Pollsters screen people by their likelyhood to vote or ask them if they have voted in previous caucuses. If these people say 8/9/10 on a scale, you have a good sample of who is likely to vote.
Actually, to speed processing and keep costs down to acceptable levels while maintaining sufficient numbers of respondents to give provide statistical validity, a number of pollsters use panels.
These panels allow comparison of choices over time.
You both have to admit that polls only make sense when they are about open primaries.
Surveys about caucuses and/or closed primaries are rather useless.