Right?
100 is so small that the results probably aren’t statistically valid.
Especially if you carefully choose the samples.
Central limit theorem. If you randomly select your sample, the probability of a sample of 10,000,000 having a substantially different percentage of bot accounts is very very small.