Americans are heading to the polls by the millions today in what seems like the most important election in American democratic history. More than 57 million had already voted before the day began. All 435 House seats are on the ballot, voters in 35 states are choosing senators, and 36 states are electing governors.

However, long before the counts are finished tonight, the networks will start talking about exit polls, and the group chats will start reacting to them. Knowing what an exit poll measures, and what it doesn’t, makes election night a lot easier to sit through.

An Exit Poll Starts With a Clipboard Outside the Polling Place

Exit polls question people right after they vote, as they walk out of their polling place, according to the American Association for Public Opinion Research (AAPOR), the main professional group for U.S. pollsters. This year, the survey firm SSRS is conducting the exit polls for a group of news organizations that includes NBC News, ABC News, The Associated Press, CBS News, CNN, Fox News, and MS NOW, NBC News reported on Tuesday. The outlets design the questionnaire jointly and receive identical data, and each one analyzes and reports the results on its own.

In randomly chosen precincts, interviewers count voters as they leave and approach them at fixed intervals, such as every third or fifth person, all day long. The paper form has about 25 questions about whom people voted for, which issues mattered most, and who they are, including their age, gender, race, and education. Voters fill it out themselves and drop it in a box, so their answers stay as secret as their ballots, and respondents can answer in English or Spanish.

Since so many people now cast their ballots early or by mail, the exit poll also includes a pre-election survey, conducted by mail, phone, text, and online, and interviewers later follow up to confirm that those respondents voted and whom they chose. In all, the national exit poll will include more than 75,000 voters, and about 4,000 voters will be surveyed in each of 16 states, including Texas, Arizona, Georgia, and Florida.

A Spoonful of Soup Can Tell You How the Whole Pot Tastes

Surveying somewhere between 1,000 and 1,500 people may suffice to gauge opinion across the whole country. However, that precision tends to shrink in tight races and smaller areas.

“If you think about a big pot of soup, you don’t have to eat the entire pot of soup to tell me the soup is spicy,” Mileah Kromer, then an associate professor of political science at the University of Maryland, Baltimore County, told CBS Baltimore in 2024. “You just need one spoonful, and that’s what we’re talking about in terms of random sampling; those 1,000 Marylanders are a spoonful of Marylanders.”

In a random sample, every person has a known chance, ideally an equal one, of landing in the survey, according to Pew Research Center. Because some groups answer surveys more readily than others, pollsters then use a technique called weighting, counting some characteristics, like age, race, gender, education, and region, a little more or a little less so the sample mirrors the population. People without a college degree, for instance, are less likely to take polls than college graduates, so their answers are “weighted up” to reflect their real share of the population, Pew explained.

Polls that let people sign themselves up online, often for gift cards or similar perks, are less accurate than random samples, and the gap is widest for Latinos, young adults, and other minority groups. In one experiment, Pew asked people in an opt-in survey if they held a license to operate a nuclear submarine, and 24% of respondents who said they were Latino answered yes, compared with 2% of non-Hispanics. Pew’s research suggests that some insincere respondents claim to be Hispanic or young in the hope of qualifying for more surveys and the rewards that come with them. Pew pointed to a widely covered opt-in survey from December 2023 as a likely case, since it reported that 20% of adults under 30 and 12% of Hispanics agreed that “the Holocaust is a myth.” When Pew asked young adults the same question on its own random-sample panel, only 3% agreed.

A careless poll, in other words, can misrepresent what our community thinks. Quinnipiac University, for its part, conducts its interviews in Spanish when a respondent prefers it.

The First Exit Poll Numbers Tonight Are a Rough Draft

Because the mix of voters shifts over the course of the day, an exit poll sample “is not representative of the entire electorate until the survey is completed at the end of the day,” according to AAPOR. For that reason, the major news outlets have agreed to hold back any exit poll numbers that could hint at a winner until every polling place in that state has closed. NBC News said it won’t share exit poll numbers of any kind until 5 p.m. ET, and it updates and reweights the data throughout the night as interviewers call in new responses.

Just under half of the voters approached by exit poll interviewers agree to take part, and those who say no can differ from those who agree. After the count, pollsters reweight the exit poll to the real results, so the final data can serve what AAPOR calls its “primary and most important purpose”: showing how different groups voted and why. Similarly, NBC News described its own use the same way, saying its Decision Desk draws on part of the exit poll data when it projects races, but that “the exit poll is primarily used to help tell the story of the election.” According to AAPOR, exit polls are only one ingredient in those projections, which also lean on statistical models, past turnout, and, in close races, real vote totals from precincts and counties.

That said, we must keep in mind every poll also comes with a margin of error.

In Caltech’s example, if a poll shows Candidate A at 51% and Candidate B at 49% with a margin of error of plus or minus 4 points, A’s true support most likely falls between 47% and 55%, and B’s between 45% and 53%. In a close race, the margin of error is often as large as or larger than the gap between the candidates, Pew noted. It also leaves out other kinds of error, such as poorly designed questions or biased responses. Polls conducted in the final stretch of a race have historically been off by under 2 points on average, according to a study Pew cited. Preelection polls overstated support for Democratic candidates in 2016 and 2020, then produced more accurate results in 2022. Over the years, polling errors have favored both parties.

Not Every Poll on Your Feed Today Is a Real One

Fake polls fabricate or tamper with data to distort how voters appear to feel, to change their behavior, according to an August analysis by the Brennan Center for Justice. Campaigns on both sides have at times publicized polls that put their own candidate ahead while keeping the methods to themselves. In 2024, for example, a doctored CNN graphic giving Kamala Harris a lead over Donald Trump among Texas early voters spread on social media, and some users took it as false proof that the election was rigged, even though CNN never aired it and the image was quickly debunked.

Ian Anson, a political scientist at UMBC, told CBS Baltimore in 2024 that people “might see a poll on Facebook that says, ‘How many of you like cats?’ and that is not quite a scientific poll.”

The Brennan Center suggests asking three questions of any poll: who conducted it, how it was conducted, and whether the methodology is public, and why it was released at that moment. If a poll’s full questionnaire isn’t available, readers should be skeptical, and if a question sounds loaded when read aloud, it probably is.

Polls Shape How Voters Think, Though Not Always the Way We Assume

Critics sometimes argue that polls push people to jump on the bandwagon, but the evidence for that effect is “mixed and complicated,” according to Pew. A 2012 working paper by Neil Malhotra of Stanford Graduate School of Business and David Rothschild of Microsoft Research found that while some voters switch sides to join a winning team, more of them use poll results as information. When participants answering questions about public policy were told that many experts backed a position, their views moved 11.3%, compared with 8.1% for a general poll showing majority support and 6.2% for the views of “people like me.”

“Majorities can cascade, which is not good if we want to preserve minority rights or worry about herding,” Malhotra told Stanford. “But we see that people are also looking for information and try to learn from the wisdom of crowds.”

Tonight, that information will arrive in pieces, from exit polls, early counts, and projections, and it will keep changing as more ballots are counted.

“We need to always remember, this is just a snapshot,” Anson said. “We’d love to know for certain, but ambiguity is the way we have to live our lives as scientists.”

For regular people, and for the whole world whose future hangs in the balance, that snapshot should not, under any circumstance, determine why we vote.

Tags: 2026 midterm elections, Exit Polls