A working paper from Stanford Digital Economy Lab studying the differences between attempts at measuring the number of US workers that work remotely found that half of US workers work remotely at least some days per week.
The differences between various surveys (Gallup, CPS, The Atlantic) can be explained by exclusions of those who are self-employed (more likely to work remote), questionnaire design and the intensity of remote work, industry, and whether or not the survey was measuring COVID-19 effects.
The CPS in particular undercounts relative to other surveys because it focuses solely on remote work due to COVID-19 and not those who worked remotely prior to the pandemic or permanently work remotely now.