before the polls close. The weekly "Time" reported in 2008 that NEP representatives left quarantine at 5:00 PM Eastern Time. How exactly this looks in 2026 has not been publicly described.

Decision desk: exit poll is just one component

Winners are announced on television by teams called decision desks. AAPOR emphasizes that projections are only partially based on exit polls. Analysts use statistical models that combine survey data with historical results from sampled precincts, and in closely contested races, also with actual results first from those precincts and then from all counties in the state. The report summarizing the 2004 elections shows the caution of these rules: the first of several conditions for announcing a winner was that the leader's margin must be at least 2.6 times greater than the standard error of that difference, corresponding to a 99.5 percent probability that the leader wins, and that only within the margin of error.

In 2026, television networks use AP data on results, but each decision desk analyzes it independently. AP states that its team does not announce a winner solely based on the Voter Poll. It takes into account results, voting history, registration and turnout data, demographics, and other information. Only in races with a clearly one-sided history can the poll be the last element that allows announcing a winner at the moment the polls close. AP does not announce a result until all polls in a given jurisdiction have closed as scheduled.

“Too early to call” and “too close to call”

In reports, two similarly sounding terms often appear. AP uses the phrase “too early to call” when the counting of votes is still ongoing. The phrase “too close to call” has been a formal designation by AP since 2019 and appears only when all ballots have been counted except for provisional and late mail-in votes, and a winner is still not apparent. AP may also withhold announcing a winner if the margin between the top two candidates does not exceed half a percentage point.

Fair warning: why exit polls “become” accurate

AAPOR states explicitly that after the votes are counted, exit poll results are adjusted to match the actual election outcome. This is a deliberate process: it ensures that the data serves its primary purpose, which is to explain how different groups voted. However, this means that the alignment of the final numbers from the exit poll with the official result does not prove its accuracy, as they have been adjusted to that result. A better measure is what the data showed earlier. Therefore, numbers from early evening can change by morning, sometimes significantly, as in 2004.

Known limitations

• Refusals and differences in willingness to respond. Just under half of the approached voters complete the survey. AAPOR notes that if voters from one party refuse more often, the exit poll may overstate the result for the other party. Adjustments for age, gender, and race do not eliminate differences that the interviewer cannot see, such as opinions. Lenski stated in 2021 that some voters do not trust the media conducting the surveys and participate less frequently.
• Early voters. They cannot be asked before the polls, so they must be surveyed by other methods, which have their own weaknesses. Lenski pointed out that text messages more often reach younger voters, phone calls reach older voters, and emails reach middle-aged individuals.
• The debate over the “shy voter.” After 2016, there was discussion about whether some voters concealed their choice from interviewers. The AAPOR committee found that some Donald Trump voters revealed their choice only after the elections, which could have been due to either a late decision or hiding preferences. Other tests of this hypothesis, including comparisons of polls with and without an interviewer, did not provide evidence to support it. This discussion mainly concerned pre-election polls, not exit polls themselves.

Despite these limitations, exit polls provide, as NBC writes, the first insight into who is voting this year and what drives their choices. Lenski summarized it in 2021 by saying that the issue is not whether polls work, but what level of precision can be expected from them. Elections decided by one or two percentage points require accuracy that survey research has never been designed to achieve.

Exit poll in Poland

In the Polish election evening, there is a moment that the entire country waits for: precisely at 9:00 PM, a bar chart with results appears on the screens of TVP, TVN, and Polsat. These are not yet the data from the National Electoral Commission, but an exit poll, which is a survey conducted among people who just minutes earlier left the polling station. Since the presidential elections in May 2015, one company, Ipsos, has been conducting it for the three largest stations. During the first round of the presidential elections in May 2025, Ipsos's Chief Operating Officer Paweł Predko was responsible for the electoral exit poll in Poland for the 27th time.

Previously, it varied. In 2005, during parliamentary and presidential elections, TVN and TVP commissioned surveys from several centers: GfK Polonia, OBOP, and PBS. In 2010 and 2011, TNS OBOP conducted the survey for TVP. Only since 2015 have competing stations contributed to one common poll, and Ipsos cannot independently disseminate its results, as it is the stations that decide how and when to present them.

How it works in practice

Ipsos describes its method quite precisely. From the list of all voting districts published by the National Electoral Commission, polling places are randomly selected. Closed districts (hospitals, detention centers, social welfare homes) and districts abroad are excluded. The sampling is stratified: so that provinces, villages and cities of various sizes, as well as small, medium, and large districts are proportionally represented in the sample. According to Ipsos's website, the survey usually involves 800–900 polling places, and in the 2024 European Parliament elections, there were 845. In the 2023 parliamentary elections, as reported before the voting by Infor.pl, there were 900 polling places and about 1850 interviewers. In the first round of the presidential elections in 2025, the sample was smaller: 500 polling places out of more than 32 thousand across the country.

In front of each selected polling place, there are two, and in large districts, three interviewers. They do not choose respondents at their discretion. They approach every nth voter leaving, according to a predetermined step. Ipsos states that on average this is every tenth person, and according to OKO.press in 2025, in smaller polling places it was every sixth voter, and in larger ones every ninth. The goal was around 100 completed surveys per polling place, totaling just over 50 thousand. According to Ipsos, nearly 2 thousand interviewers work on a large exit poll in close to a thousand districts.

The most important aspect is how the question is asked. The interviewer does not ask aloud whom someone voted for. The voter receives a paper form, marks who they voted for, adds their gender, age, education, employment status, and how they voted previously, and then drops the paper into a sealed Ipsos ballot box. Participation is voluntary and anonymous. Interviewers enter the data into an application on their smartphones, which sends it to headquarters almost in real-time. They also count refusals, as this information is useful later when weighting the results.

The problem of refusals

Not everyone wants to fill out the survey. According to OKO.press, in the first round of 2025, about 18 percent of those approached refused, similar to the 2023 parliamentary elections, while in 2020 it was about 10 percent. Older individuals, residents of smaller towns, and those with lower education levels refused more often. Interviewers therefore record the estimated age and gender of the person who refused, and statistical models attempt to estimate how those who did not respond might have voted. When OKO.press recalled that in the 2019 European Parliament elections, the exit poll underestimated the result of PiS by 3.3 percentage points, Predko responded that the model proved insufficient at that time and was improved.

Why exactly at 9:00 PM

The answer is in the Electoral Code. Article 39 states that voting takes place continuously from 7:00 AM to 9:00 PM. Article 107 prohibits campaigning on the day of voting and for 24 hours before it. This is the popular "election silence." Article 115 adds that from 24 hours before the day of voting until the voting ends, no results of pre-election surveys or surveys conducted on the day of voting may be made public. This prohibition applies within the territory of Poland.

The penalty is exceptionally high. According to Article 500, anyone who publishes survey results during this time is subject to a fine of 500 thousand to a million zlotys. Therefore, stations that know well in advance what the survey data looks like only show them when voting ends. At 9:00 PM, the chairperson of the commission announces the end of voting, but people who arrived before that hour can still cast their votes.

When the silence lasted longer

There have been instances when all of Poland waited longer for results because voting was extended in one or two polling places. In the elections on October 21, 2007, in two polling places in Warsaw, at Cybisa and Lokajskiego Streets, there were no ballots available. The ballots had been printed in sufficient numbers and were waiting in reserve, but the commissions did not collect them in time. The National Electoral Commission postponed the end of the silence several times until it finally ended nationwide at 10:55 PM.

During the second round of the presidential elections on May 24, 2015, in a polling place in Kowale in the Cieszyn district, an 80-year-old woman who was waiting for a ballot fainted at 5:35 PM. Despite resuscitation efforts, she died. The polling place was closed for about an hour and a half, and voting there was extended. The National Electoral Commission extended the election silence nationwide until 10:30 PM, and only at that hour did TVN announce the first results from Ipsos.

It was different on October 15, 2023, in Wrocław. In commission number 148 in the Jagodno estate, 1,649 people had registered in the last days, while the original list counted 2,453 voters. Regulations allowed the chairperson of the commission to request additional ballots only after 80 percent of those received had been issued, which happened at 6:28 PM. People stood in line for up to five hours, and by midnight the line was 450 meters long, with voting ending at 2:45 AM. These voters were able to vote because they arrived before 9:00 PM. The head of the National Electoral Bureau, Magdalena Pietrzak, stated at that time that everyone who arrived at the polling place before 9:00 PM had the right to vote. In comparison, during the second round in 2025, the chairperson of the National Electoral Commission, Sylwester Marciniak, stated according to reports from the commission's conference that there were no queues outside the polling places, so there was no need to extend the silence.

Late poll, or the real numbers from some commissions

An exit poll is a survey. A late poll is something different. After 9:00 PM, interviewers do not go home; they wait in the selected polling places until the commission counts the votes and posts the protocol with the results. These real numbers are sent to Ipsos and gradually replace the survey data. Because they come from official counting, there are no issues with refusals, and they include all voters in that polling place, not just those selected for the survey.

In the first round of 2025, the first late poll, announced around midnight, combined survey data and actual results from about half of the 500 polling places. The second, at 2:20 AM, was based 90 percent on actual results from the commissions. The official results from the National Electoral Commission were announced on Monday after 9:00 AM. Late polls usually align very well with the final data from the National Electoral Commission, but the election results are determined and announced solely by the commission, in the form of a notice published in the Journal of Laws.

How accurate were Polish exit polls

Election	Exit poll Ipsos	Official result
Presidential 2015, I round	Duda 34.8, Komorowski 32.2	Duda 34.76, Komorowski 33.77
Presidential 2020, II round	Duda 50.4, Trzaskowski 49.6	Duda 51.03, Trzaskowski 48.97
Sejm 2023	PiS 36.8, KO 31.6, Third Way 13.0, Left 8.6, Confederation 6.2	PiS 35.38, KO 30.70, Third Way 14.40, Left 8.61, Confederation 7.16
Presidential 2025, I round	Trzaskowski 30.8, Nawrocki 29.1	Trzaskowski 31.36, Nawrocki 29.54
Presidential 2025, II round	Trzaskowski 50.3, Nawrocki 49.7	Nawrocki 50.89, Trzaskowski 49.11

In the first round of 2015, the exit poll accurately predicted Andrzej Duda's result but underestimated Bronisław Komorowski's support by about 1.6 points. In the 2019 parliamentary elections, Ipsos, as it states, accurately predicted PiS's result to within 0.1 percentage points. In 2020, with a result of 50.4 to 49.6, Ipsos stated that the difference was within the margin of error and could not indicate a winner.

The elections on October 15, 2023, had a turnout of 74.38 percent, the highest in all elections since 1989. The average error of the exit poll was, according to Ipsos, 0.78 points, and the late poll 0.3 points. In the first round of 2025, the average error for the seven leading candidates was about 0.3 points (the text stated 0.33 once and 0.34 once), and in the late poll at 2:20 AM – 0.13 points. Predko stated then that he did not know if exit polls had ever been so accurate. The late poll even reversed the order of two candidates: Szymon Hołownia surpassed Adrian Zandberg, and the National Electoral Commission confirmed this. The difference was 0.13 points, about 26 thousand votes.

Second round 2025: when the survey indicated differently than the ballot boxes

On June 1, 2025, at 9:00 PM, TVN24, Polsat News, and TVP announced the result of Ipsos's exit poll: Rafał Trzaskowski 50.3 percent, Karol Nawrocki 49.7 percent, with an estimated turnout of 72.8 percent. The stations immediately reminded that the margin of error in the exit poll is plus or minus 2 percentage points for each candidate. The lead of 0.6 points was therefore much smaller than this margin.

After 11:00 PM, the first late poll appeared, already based on data from selected commissions: Nawrocki 50.7, Trzaskowski 49.3. After 1:00 AM, the second late poll showed 51 to 49 for Nawrocki, with an error of plus or minus 0.5 points. The official announcement from the National Electoral Commission on June 2, 2025, stated: Karol Nawrocki 10,606,877 votes, or 50.89 percent, Rafał Trzaskowski 10,237,286 votes, or 49.11 percent. The difference was 369,591 votes. Turnout: 71.63 percent.

Thus, the exit poll indicated a different winner than the ballot boxes. Formally, the difference between the exit poll and the result was within the announced error. A high rate of refusals to participate in the survey was indicated as a possible reason, which is the same problem the center had already faced in the first round. This case illustrates well what an exit poll is: it is a very good tool when the differences are clear, but with a result close to 50 to 50, it cannot decide elections. OKO.press noted after the first round that a researcher can do everything perfectly, and still, it may turn out that they were wrong.

How Polonia votes in the USA

A Pole living in the United States can vote in elections for the Sejm and Senate, in presidential elections, and in elections to the European Parliament. Voting districts abroad are created by a regulation of the Minister of Foreign Affairs if at least 15 voters are present in a given location and the results can be promptly transmitted. In 2025, 511 districts were established abroad and 5 on ships.

To vote, one must register in advance with the voter roll maintained by the consul responsible for the place of residence. According to Article 35 of the Electoral Code, the application must be submitted no later than 5 days before the election day: on paper with a handwritten signature, through the online service of the Ministry of Foreign Affairs (eWybory), or by email to the consul as a scanned signed application. One must provide, among other things, their PESEL and the number of a valid Polish passport, as outside the European Union, the passport is the document based on which one votes. In 2025, the deadline for the first round was May 13, and for the second round, May 27.

Voting is only done in person, at the polling place. The Code excludes mail-in voting in districts abroad and on ships, and voting by proxy is also not allowed abroad. Polling places are open from 7:00 AM to 9:00 PM local time. If voting would end the next day according to Polish time, it is conducted the day before. Therefore, Poles in the USA, Canada, Mexico, and South America vote on Saturday: in 2025, May 17 for the first round and May 31 for the second.

Votes from abroad are counted by district commissions at consulates and in other locations, just like commissions in the country. Formally, foreign districts belong to the electoral district in which the Warsaw district of Śródmieście is located. In elections to the Sejm, this means that votes from Polonia count in the Warsaw district. In presidential elections, this does not matter, as one result is counted for the entire country.

How significant are votes from abroad

In the second round of 2025, in 511 districts abroad, Rafał Trzaskowski received over 383 thousand votes, or about 63.5 percent, while Karol Nawrocki received over 220 thousand, or about 36.5 percent. The turnout among those registered was 84.26 percent, clearly higher than in the country, which is not surprising, as one must register themselves to the roll abroad. In the United States, in 57 districts, the result was reversed compared to most European countries: Nawrocki received 28,070 votes (about 56.6 percent), Trzaskowski 21,479 (about 43.4 percent), with a turnout of 86.38 percent. A similar situation occurred in Canada, where Nawrocki garnered about 62 percent.

For scale (these are our calculations based on PKW data): the votes cast abroad for both candidates total about 604 thousand, out of nearly 20.85 million valid votes in the entire election, which is just under 3 percent. Trzaskowski's lead abroad was about 163 thousand votes, less than half of Nawrocki's lead nationwide. Therefore, these votes did not change the outcome, but in a very close election, several hundred thousand votes from abroad is a number that no campaign can ignore. It is also worth noting that Ipsos's exit poll does not cover polling places abroad at all, and the late poll uses results from the same sampled polling places in the country, so votes from Polonia only appear in the PKW data.

Exit polls around the world

Since the first exit polls in 1967, each country has developed its own style of election night and its own rules for publication. Here are a few examples that illustrate how differently the same issue can be approached.

United Kingdom: ten o'clock in the evening

In the British election night, the most important hour is 10:00 PM. Polling stations close, and at that very moment, BBC, ITV, and Sky News present a joint exit poll. Data is collected by Ipsos, and the forecast is prepared by a team of experts, including political scientist Sir John Curtice. In 2024, interviewers stood outside 133 polling stations across the UK from 7:00 AM to 10:00 PM, conducting 21,514 interviews, with over 550 people involved in the study. Selected voters fill out a replica ballot and anonymously drop it into the ballot box.

The British method has an important detail: where possible, interviewers return to the same polling stations as in previous elections. It is not just about how many percent a party received in a given polling station, but how its support has changed compared to the last time. This change is then transferred to similar constituencies. This makes sense in a system where 650 MPs are elected in single-member constituencies, and the number of seats is what matters most. In 2024, the exit poll gave the Labour Party 410 seats, and they won 412.

The most famous was the evening of May 7, 2015. Pre-election polls predicted a hung parliament, with no majority for anyone. The exit poll gave the Conservatives 316 seats, 10 less than needed for a majority, and the Liberal Democrats only 10 seats instead of the previous 57. Former Liberal Democrat leader Paddy Ashdown announced that if this result proved true, he would eat his hat. Ultimately, the Conservatives won 330 seats and an outright majority, while the Liberal Democrats won 8. The exit poll turned out to be closer to the truth than all the pre-election polls, and in 2017, the House of Lords established a special committee to examine political polling.

The British exit poll has also made mistakes. In 1992, two exit polls for different stations predicted a hung parliament, but John Major's Conservatives gained a majority. The reasons cited included the so-called shy Tory effect, weak demographic data, and poorly chosen polling points. Since 2005, BBC and ITV have joined forces, and since then, the results have usually been very close. In 2017, the exit poll was off by three to four seats for the two main parties (depending on whether the Speaker of the House was counted as a Conservative), and in 2019, it was off by three seats for the Conservatives. Publication is protected by law: under the Representation of the People Act 1983, announcing a forecast based on what voters say after voting before polling stations close carries a fine or a penalty of up to six months in prison.

Germany: Prognose at 6:00 PM, then Hochrechnungen

In Germany, polling stations close at 6:00 PM, and at that very minute, the "Prognose" appears on television. It is produced by two centers: Infratest dimap for ARD and Forschungsgruppe Wahlen for ZDF, which has been working for this station since 1974. The forecast is based on exit polls: throughout the day, interviewers randomly select exiting voters in front of randomly chosen polling stations and ask them to fill out a written survey, which includes questions about age, gender, education, employment, and religion, in addition to asking about their vote.

Germany has an additional problem: a large number of mail-in votes. The exit poll only includes those who voted in person, so Forschungsgruppe Wahlen must estimate the mail-in votes and combine them with the survey results. Later in the evening, "Hochrechnungen" appear, which are estimates based on the official counting of votes in randomly selected districts, which correspondents relay to the studio. With each passing hour, more votes are counted, making the estimates increasingly accurate. According to ARD, cited by CORRECTIV, in 2025, Infratest dimap observed public counting in 560 selected polling stations.

The forecast can be inaccurate for parties balancing on the 5 percent threshold. According to CORRECTIV, in the 2025 Bundestag elections, the first ZDF forecast gave the FDP and BSW results at the threshold level, while the preliminary results showed both parties below it. The Bundestag election law prohibits the publication of survey results from voters after voting before the end of voting. Violating this ban carries a fine of up to 50,000 euros.

France: at 8:00 PM not an exit poll, but the first count

The French have a different custom that surprises many observers. Their evening "estimations" at 8:00 PM usually do not come from exit polls. Before the 2022 presidential elections, the Commission des sondages obtained assurances from eight major research institutes (BVA, Elabe, Harris Interactive, Ifop, Ipsos, Kantar, Odoxa, and OpinionWay) that they would not conduct exit polls on election day.

So where do the results come from? Some polling stations close at 7:00 PM, and the last ones at 8:00 PM. The institutes have selected "bureaux tests," or test polling stations, where they observe the first count of ballots after closing. Based on this, they estimate the national result. The Commission emphasized in 2022 that with an hour's difference between the closing of the first and last polling stations, a reliable estimation cannot be made earlier than around 7:45 PM.

The publication ban stems from a 1977 law on polling: on the eve and on election day, no polls may be published or commented on until the last polling stations in mainland France close at 8:00 PM. The issue arises with media from neighboring French-speaking countries. In April 2012, some Swiss and Belgian media announced that they could and wanted to provide the first estimates as early as 6:30 PM, and the Paris prosecutor warned that anyone who disseminated results before 8:00 PM would be prosecuted, including foreign media and internet users. The Commission in 2022 appealed not to believe any "results" circulating online before 8:00 PM. A curiosity for the Polish diaspora: French citizens voting in embassies and consulates in both Americas also cast their votes the day before, on Saturday.

Mexico: conteo rápido instead of a survey

Mexico has taken a different path. The official electoral institute INE conducts a "conteo rápido," or quick count. In the presidential elections on June 2, 2024, 7,500 polling stations were randomly selected from over 170,000. After the polling stations close, INE employees record the results from these stations and relay them to headquarters. Estimates are prepared by the expert committee COTECORA and presented in the form of ranges rather than a single number. The list of selected polling stations is confidential, secured with three keys held by different individuals, so that no one knows it in its entirety and attempts to influence those specific stations.

Ukraine 2004 and Israel 2015: when exit polls become an event

In Ukraine, exit polls became part of grand politics. In the second round of the presidential elections on November 21, 2004, Viktor Yushchenko led in all exit polls, with a margin of 2 to 19 percentage points. However, the Central Election Commission announced the victory of Viktor Yanukovych, who, according to them, had 49.42 percent compared to Yushchenko's 46.69 percent. The Orange Revolution erupted, and on December 3, the Supreme Court annulled the results of the second round, stating that the scale of irregularities made it impossible to determine the true result. In the repeated vote on December 26, Yushchenko received 52.77 percent. Since then, exit polls have been cited as one of the tools to verify whether official results are credible.

In Israel, on March 17, 2015, three television exit polls showed an almost dead heat: Channel 1 and Channel 10 gave Likud and the Zionist Union 27 seats each, while Channel 2 gave Likud 28 against 27. After counting 99 percent of the votes, Likud had 30 seats, and the Zionist Union had 24. The difference between the two parties turned out to be six seats instead of zero.

Where exit polls are banned or restricted

In many countries where exit polls are allowed, the same rule applies: they cannot be published while voting is ongoing. The United Kingdom, Italy, France, Germany, India, and Singapore consider publication before polling stations close to be a crime or offense. Singapore has gone the furthest: it has banned exit polls due to voter privacy, and since 2015, the electoral office has published "sample counts" instead, which are results from counting a sample of ballots. A comparative study by ARTICLE 19 from 2003, made available by the OSCE, covered 16 countries. Among the countries without a ban on publishing pre-election polls were Australia, South Africa, and the United States, with South Africa prohibiting the announcement of exit poll results.

Estimates on election night also face challenges where many people vote differently than in person. In Austria, in the second round of the presidential elections in May 2016, Norbert Hofer led in the evening results without mail-in votes, but the votes sent by mail (over 885,000 such ballots were issued) tipped the scale in favor of Alexander Van der Bellen.

Curiosities

• In Wrocław's Jagodno, where in 2023 polling station no. 148 was open until 2:45 AM, in March 2024, four gnomes were placed in line outside the neighborhood council's office, holding ballots and pizza. This is a memento from election night when a local pizzeria delivered hundreds of pizzas to those waiting.
• A voting district abroad can be established wherever at least 15 Polish voters are present, provided that results can be quickly relayed to the country.
• In the second round of 2025, the highest turnout among Polish municipalities was in Krynica Morska – 88.38 percent, while the lowest was in the municipality of Lasowice Wielkie – 47.67 percent.
• In the same round, Poles in Vietnam supported Rafał Trzaskowski by 91 percent, while in Kosovo, 74.41 percent of voters chose Karol Nawrocki. In China, exactly 1,000 valid votes were cast for both candidates.
• The Polish diaspora is changing from election to election. In the second round of 2020, Rafał Trzaskowski received about 74 percent of the votes abroad, while in the second round of 2025, he received about 63.5 percent.
• In the second round of the Krakow mayoral elections in 2024, the Ipsos exit poll was accurate to within 0.06 percentage points.
• Before the polling station opens, the Polish commission checks whether the ballot box is empty, then closes and seals it. From that moment until the end of voting, the ballot boxes may not be opened.
• Trustees in polling stations in the country can record the work of the commission with their own devices, from checking the ballot box in the morning until signing the protocol.
• The Polish ballot must be printed only on one side, and the boxes for the "x" mark must be of the same size for all candidates. In the elections to the Sejm, the ballot can therefore consist of many permanently connected sheets.
• In 2005, the joint exit poll by BBC and ITV predicted a Labour majority of 66 seats. That is exactly what it was.
• Exit polls serve not only television stations. In Georgia in 2024, exit polls for private television stations showed results that were clearly different from the official ones, which critics considered a signal of manipulation. In Venezuela, during the 2004 referendum, the exit poll also diverged from the official result, but observers from the Carter Center did not confirm any fraud.
• In India in 2014, the Election Commission determined that exit polls could be shown on May 12 at 6:30 PM, after the last stage of voting. One of its members later announced that this would only happen on May 16, but the commission confirmed the original date the next day.
• In South Korea, like in Poland, three television stations – KBS, SBS, and MBC – use a joint exit poll. In 2017, its project was led by Kantar, and in recent years it has been conducted jointly by Ipsos, Korea Research, and Hankook Research.