One Nation’s lead was a coalition of the discontented. In August it started coming apart.
The Capital Brief/DemosAU August poll records the first statistically significant fall in One Nation’s national primary vote since its rise began. What moved, who moved it, and what the seat model does with a five-point shift.
Fieldwork 18–20 August 2026 · n = 1,562 · effective n = 754 · effective margin of error ±3.6 per cent.
The numbers
One Nation’s national primary vote fell from 29 per cent in July to 24 per cent in August, placing it third behind Labor (26) and the Coalition (25). The five-point fall exceeds the 95 per cent confidence bound for a change between these two waves (±3.8 points) and is the first movement in the series this year that clears that bar. The Coalition’s four-point rise clears it as well. Labor did not move.
DemosAU’s Monte Carlo seat projection, which applies a uniform national swing to the May–July MRP model, now puts the Coalition (32–50 seats) ahead of One Nation (26–39) as the larger non-Labor bloc, with Labor on 61–73. Seventy-six seats form a majority in the 150-seat House of Representatives. On these ranges no party reaches it.
One line falls five points, and the line it had passed comes back
National primary vote, May 2025 election to August 2026. The axis is time-true, so the eleven months between the election and the first 2026 reading are drawn as eleven months. A hollow marker is an election result rather than a poll; the whisker on the August point is that wave’s own ±3.6 points. Select a party in the key to follow it alone.
Three of the five movements on this figure are measurements rather than noise: One Nation −5 (z = 2.6), the Coalition +4 (z = 2.1) and Others +3 (z = 2.1) each clear the 95 per cent threshold. The Greens’ two-point fall (z = 1.4) does not, and Labor did not move at all. Election baselines are taken from the swing figures published in the DemosAU report; undecided respondents, 5 per cent of the sample, are excluded from the primary vote.
CAPITAL BRIEF/DEMOSAU. FIELDWORK 18–20 AUG 2026, N=1,562, EFFECTIVE N=754, MOE ±3.6%.
A five-point vote change moves the Coalition twenty-eight seats, and nobody reaches 76
The August poll against the May–July MRP the projection swings from. Every range is drawn against the same 76-seat majority line. The Coalition’s Liberal and National split is a toggle, because the record approved the combined range as the figure.
The ranges are the method reporting its own uncertainty, not a hedge: each is drawn from 20,000 simulations. What the figure shows is a swap in the order of the two non-Labor blocs, on a Labor count that barely moves.
DEMOSAU MONTE CARLO SEAT PROJECTION, 20,000 SIMULATIONS, UNIFORM SWING FROM THE MAY–JULY 2026 MRP. SEAT-LEVEL MARGIN ±5%.
The leaders, in the same wave. On preferred prime minister, Albanese leads on 35 per cent (no change), Taylor is on 23 (+1) and Hanson on 22 (−4), with 20 per cent saying they do not know (+3). Head to head, Albanese leads Hanson 43–32 with 25 per cent undecided, and Taylor 38–36 with 26 undecided. On net personal ratings Albanese is on −19 (+4 since July), Taylor −4 (−1) and Hanson −17 (−8). Hanson’s is the only rating of the three moving away from zero.
Where the vote moved
The August report’s recontact table shows where the July One Nation vote went. Of respondents who voted Coalition in May 2025, 67 per cent now say Coalition and 31 per cent say One Nation; in the July dashboard the One Nation share of that group was 37 per cent. Of 2025 One Nation voters, 96 per cent remain with the party. The fall is therefore a partial return of 2025 Coalition voters, not an erosion of One Nation’s 2025 base.
The gender split is the second half of the story. DemosAU reports Hanson’s net rating among women moving from +2 in June (39 positive, 37 negative) to −26 in August (26 positive, 52 negative). The One Nation primary vote among women is 22 per cent against 26 among men; DemosAU puts the fall since June at 10 points among women and 3 among men. A subgroup of roughly half the effective sample carries a margin of about ±4.4 points on a proportion near 25 per cent, so the size of the women’s fall is established; its exact magnitude is not.
Two structural features of the One Nation vote persist in the crosstabs. It is a regional vote: 34 per cent in regional and rural Australia against 20 in inner metropolitan areas. It is an older vote: 27 per cent among those 55 and over, where the Coalition is on 32.
One Nation’s own base did not move. The vote it had borrowed did.
Each row is a 2025 voting cohort and how it divides in August 2026. Rows sum to 100. Select a party in the key to follow it down every row.
Read across the Coalition row and the argument is there: 31 per cent of 2025 Coalition voters still say One Nation, against 37 in July. The sixth row is where the rest of the borrowed vote sits — among people who did not vote in 2025, One Nation takes 40 per cent, more than in any other cohort.
CAPITAL BRIEF/DEMOSAU AUGUST 2026 REPORT, RECONTACT BY 2025 FEDERAL VOTE.
What the seat model is doing
The seat ranges move far more than the primary votes because Australia’s single-member, full-preferential lower house translates a three-way contest into seats non-linearly. A party on 24–29 per cent nationally wins seats only where its vote is concentrated and where preference flows from the other parties do not lock it out. DemosAU flagged this in its 6 August note on primary vote sensitivity: near the crossover point, a few points of national swing can move dozens of seats between the Coalition and One Nation without changing Labor’s count. The August result is that sensitivity operating in the Coalition’s direction.
Three caveats apply to the projection and are stated in the report. It assumes uniform swing from an MRP fitted on May–July data, so state-level variation since then is not modelled. It carries a stated ±5 per cent seat-level margin. And it is a projection from one wave of one pollster; the MRP itself, with 9,343 respondents, remains the stronger instrument for seat-level claims until the next one is fielded.
What to watch
The next federal election is due by May 2028. The relevant questions for the next wave are whether the Coalition’s recovery among 2025 Coalition voters continues, and whether Hanson’s rating among women stabilises or continues to fall. The state series shows the same pattern in NSW, where the Coalition regained four points from One Nation over the same fortnight, and the opposite pattern in Western Australia, where One Nation rose three points ahead of the Secret Harbour by-election. Those two results are covered in the companion analyses.
One Nation’s July lead was assembled from other parties’ 2025 voters and from people who did not vote in 2025, and that is the part of it that moved. Its own base held at 96 per cent. For anyone reading this as a trend rather than a wave: the number to watch is not One Nation’s primary vote but the Coalition share of the 2025 Coalition cohort, which has gone 57 to 67 in two months. The seat model will keep amplifying that, because near the crossover point it amplifies everything.
Method and sources
Capital Brief/DemosAU. Internet panel, n = 1,562, effective n = 754, effective margin of error ±3.6 per cent at 95 per cent confidence. Rim weighting on age × gender, education, personal income, residential tenure, language, state × region, 2025 federal vote and left–right self-identification. Voting intention was the first substantive question after demographics. Undecided respondents were asked a leaner question; the remainder were excluded. No two-party preferred figure is published for this wave. Full methodology statement, question wording and crosstabs are in the report PDF.
Wave-on-wave differences in this analysis are tested as differences between independent proportions using each wave’s effective sample size. A five-point One Nation fall (z = 2.6), a four-point Coalition rise (z = 2.1) and a three-point Others rise (z = 2.1) each exceed the 95 per cent threshold; the two-point Greens fall (z = 1.4) does not. What that threshold is, and why a margin of error covers a single figure rather than the gap between two, is set out in our guide to how to read an Australian election poll.
Estimated figures in Figure 3
The recontact chart on page 11 of the report prints a figure against its larger segments and leaves the smaller ones unlabelled. The report is a vector drawing rather than an image, so those segments are measurable: each was read from the drawing’s own geometry, and each landed within half a point of a whole number. Every row then sums to exactly 100 across the five parties, which is the check that the reading is right rather than merely plausible.
Values read that way are marked est. in this figure’s table view and in its tooltips. They are: the Greens and Others shares of the 2025 Labor row; every share but the Coalition’s and One Nation’s in the 2025 Coalition row; the Coalition share of the 2025 Greens row; and every share but One Nation’s in the 2025 One Nation row. Ace Strategies confirms these before publication.
One figure quoted in the analysis above is not drawn from this report at all. The size of One Nation’s fall among women since June — 10 points against 3 among men — is DemosAU’s own comparison across two waves, and the June report is not held here. It rests on DemosAU’s release rather than on a table this analysis has read.
Sources and related reading
How this poll was conducted
- Fieldwork
- 18–20 August 2026
- Sample
- 1,562 respondents (754 effective, after weighting)
- Margin of error
- ±3.6%
- Population
- Australia
- Method
- Internet panel survey, rim weighted by age × gender, education, personal income, residential tenure, language, state × region, 2025 federal vote and left–right self-identification. Undecided respondents were asked a leaner question; the remainder were excluded.
- Commissioned by
- Capital Brief