Israel's pollsters and editors owe an explanation after failing voters - comment

The polling stations had been closed for ten minutes when N12, Channel 12's news site, told its audience what would happen next. President Isaac Herzog, it assessed, would ask Gadi Eisenkot, the former army chief challenging Benjamin Netanyahu, to form the next government. There was a caveat, as there usually is, about the actual results and the party leaders' recommendations.

Four hours later, the caveat had swallowed the headline. The channel's revised exit poll put Netanyahu's allied parties ahead of Eisenkot's camp, 52 seats to 49. The initial projections on Channels 12, 13, and the public broadcaster Kan had all placed the four main opposition parties in front. An evening of confident talk had been built on numbers that were already moving.

Israelis vote for parties competing for 120 Knesset seats. A coalition needs 61, and leading the largest party guarantees neither the president's invitation nor a government. A few seats decide who sounds like the winner and who holds the leverage.

By 2:52 p.m. on Wednesday, with 85.6% of the vote counted, Likud had 26 seats and Eisenkot's Yashar 21. Netanyahu's coalition parties held 53, the four Jewish-led opposition parties 48, and the Arab parties 19. Counting continues; neither camp has a majority, and opposition to Netanyahu doesn't make every party outside his bloc a willing partner.

All three outlets committed a serious error about the people journalists are paid to understand

Set that beside the final campaign polls. Channel 12 gave Likud 21 seats, Channel 13 gave it 19, and Kan 20. All three put Yashar first. They missed Likud by five to seven seats and named the wrong largest party, a serious error about the people journalists are paid to understand.

The disagreement was visible for months. In August, Channel 12 had Likud trailing Yashar 22 to 24. On September 17, Tzuriel Sharon's poll for i24NEWS gave Likud 27 and Yashar 21. In the final week, Channel 13 had Likud at 19 while Channel 14 had it at 30. Voters change their minds, and an August survey isn't a forecast. But an eleven-seat gap over one party, days before the vote, should have been a central campaign story.

Our own October 22 report led on Eisenkot's polling advantage and cited surveys putting Likud at 19. It explained that neither camp had a majority, but it should have made the disagreement among pollsters impossible to miss. I am responsible for what we publish. Most of our readers follow Israel from abroad and relied on us to show them how open this contest remained. That required no advance knowledge of the result.

Sharon and Shlomo Filber, the two pollsters who saw a stronger Likud, founded Direct Polls in 2019, with Filber as the face of its Channel 14 polling. In 2025, Sharon became sole owner and began polling for i24NEWS.

Sharon's final poll put Likud at 26, exactly where the count stands. His opening election-night estimate gave Netanyahu's five coalition parties 55 seats, two above their current 53 and closer than the other four channels. On X, he has since declared Direct Polls Israel's most accurate polling company, writing that "criticism and pressure are no reason to change a professional finding."

Filber, who took fire for his strong Likud numbers, ended the campaign at 30, closer than the three broadcasters. His opening exit poll, though, gave Netanyahu's bloc 60 seats, seven too many; the other channels' 50 was three too few. Reading a campaign well doesn't make election night right. Sharon's claim will need the completed count and a party-by-party comparison, since a close bloc total can hide errors that cancel out.

Before the vote, Sharon told the business daily Globes that his company recruits its own panel and keeps its analysis in-house, and that competitors' reliance on external panels helps explain their numbers. His account needs independent checking, but it raises the question editors should have pressed during the campaign: what is agreement among several polls worth if they may be drawing on similar pools of respondents? An editor should be most curious about the evidence that unsettles the story he has been publishing.

Jewish tradition has long been wary of counting people. The Talmud forbids counting Jews directly, even for a good purpose, and King David's census ended in a plague. Many synagogues still count a minyan, the quorum of ten for communal prayer, by reciting a ten-word verse, one word per person. The instinct seems sound to me. A count holds authority over the counted. A number on a screen tells people who they are and what is possible for them, and it rarely arrives with the hesitation it deserves.

I can't quantify how many Israelis changed their vote because of this coverage. Someone drawn to a small party may have moved to a larger one; someone else may have decided her candidate was finished. What can be examined is the information newsrooms gave voters and the confidence with which they gave it.

Polling companies should explain themselves

Once counting ends, polling companies should explain how their samples, turnout assumptions, and treatment of undecided voters produced these errors. A claim that voters moved late needs evidence of when and where they moved. Editors should publish those findings prominently, alongside their own account of the coverage, and not in a technical report few will read.

The 2:15 a.m. update corrected the numbers. The coverage that came before it still needs explaining, by the editors who approved it, myself included, before the next poll arrives.