Household headship rates from a census are the basis of the modelling that produces time series of total household counts for fiscal allocation and household survey weight calibration. Common practise is to break down headship rates by age-sex-race cells, meaning the quality of headship rates and the counts they produce depend directly on the quality of the demographic characteristics of household heads. In light of existing data quality concerns regarding Census 2022, I compare South African census and survey data for the period 1996-2024 and show that there is a high level of agreement between these sources regarding household headship rates, but this changes with Census 2022. Between Census 2011 and 2022, headship rates reduced for men and increased for women more than they did in the survey data from the same period. This outcome in the census is closely connected to two complementary changes regarding marital status, also not reflected in the survey data to the same degree. The first is an inversion of the stable historical rates at which husbands and wives headed households. The second is preliminary evidence of too few married people in Census 2022 based on a birth cohort analysis of the census data. Most household heads in South Africa are married men, and potentially too-low headship rates for married men combined with potentially too few married men is consistent with error in the direction of an undercount of households. My findings suggest that including household headship rates from Census 2022 in a household projection model might underestimate households going forward (important for fiscal allocation reasons) and backwards (important also for survey weighting reasons) in time.