Brazil’s presidential runoff has undergone a striking compression in just over two weeks. Flávio Bolsonaro emerged from the October 4 first round ahead of President Luiz Inácio Lula da Silva, and early runoff surveys suggested his advantage could be several points wider. On October 23, however, a new Ipespe/Abrapel poll put the candidates just one percentage point apart among valid votes. The change is real in the institute’s reported measurements — but it is not proof that either candidate has secured an election-day lead.
From first-round upset to a narrowing runoff
The first-round count gave Flávio Bolsonaro 47.03% of valid votes to Lula’s 45.16%, according to figures reported from Brazil’s electoral authorities. That 1.87-point advantage was enough to put Bolsonaro first but not enough to avoid a runoff on October 25. It also set a different starting point from the expectations generated by some pre-election polling averages.
In the first Ipespe/Abrapel runoff tracking round, conducted October 6–8, Bolsonaro stood at 52.7% of valid votes and Lula at 47.3% — a 5.4-point gap. The institute’s subsequent round, conducted October 13–15, measured 51.2% to 48.8%, reducing that gap to 2.4 points. Its latest round, conducted October 20–22 and released Friday, put Bolsonaro on 50.5% and Lula on 49.5%: a gap of just one point.
That is a 4.4-point reduction in the measured candidate gap within this one tracking series, from the first post-election round to the latest. It should not be interpreted as a 4.4-point shift in individual voters or as proof of a statistically significant trend without the underlying survey design and uncertainty of the change.
What the October 23 poll actually shows
Ipespe/Abrapel interviewed 1,500 Brazilian voters by telephone between October 20 and 22. Its reported margin of error is ±2.6 percentage points, with a confidence level of approximately 95%. The poll is registered with Brazil’s Superior Electoral Court under BR-04350/2026. The 50.5% and 49.5% figures refer to valid-vote estimates, excluding blank and invalid votes and undecided responses; they are not shares of everyone interviewed.
In its all-respondent stimulated question, the institute measured 48% for Bolsonaro and 47% for Lula. The valid-vote conversion is useful when comparing to the official election result, but it assumes that excluded groups will not change the final allocation in a way the survey cannot observe.
Source: CNN Brasil’s October 23 report on Ipespe/Abrapel and the institute’s account of its first runoff round.
Why other polls matter
A close finish in one polling series does not erase the spread between pollsters. The AtlasIntel/Bloomberg survey conducted October 13–18 and published October 19 measured Bolsonaro at 51.1% and Lula at 48.9% among valid votes. Veritá’s October 15–19 fieldwork, published October 21, measured 50.7% and 49.3%, respectively. These surveys likewise show narrow numerical Bolsonaro advantages, but their different samples, modes and fieldwork periods prevent a simple like-for-like comparison.
Other surveys have reported a Lula edge. Votendi’s October 22 report examined a Vox Brasil result putting Lula at 50.6% and Bolsonaro at 49.4%. Reporting the existence of differing poll measurements is not the same as declaring either candidate ahead in the overall polling landscape. Each survey needs to be read with its sample definition and treatment of undecided voters.
For a consistent view of individual survey observations, read Votendi’s Brazil presidential runoff tracker and its latest Datafolha and Vox coverage. The tracker snapshot previously available to us placed its weighted average at October 20; a weighted average with that cutoff must not be labeled a complete October 23 estimate.
The question for Sunday is turnout as much as preference
The runoff is not a repeat of the first round. Some voters whose preferred candidate was eliminated must choose again, vote blank or invalid, or abstain; those who did not vote earlier may turn out this time. Small net differences in turnout and in the distribution of those choices can matter when measured candidate shares are this close.
There is also a methodological distinction between a poll lead and a reliable prediction. A one-point observed margin with a 2.6-point individual sampling error does not establish which candidate will finish first. Nor does a series of increasingly close snapshots explain, by itself, why respondents answered differently.
A tight election, not a declared outcome
Friday’s data place the Ipespe/Abrapel series near parity after starting the runoff campaign with a clearly larger Bolsonaro advantage. The independent polls published in recent days document a range of close outcomes rather than a settled national consensus.
Brazil votes on Sunday, October 25. The result will be determined by ballots counted by the electoral authorities, not by one survey or any single polling average. Votendi will continue separating dated poll results, modeled averages and official returns as the contest concludes.
