Who Uses Generative AI for Schoolwork in Brazil? A Descriptive Note on Household Class and Joint Use of Web Search in TIC Kids Online Brasil 2025

TIC Kids Online Brasil 2025 (fieldwork March-September 2025) is a national household survey of internet users aged 9 to 17 with generative AI items; its published marginal table puts the share using generative AI for school research or study highest in AB households. This note re-analyses the public microdata (2,370 respondents) under a written analysis plan made after the published marginals were available. The plan's Japanese original is unpublished (an English translation is included) and its timing relative to the first results cannot be verified, so the analysis is not called pre-registered. The plan set two primary quantities: (H1) the DE minus AB gap in the share using generative AI for school, adjusted for parental education, age, sex, region, rurality and school internet use; and (H2) among those users, the DE minus AB gap in the share who did not also search the internet for schoolwork. The plan's first stop condition fired: before the study outcomes were estimated, the ratio of the replicate-weight margin of error for a non-study item to the margin Cetic.br publishes was 0.785 (0.783 under the two alternative formulas), outside the required 0.80 to 1.25, so the variance formula is unverified. Under the plan no decision on H1 or H2 is made, so only point estimates are reported, with no intervals and no test. Weighted use of generative AI for school is 64.6% (AB), 56.8% (C) and 61.0% (DE); the adjusted DE minus AB coefficient is -4.4 percentage points. Among these users, the share who did not search the internet for schoolwork is 4.2% (AB), 11.9% (C) and 8.6% (DE), a DE minus AB difference of +4.4 points (+6.3 adjusted). Their sampling error is not established. The ratio of replicate to published margins is 0.785 to three decimals in every cell compared, a discrepancy the methodology report does not explain, bearing on any difference in the published tables.