That sounds like they have a model of population demographics and try to “correct” their sample for various biases that effect likelihood of reaching and getting a response from people by reweighting the actual sample to match the expected demographics.

If actual views on the target issues are predicted well by demographics and independent of the response biases beyond their correlation with the demographics, this should work well.

If they aren't, though, this potentially magnifies the impact on the overall result of subsamples that are potentially most unreliable because of small size.