The decennial census is foundational to American governance at all levels. For example, census population data determine the allocation of seats in the US House of Representatives, the distribution of more than $1.5 trillion annually in federal funds for programs, private-sector marketing campaigns, and more. Given the importance of the decennial census, the US Census Bureau has constantly evolved to use up-to-date technology and implemented an online response option for the first time for the 2020 Census. Because high self-response rates are crucial for minimizing costs and ensuring high-quality data, this study seeks to identify publicly available demographic characteristics that predict low response rates on the 2020 Census. We use difference-in-difference-in-difference and multivariate regression methods and find that previous survey response rates, the population age 65 and above, broadband access, and the so-called low response score are useful to identify census tracts with low self-response rates. With this information, we further consider two similar methods to determine theoretical alternative tracts for the Census Bureau’s Internet Choice outreach. We compare the tracts those methods would have classified as Internet Choice with the Census Bureau’s 2020 allocation and compare the tracts with each other. These comparisons illustrate that Internet Choice tract allocation could have been substantially different using an alternative classification process, even when using identical variables. Additionally, both of our methods would have led to 2020 Internet Choice tracts being more urban, racially diverse, and socioeconomically vulnerable than the actual 2020 Internet Choice tracts.