Pollsters › Brazil

AtlasIntel: final polls against certified results

2 elections measured against the certified count.

Based in
Brazil
Countries polled
Brazil
Final polls held
2 (2 set against a certified count)
Modes
online: Random Digital Recruitment (Atlas RDR); online: Atlas Random Digital Recruitment (RDR)
Polls registered with
TSE PesqEle
Website
atlasintel.org

Brazil runoff 2026

for Bloomberg · fieldwork 2026-10-18 – 2026-10-23 · published 2026-10-24 · n = 4,990 · TSE PesqEle BR-06380/2026 · online: Random Digital Recruitment (Atlas RDR). Base: votos válidos (excluem brancos, nulos e indecisos). Source: bloomberglinea.com.br/brasil/flavio-bolsonaro-tem-508-dos-votos-validos-contra-492-de-lula-aponta-atlasintel/ (read 2026-10-25)

Mean error 2.6 points · margin error 5.2 points towards Flávio Bolsonaro · leader in final poll: different

ContestantPoll %Result %Error (pts)
Flávio Bolsonaro (PL)50.848.2+2.6
Lula (PT)49.251.8−2.6

Read in Bloomberg Línea, the Latin American partner of the commissioning outlet, Bloomberg.

Brazil general election 2026

for Bloomberg · fieldwork 2026-09-27 – 2026-10-02 · published 2026-10-03 · n = 4,945 · TSE PesqEle BR-00999/2026 · online: Atlas Random Digital Recruitment (RDR). Base: votos válidos (excluem brancos, nulos e indecisos). Source: cdn1.atlasintel.org/pesquisa_atlas_bloomberg__nacional_261003_1_2c124be0d65430b5.pdf (read 2026-10-25)

Mean error 1.7 points · margin error 4.8 points towards Lula · leader in final poll: different

ContestantPoll %Result %Error (pts)
Lula (PT)47.045.2+1.8
Flávio Bolsonaro (PL)44.147.0−2.9
Renan Santos (Missão)4.62.2+2.4
Augusto Cury (Avante)2.12.9−0.8
Ronaldo Caiado (PSD)1.42.2−0.8

Method

Each figure is the pollster’s final published poll before the vote, read from its own release, the commissioning outlet’s report, or an official registry; never from Wikipedia or an aggregator. It is set against the certified count as a share of valid votes. Error is poll minus result, in percentage points; the mean is over the contestants the poll and the count both name. Margin error is the certified winner’s lead over the certified runner-up as the poll had it, minus as it was, and names the one the poll leaned towards. Where the poll’s base differs (for example, all respondents including undecided), the comparison says so. These are measured facts, not grades: we rank no pollster.

JSON · Report a mistake