You need to though. If you only do phone call polls you're only going to poll people with (landline?) phones who answer (random?) calls and have the time/enthusiasm to respond when you call. That heavily skews demographics.
A lot of people will only do online polling, and it's by far the easiest way to cover a lot of demographics (need to supplement with other methods as well if you don't want a poll of "internet users" - but that is a far larger and varied group than most other methods would produce)
Getting a different error is great! Then if you do an online poll and a phone poll, you can combine the results to get a composite poll that's more accurate than either method alone.
Both oversample people that want to take polls, that's the problem.
In fact, online polls are worse because you now have to make sure the "people" answering are actually people and are actually able to influence whatever polling data you're trying to gather.
They've gone from gathering a dataset that's been becoming less reliable to one that is even less reliable.
A lot of people will only do online polling, and it's by far the easiest way to cover a lot of demographics (need to supplement with other methods as well if you don't want a poll of "internet users" - but that is a far larger and varied group than most other methods would produce)