Twitter doesn't say that 5% of all accounts are bots. While I don't have inside knowledge of how Twitter does things, the industry standard is to: 1. Build a c

Twitter doesn't say that 5% of all accounts are bots. While I don't have inside knowledge of how Twitter does things, the industry standard is to: 1. Build a classification model to determine whether an account is a bot or not. 2. Measure the precision/recall curve of that model, and pick a threshold where the precision would be >95% (i.e. >95% of what the model labels as "humans" are really humans). (Higher recall leads to lower precision and vice versa. Higher recall means a higher mDAU, so it's a balancing act between a high number versus a high confidence in that number.) 3. Let the model loose on all accounts. 4. The number the model tags as humans is the "mDAU", though <5% of these might be false positives. While this model takes a lot of training data to build, verifying its continued precision/recall characteristics can be completed with very little ongoing sampling - easily handled by the "100 a day" figure Musk has claimed.