Using speech recognition software adds to the likelihood. Dragon, for example, is really bad at its/it's, despite this almost always being eminently clear from the context. It's a weakness in the contextual match engine which remained in place last time I upgraded, though I've skipped the last two releases as they have become increasingly expensive for diminishing accuracy returns over the years. Maybe its [sic, for lols] fixed now.