In this paper, we systematically analyze writing variations of Swiss German in two existing corpora with standard German glosses, a corpus of 10,000 short text messages and a corpus of transcribed oral history recordings (90,000 tokens). We show that neither resource is sufficient for assessing factors in writing variations of users and describe a data collection project involving a citizen science community for solving this problem. Laymen will independently and redundantly transcribe 1,200 short samples (15-20 seconds) of audio material in Swiss German according to their own best practice.