Federal, May 2026

01

Three parties inside five points, and the poll orders none of them

Two points separate One Nation from Labor, three separate Labor from the Coalition and five separate One Nation from the Coalition. Each of those gaps is smaller than the interval that applies to it, so this wave settles no part of the order.

Each bar is the 95 per cent interval on a party's May first preference, from the effective sample of 754, and the dot is the share itself.

The poll excludes 7 per cent of respondents who were undecided. The election baselines are reconstructed from the swings DemosAU states and sum to 100.0. No two-party preferred figure is published; the report records that none was calculated.

The useful question of a poll with three parties inside five points is not which one is printed at the top but which orderings its sample can carry. One Nation's two points over Labor sit against a gap interval of about ±5.2. Labor's three over the Coalition sit against about ±5.0. One Nation's five over the Coalition sit against about ±5.1, which the gap falls a tenth of a point short of. All three comparisons come back the same way. What this wave's topline establishes is that One Nation, Labor and the Coalition are level with one another, and that the Greens and the Others are below all three.

The changes since April are smaller still: One Nation up two, Others down two, and Labor, the Coalition and the Greens unmoved. Against the May 2025 election the picture is different in kind. One Nation's 21.6-point rise, Labor's 8.6-point fall, the Coalition's 8.8 and the Others' 5.0 all exceed this poll's own interval several times over, because an election result is a count rather than a sample and the comparison carries only the poll's error. The Greens' 0.8 is the one movement since 2025 that this wave cannot distinguish from no movement at all.

02

Labor's 26 per cent is worth more seats than One Nation's 28

The report publishes a projection of the 150-seat House as a range for each party and states no method for it. The ranges are reproduced here as published, drawn end to end, with no midpoint and no most-likely figure added.

A majority of the 150 seats is 76. The top of Labor's range is 74, so no party reaches a majority at the top of its own range in this projection.

Labor polls two points below One Nation and is projected eighteen to seven seats above it. That is not a contradiction; it is what single-member districts do to a vote that is spread evenly against one that is concentrated. A party whose support sits at roughly its national level in most seats wins where it is narrowly ahead, and a party whose support is piled high in some places and thin in others banks surplus votes it cannot use. Section 05 shows One Nation at 38 per cent in regional and rural Australia and 18 in the inner metropolitan areas, which is the shape that produces this result.

Two cautions belong with these ranges. The report states no method for the projection, so this briefing reproduces the numbers and does not describe how they were reached. And a range is not a point: the distance between 16 and 28 Coalition seats is most of the difference between a diminished opposition and a marginal one, and nothing in the published figures says where inside that span the projection is most likely to fall.

03

Asked to imagine Hanson retired, the same sample gives the Coalition twice the seats

The report put a second voting-intention question to the same respondents, prefaced by the assumption that Pauline Hanson had announced her retirement at the age of 71. One Nation falls seven points and the Coalition rises four.

The question as asked excludes 7 per cent of respondents who were undecided and the hypothetical excludes 6 per cent.

Each bar is the published seat range for the hypothetical question, on the same scale as section 02.

Seven points leave One Nation and four of them arrive at the Coalition, which is the single most informative figure in this wave. It says that a little over half of the gap between the Coalition's 23 per cent and the 31.8 it won in 2025 is held by voters whose attachment is to a person rather than to the party that person leads. The remaining three points go to Labor and the Greens, and the Others column does not move at all.

The seat effect is larger than the vote effect. The Coalition's range runs 16 to 28 as the question was asked and 36 to 52 under the assumption, so four points of primary vote are worth roughly twenty seats at either end of the range. Labor's range crosses a majority of 76 in the hypothetical and does not in the question as asked. That is the arithmetic of a three-way contest: One Nation at 21 rather than 28 stops being competitive in enough seats that the two older parties take back most of what it was winning, and they do not take it back evenly.

This is a question, not a forecast. It measures what respondents said in May 2026 when prompted with a scenario, and such a prompt tends to produce a cleaner answer than an actual retirement would. It is recorded here because the report asked it and because the size of the response is the thing worth knowing.

04

The budget's tax changes are read as a cost rather than a reform

Fifty-three per cent agree the changes will make it harder for the average Australian and 44 per cent that they will hurt the economy, against 27 and 28 per cent agreeing they will make Australia fairer or housing more affordable.

Each bar is the net agreement with a statement, the agreeing share less the disagreeing share, in points. The note gives the three underlying shares.

Asked for an overall verdict on the 2026 Federal Budget, 43 per cent called it generally bad, 23 per cent generally good and 34 per cent were unsure. The three specific measures tested run the same way without running far: ending negative gearing for existing homes bought after 12 May 2026 draws 34 per cent approval against 29 per cent disapproval, replacing the 50 per cent capital gains discount with inflation-linked indexation draws 28 against 29, and a minimum 30 per cent tax rate on family trusts draws 27 against 34.

The don't-know shares are what make this battery worth reading. They run from 33 to 43 per cent across the seven questions, which is high for a budget already handed down, and they are largest on the two measures with the most technical descriptions. A reader should treat the net figures above as the settled opinion of between half and two-thirds of the electorate rather than of all of it, and the budget's reception as unformed rather than as rejected.

05

One Nation's vote is regional and the Greens' is young

DemosAU publishes no cohort sample sizes for this wave. If a three-way split holds about a third of the effective sample of 754 each, a gap between two of those groups carries roughly ±7.9 points, and a gap between two halves about ±6.4.

Every row sums to 100 in the report. The undecided share of 7 per cent sits outside these totals. The report also publishes a breakdown by personal income, which runs from 24 per cent Labor and 29 One Nation below $45,000 to 28 and 21 above $125,000. Cohort figures rest on subsamples and carry wider margins than the ±3.6% national topline.

Three divisions in this table are wide enough to survive their intervals. One Nation takes 38 per cent in regional and rural Australia against 18 in the inner metropolitan areas, a twenty-point gap. The Greens take 33 per cent of under-35s against 4 per cent of the over-55s, which is twenty-nine points and the widest single contrast in the wave. And Labor takes 39 per cent of respondents who speak a language other than English at home against 23 of those who speak English, a sixteen-point gap that holds even on the smaller subsample the first group must represent.

The gender row is worth stating precisely because it runs against expectation and does not clear. One Nation is on 30 per cent among women and 28 among men, and Labor on 24 and 28. Both gaps are inside the roughly ±6.4 points a two-way split of this sample supports, so this wave establishes no gender division in either direction. The Coalition's figures are flat where this series usually finds structure: 22 and 24 by gender, 19 to 27 across education, and 22 to 24 across the three location bands. Where it does concentrate, it concentrates on age, tenure and income together, at 29 per cent of the over-55s against 13 of the under-35s, 29 of home owners against 15 of renters, and 32 of those earning more than $125,000 against 19 of those earning less than $45,000. Each of those three gaps clears the roughly ±7.9 points a three-way split of this sample supports.

06

One Nation holds 98 per cent of its own voters and 37 per cent of the Coalition's

The report breaks this wave's vote down by how each respondent says they voted in May 2025. The row that carries the Coalition's position is its own.

The report draws this table as stacked bars and prints a figure only where a segment is wide enough to hold one. An en dash marks a cell the report leaves unlabelled; each of those rows balances to 100 across the cells shown plus the unlabelled remainder, and this briefing does not write a figure into a cell the report does not print.

Thirty-seven per cent of the people who say they voted Coalition in May 2025 now say they would vote One Nation, against 60 per cent who would vote Coalition again. No other row in the table loses a third of itself to a single destination. Labor's seventeen points to One Nation is the next largest and is less than half the size, and the Greens and One Nation hold 86 and 98 per cent of their own.

That 98 is the figure to keep. A party polling 28 per cent that almost no one leaves is in a different position from one polling 28 per cent with ordinary churn, because it means the whole of the movement in this series has to come from somewhere other than One Nation's own voters changing their minds. It also puts a floor under the retirement counterfactual in section 03: the seven points that left One Nation when Hanson was removed from the question came out of a column that is otherwise almost perfectly retentive.