An independent tool to explore and analyze data from every major poll for Israel’s 2026 Knesset election.
Compiled by Ron Levi, University of Toronto · Built, analyzed, and annotated with Claude Code
Work in progress — corrections, suggestions and advice are very welcome: ron.levi@utoronto.ca
Analysis & data
ℹ️ When is the election? — now fixed: 27 October 2026 · tap to read more
ℹ️ How Israeli elections work — 120 Knesset seats, 61 needed to govern · tap to read more
PM suitability — the "who is more suited to be Prime Minister" question (e.g. Netanyahu vs Eisenkot), shown as a % over time. This tracks who respondents rate as more suited to be Prime Minister — a separate question from how many seats a party wins. Pick a Matchup; the number in brackets is how many polls asked it (more polls generally mean a steadier line). Different pollsters word the question differently, so filter to one pollster for a strictly comparable trend.
Explore — the raw data, one row per poll × party. Search or use the filters to narrow it, click a column heading to sort, then Download saves exactly what you've filtered. The verification column shows how thoroughly each figure was checked — open the key below. Tip: filter to one pollster + one date to pull a single survey; leave filters empty to get everything. We list individual polls only — not other aggregators’ or models’ averages, so nothing is double-counted.

Download the full poll database (CSV) — the entire dataset in one file, one row per poll × party, rebuilt daily. Each figure is an individual poll by the named pollster — every row records the pollster, who commissioned it, the source, and our verification level. Compiled by Ron Levi (University of Toronto) from those published polls; free to reuse with attribution (CC BY 4.0). The Download button under the table below instead saves only the rows you’ve filtered.

What the “verification” column means

two_source_confirmed Confirmed against two independent sources — the strongest level (the N12/Midgam figures, cross-checked with Wikipedia).

primary_verified Independently checked, seat-by-seat, against the pollster’s own published release or a primary news report.

primary_confirmed Read directly from the pollster’s release.

aggregator_only Taken from Wikipedia’s poll-aggregation table, not independently re-checked.

conflict Sources disagreed; the value judged most reliable was kept and flagged here.

primary_manual Hand-entered by us from the pollster’s own report because it is newer than Wikipedia’s table (which lags the Hebrew press by a few days). Shown with an “awaiting confirmation” tag; it clears to two_source_confirmed automatically once Wikipedia carries the poll.

unconfirmed Not yet checked.

When a source poll doesn’t add to 120

Each poll is recorded exactly as its source published it, even where the seats don’t total 120 — for instance one Direct Polls survey (16 Jun) sums to 127 and one Lazar survey (11 Jun) to 117 (each is flagged in its notes column below). Nothing in this raw table is changed. But before any average is taken, every poll is rescaled to the 120-seat house, so a source’s over- or under-count cannot tilt the aggregates on the other tabs.

When the same poll appears under two labels — how we avoid double-counting

A single survey occasionally enters the source record more than once — the same Channel 13 poll was logged first under one polling firm (Maagar Mochot) and later another (Midgam) as the attribution was corrected, and some polls appear under two nearby dates (a date later revised, or a fieldwork-end date shown next to a publication date). After a full audit of the record in July 2026 we adopted a consistent method for these, so no single poll is weighted twice in the averages:

How we identify a duplicate. Two entries are the same survey when they carry an identical seat breakdown from the same commissioning outlet. A poll’s identity is keyed on the outlet that ordered it plus its numbers — not on the polling firm’s name or the exact date, since both can differ across sources for one real survey.

Which copy we keep. The best-sourced one — the copy with a real sample size and an independent, seat-by-seat check, over a copy taken only from an aggregator. The duplicate is dropped.

When we do not merge. Identical numbers that sit far apart in time (weeks or more) are left as separate polls, because a quiet race can genuinely produce two matching results; we flag those for a source check rather than assume. We merge only where the evidence is unambiguous — and say so here when we don’t.

Click a column header to sort. Scenario rows hidden by default.
Issue-specific polls — who the public trusts on each problem. Beyond the seat and PM-suitability numbers, pollsters sometimes ask who would handle X better — security, the economy, and so on. This is who came first on each. A single-poll snapshot (one date, not a trend yet); a leader can top the seats and still own no issue. Pollsters now ask often enough whether the prime minister is trusted to run the war that the answers can be followed over time, so they open the tab as a running series, ahead of the dated snapshots.

About this tracker

An independent tracker of opinion polls for Israel’s 2026 Knesset election. It gathers the seat projections published by the major Israeli pollsters into one database, and gives you tools to explore them — trends, blocs, a coalition builder, and an “if the vote were held today” simulation.

Where the numbers come from

Israeli pollsters publish their results (how many of the 120 seats each party would win) but not their raw surveys, and no official body collects them — so this site aggregates published results (a “poll of polls”) rather than re-analyzing raw survey data — a true meta-analysis would need the underlying surveys, which pollsters do not release. The backbone is the Wikipedia poll-aggregation page; each figure is cross-checked against the pollster’s own release or the Hebrew press where possible, and every number is labelled by how well it was confirmed. Coverage is ~100 polls from 8 pollsters since January 2026. It is checked for new polls automatically once a day; the header at the top shows the span of dates the polls cover. If a poll later disappears from Wikipedia, we keep our archived copy of it (flagged “archived”), so nothing we’ve captured is ever lost.

One survey, one entry — guarding against double-counting

Because this site pools published results, its central discipline is that each real survey is counted once. That is harder than it sounds: the same poll can enter the record under different labels, and counting it twice quietly tilts every average. A full audit of the record in July 2026 turned up three ways it happens:

The rule we settled on: two entries are the same survey when they carry an identical seat breakdown from the same commissioning outlet — a poll’s identity is the outlet that ordered it plus its numbers, not the polling firm’s name or the exact date, since both of those drift across sources. We keep the better-sourced copy (a sample size, an independent check) and drop the duplicate; where identical numbers sit far apart in time and the case is ambiguous, we keep both and flag it rather than guess. The point is not housekeeping: before this pass, a strongly anti-Netanyahu poll was being counted twice, which made the opposition’s lead look steadier than the once-counted data supports.

Which parties count as being “in the race”

Israel’s electoral threshold is 3.25% of the vote, worth roughly four of the 120 seats. A party below it wins nothing at all. This matters for an average of polls because the threshold has already been applied before we see a number: pollsters publish seat projections, not vote shares, so a party a pollster leaves out (or shows at zero) is one that pollster has already judged to fall short. We are never applying the threshold ourselves; we are deciding what to do when pollsters disagree about whether a party clears it.

A party is counted in the two-week average if either test passes: it is seated in at least half the window’s polls, or in at least half of the six most recent. Where a party is counted, we average only the polls that actually seat it, so a firm leaving it out lowers its chance of being counted rather than dragging its seat number toward zero.

The second test was added on 30 July 2026, and the reason shows what is at stake. Yoaz Hendel’s list (now Zionist Home) crossed the threshold partway through the window: it appeared in 7 of 17 polls, under half, so it was dropped from every average, yet three of the last four pollsters seated it at four or five. The first test alone counts how many polls seat a party and ignores which, so a rising party stays invisible for up to two weeks, exactly when it is newsworthy. The older test is kept alongside because the second alone would admit a single fresh outlier. Blue & White fails both, seated in 1 of 17 polls and by no pollster since 16 July. That is the point: one stale reading must not put a party on the board.

Two consequences worth knowing. Adding a party changes every other party’s seats, because the projection is rescaled to total exactly 120 — counting Hendel’s four seats moved the blocs from 54/55/11 to 53/58/9, without changing the finding that neither bloc reaches 61. And a party sitting near the line can enter or leave the board between refreshes; when that happens the shift is arithmetic, not a surge in support.

The same rule now governs the nowcast simulation, which had a related flaw: it decided membership after evening out house effects, so max(0, 0 − a negative lean) could invent a seat share for a party a pollster had placed below the threshold. That was enough to put Balad — which no pollster seats at all — and Blue & White into the simulation. Membership is now judged on the raw published seats in every panel: a house effect may move a party’s seat number, it may not decide whether the party is running.

How to read it honestly

What it does — and doesn’t — claim

The Nowcast is a snapshot of where things stand now if the polls are right — not a forecast of the October result. We deliberately do not: predict the eventual outcome (that needs a history of how Israeli polls have missed past results), grade or weight pollsters by track record (same reason), or break results down by demographics (that needs the raw surveys, which we don’t have). We can show how the published numbers move, how much they disagree, and what coalitions they imply. When we simulate, we let a whole bloc’s polls miss together, because real polling errors are correlated rather than independent, so the coalition odds carry a realistic spread rather than a falsely tight one; the full method (including this “correlated-error” term) is in the method note on the Nowcast tab.

How much these findings hold up

The tools, in one line each

Frequently asked questions

When is the 2026 Israeli election?

The election is set for 27 October 2026. The Knesset passed a dissolution bill and is set to dissolve on 17 July 2026, with the vote held on 27 October — the end of the current four-year term.

How many seats are needed to form a government in Israel?

61 of the 120 Knesset seats. No single party wins a majority, so governing means assembling a coalition of parties that together hold at least 61.

Who is ahead in the polls, and can Netanyahu form a coalition?

The tracker pools every major poll into an “if the vote were held today” nowcast. The pro- and anti-Netanyahu blocs are close, and in most polling neither reaches 61 on its own — see the Nowcast and Coalition tracker tabs for the current picture.

Can the opposition form a government without the Arab-majority parties?

Rarely in the polling: most surveys put the anti-Netanyahu bloc short of 61 without the two Arab-majority parties (Hadash-Ta’al and Ra’am). Use the Coalition builder to test any combination against 61.

Which pollsters are included?

All major Israeli pollsters (Midgam, Lazar, Kantar, Maagar Mochot, Direct Polls, Filber, Yossi Tatika and others), taken from each firm’s own release and cross-checked against Wikipedia.

What is a “nowcast”?

An estimate of the current state from the latest polls: what the seats would be if the vote were held today, rather than a forecast of the eventual October result.

Who will be the next Israeli prime minister?

The tracker doesn’t call a winner, but the PM suitability tab tracks head-to-head “who is more suited to be prime minister” polling among the main contenders (Netanyahu, Eisenkot, Bennett), and the PM tab also carries a live prediction market on who forms the next government.

Who made this

Compiled by Ron Levi, University of Toronto (ron.levi@utoronto.ca), built, analyzed, and annotated with Claude Code (Anthropic). It is a compilation of publicly available polls — not original polling, and not affiliated with any pollster or news outlet. To avoid double-counting, it lists only individual published polls (each pollster’s own survey) and never other aggregators’ or models’ averages. Bloc alignments (pro- vs anti-Netanyahu) reflect the current moment, not a fixed feature of the parties. It is a work in progress, and corrections, suggestions and advice are very welcome — please write.

SQL is an optional power tool — you don't need it; the other tabs answer most questions. It lets you ask precise questions by typing a short formula against one table called polls (one row per poll × party). Easiest way to start: click one of the grey example buttons below to run it, then edit it. (For yes/no fields, write scenario = false.)
Columns & values you can use — click any to drop it into your query
Columns:
Values:
Table polls — one row per poll × party. Booleans: scenario = false.
Nowcast — if an election were held today. What the most recent polls imply about who could form a government right now — not a forecast of the October result. (A “nowcast” estimates the current state from current data, as opposed to a forecast of the eventual outcome.)
Where the race stands — 30 Jul 2026

On the two-week polling average through 30 Jul 2026, neither side reaches the 61 seats a government needs. The anti-Netanyahu parties sit around 58, Netanyahu's coalition around 53, and the two Arab-majority lists (Hadash-Ta'al and Ra'am) near 9, currently aligned with neither camp. That leaves the opposition short of a majority unless those lists join it — the pattern in most polls since the spring.

Anti-Netanyahu ~58 · Netanyahu coalition ~53 · Arab-majority ~9 — of 120 seats, 61 to govern.
Two-week polling average, each pollster’s house effect evened out; the most recent single poll (Lazar, 30 Jul 2026) put the opposition at 61 and the coalition at 49. Rebuilt daily — this reading is from 31 Jul 2026.
Weight recent polls more All three keep the full two-week field of polls, so no pollster is ever dropped, and change only how much recent polls count, using a smooth decay rather than a hard cutoff. Even weights every poll the same (the default). Recent halves a poll’s weight for every ~4 days it is older, so most of the weight sits in the last week. Newest halves it every ~2 days, so roughly the last three days dominate — older polls still count, just far less. (A shorter hard window would instead drop whichever firms didn’t publish this week, swinging the result on cadence rather than opinion.)
Rise of challenger parties — a look at which combinations of party traits tend to go together with winning more seats. It is exploratory — it describes a pattern, it does not prove a cause.