Google Surveys makes use of the inferred demographic and location information to employ stratified sampling by distributing the surveys based on the targeted audience to our publisher network and/or smartphone (Android and iOS) users. We infer demographics through respondents' browsing history (e.g., infer geography from IP address), then we match them against existing government statistical data. Google Surveys uses post-stratification weighting to compensate for sample deficiencies to remove bias among the survey sample. This gives a more accurate result with lower root mean square error (RMSE) which also makes the results better represent the Current Population Survey (CPS).
Google Surveys performs equal to or better than existing probability and non-probability based Internet survey panels. A full description of the Google Surveys methodology can be found in this whitepaper or the product overview.