Explainer · Who voted

Exit Polls, AP VoteCast and the New Voter Poll: Who Voted, and How We Know

For years, election night had two competing portraits of the electorate: the networks' exit poll and the AP's VoteCast. In 2026 they've been merged into one survey. Here's what changed, how it works, and why its numbers move after election night.

Updated November 1, 2026 · Election Tracker Live editorial team · 6 min read · Election Day: Tuesday, November 3, 2026

Vote counts tell you who won. They don't tell you who voted for whom, or why. For that, news organizations survey voters, and those surveys shape how elections are explained for years afterward: which groups shifted, which issues mattered, what the result "meant."

The way those surveys are run has changed several times in the past decade, and 2026 brings the biggest change yet. Here's the history in brief, and how to read the numbers you'll see.

01The traditional exit poll

Since 2003, the national exit poll was run for the National Election Pool, a consortium whose members in recent cycles were ABC, CBS, CNN and NBC. Edison Research conducted the only national exit polls in the U.S. for the pool from 2004 on.

The core method: interviewers stationed at a sample of polling places hand voters a short, self-administered paper questionnaire as they leave. As more people voted before Election Day, the exit poll added interviews at early-voting sites and surveys of mail and early voters by phone, email or text.

NBC News described the 2024 version this way: interviewers at 600 polling places nationwide, for about 20,000 interviews; early voters reached by phone through voter lists in all 50 states; and in-person interviews at early-voting sites in Georgia, North Carolina, Nevada and Ohio. Results that could indicate a race's outcome were held until all polls in that state closed.

02AP VoteCast, 2018–2024

The AP and Fox News left the National Election Pool in 2017 and, with NORC at the University of Chicago, built a different kind of survey. AP VoteCast debuted in the 2018 midterms and was used in general elections through 2024.

Instead of intercepting voters at polling places, VoteCast surveyed registered voters over several days ending when polls closed. The 2022 midterm edition, for example:

• interviewed 120,896 registered voters from Oct. 31 to Nov. 8, 2022, about 96% of them online and the rest by phone;
• drew on a probability sample from state voter files, NORC's AmeriSpeak probability panel, and opt-in online panels;
• ran as 50 state-based surveys combined into national estimates;
• reported a margin of sampling error of plus or minus 0.5 percentage points for its 94,296 voters;
• was weighted to the certified vote count after the election was complete.

032026: one survey for everyone

In October 2025, the research firm SSRS acquired Edison Research, and the AP announced that ABC, CBS, CNN and NBC would join Fox News in using AP election data for 2026 and 2028. SSRS launched the Voter Poll for the November 2025 elections, describing it as merging "the methodologies of the Edison/NEP Exit Poll and AP VoteCast/FOX News Voter Analysis."

For 2026, according to the AP's explainer on what it calls the AP Voter Poll:

• Who runs it: SSRS, for the AP, ABC News, CBS News, CNN, Fox News Media, NBC News and MS NOW.
• Pre-election survey: about 75,000 voters in all 50 states, reached by phone, text, mail invitation and online panels (the SSRS Opinion Panel and the Verasight Verified Panel). It runs from Oct. 22 through Election Day; Alaska's began Oct. 19. Voters who haven't voted yet are recontacted on Election Day to confirm their choice and method.
• In-person survey: about 25,000 interviews at polling places in Alaska, Florida, Georgia, Iowa, Maine, Michigan, Minnesota, Nebraska, Nevada, New Hampshire, North Carolina, Ohio, Pennsylvania, Texas and Wisconsin.
• What's reported: the makeup of the electorate and voters' views and motivations. The AP says topline results, meaning overall support for each candidate or party, won't be reported, and that the poll is not a replacement for the vote count.
• Race calls: the AP says it doesn't call races based solely on the poll. It's one input alongside results, vote history, registration and turnout data, and demographics. In some lopsided races, it can be the final data point that lets the AP call a race when polls close. See how races are called.

04Side by side

Sources listed below. "About 100,000" combines the AP's figures of roughly 75,000 pre-election and 25,000 in-person interviews.

05Why the numbers move after election night

Surveys of voters are adjusted so that they match what actually happened. NBC described the 2024 exit poll as re-weighted as new data came in, and VoteCast was weighted to the certified vote count once the election was complete. So the share of a group that backed a candidate can look different at 9 p.m. on election night, the next morning, and weeks later.

Months after that come validated-voter studies, which check survey respondents against official voter files to confirm who actually voted. Pew Research Center's study of the 2022 midterm, for example, was published in July 2023. Those studies use smaller samples than election-night surveys but can confirm turnout, which helps when analysts disagree about which groups shifted.

06How to read them on election night

• Treat early-evening numbers as preliminary. They'll be adjusted as votes are counted.
• Watch subgroup sizes. Small groups in a single state have wide margins of error, even in a large national survey.
• Remember that in-person interviewing covers 15 states this year. Other states rely on the pre-election survey.
• Don't confuse it with a vote count. The AP says the 2026 survey won't report toplines for each candidate; winners come from the count.
• Compare across sources later. Validated-voter studies published months after the election are the best check on election-night narratives.

Related: why polls miss, who votes in midterms and what early-vote data can and can't tell you.